test(experiment): freeze client wait guidance diagnostic

Keep identical analysis contracts, question, engine and complete catalogs. Test common opt-in wait guidance while retaining polling and native latency in whole-task costs.

Co-authored-by: gensokyo-dev <342580529+gensokyo-dev@users.noreply.github.com>

Co-authored-by: Kaoru0822 <208281322+kaoru0822-kitauji@users.noreply.github.com>

.agents/skills/reverse-engineer-anything/SKILL.md

@@ -59,7 +59,7 @@ availability have different remedies; avoid repeating setup for all of them.
    absent, inspect the client's MCP launch error rather than repeating setup.
 
 REA setup never installs or upgrades Node.js, npm, Homebrew, Java, Ghidra, IDA,
-JADX, Binwalk, or Unblob. Use existing prerequisites; do not install unrelated
+JADX, Kuna, Binwalk, or Unblob. Use existing prerequisites; do not install unrelated
 software to repair
 MCP registration. For an unsupported client, use manual stdio registration
 with a version-pinned `rea-agents` package or continue through the CLI.
@@ -79,7 +79,7 @@ is within the authorized scope, use `npx -y rea-agents@latest` in its place.
 Configuration repair starts with a read-only setup plan; show paths, backups and
 skill replacement before applying authorized changes. Preserve existing approval
 and target-execution boundaries. REA setup never installs or upgrades unrelated
-software, including Node.js, npm, Homebrew, Java, Ghidra, IDA, JADX, Binwalk or Unblob.
+software, including Node.js, npm, Homebrew, Java, Ghidra, IDA, JADX, Kuna, Binwalk or Unblob.
 
 ## Route the target first
 
@@ -215,6 +215,20 @@ after opening; IDA does not supply `binary_overview`. Consult session availabili
 when composing broader workflows. Modern direct callers and unsupported dossier
 facets remain unknown; live IDA results are not replayed from snapshots.
 
+With `REA_KUNA_COMMAND` configured, `provider_id: "kuna"` answers procedure
+lists, lookups, pseudocode, assembly and `analyze_function` from Kuna's cached
+whole-program export. Callers/callees use a separate body-free graph shared by
+the session; `xrefs` returns exact-address incoming sites and filters forwarding
+aliases. These queries and inventory lookups do not trigger a C export. Indirect
+targets remain unknown and can make relationship coverage partial. Reference
+caches expire on close; a fresh CLI process repeats the native reference walk.
+Typed references, CFG, raw-memory tools and annotations remain unsupported. A cold first
+answer can come from a single-function run (its limitations say so); assembly
+waits for the export. For pseudocode and assembly, `raw_result.source` names the
+cached `.c`/`.asm` file and byte span, which you may search directly instead of
+calling per function. See the
+[Kuna provider guide](https://github.com/morluto/rea/blob/main/docs/kuna-provider.md).
+
 If the user says only "my app" without a name, path, or target already selected
 for this task, ask which app before searching for artifacts. Paths in unrelated
 saved reports and nearby test fixtures do not select the user's target. Resolve

.agents/skills/reverse-engineer-anything/references/javascript-applications.md

@@ -94,6 +94,36 @@ recorded observations, identity, hashes, exports and source ranges; it does not
 include every child syntax observation, related edge or original source file.
 An absent item fact is not evidence of absence from the whole application.
 
+For canonical identities and retained source-range counts, use a typed reference
+page instead of reading every full item:
+
+```json
+{
+  "source": {
+    "kind": "retained-evidence",
+    "evidence_id": "ev_ACTUAL_PARENT_ID"
+  },
+  "view": {
+    "kind": "javascript",
+    "collection": "modules",
+    "projection": "references",
+    "where": { "path": "main.js", "kind": "javascript-module" },
+    "offset": 0,
+    "limit": 16
+  }
+}
+```
+
+Filters match exact retained strings; omit a filter to include all values, use
+`path: null` for unavailable paths, and `path: ""` for a recorded empty path.
+Rows preserve canonical `identity` and distinct observation support states in
+`basis`. `retained_source_range_count` counts retained source-range observations,
+not runtime imports or unique reference sites. Use a row's `node_id` with the item
+view above for exact locations and full supporting observations. Follow
+`coverage.next_offset`; page exhaustion describes matching retained rows only.
+Check `source_coverage` for source selection, partial/unknown graph coverage and
+unknown counts before drawing a negative conclusion.
+
 ## Trace only the remaining question
 
 Use `trace_application_feature` on the parent application Evidence for one

.github/workflows/ci.yml

@@ -60,6 +60,7 @@ jobs:
       web_runtime: ${{ steps.plan.outputs.web_runtime }}
       web_source_map: ${{ steps.plan.outputs.web_source_map }}
       nativeaot: ${{ steps.plan.outputs.nativeaot }}
+      ghidra_bridge: ${{ steps.plan.outputs.ghidra_bridge }}
       website: ${{ steps.plan.outputs.website }}
       matrix: ${{ steps.plan.outputs.matrix }}
       test_selection: ${{ steps.plan.outputs.test_selection }}
@@ -139,6 +140,11 @@ jobs:
       - name: Validate workflow syntax and expressions
         if: needs.changes.outputs.workflows == 'true'
         run: go run github.com/rhysd/actionlint/cmd/actionlint@v1.7.7 -shellcheck= -pyflakes=
+      # Code tiers run this source-only policy test in their existing shards.
+      # Workflow-only changes need the guard, without preparing runtime artifacts.
+      - name: Check action pins for workflow-only changes
+        if: needs.changes.outputs.workflows == 'true' && needs.changes.outputs.code != 'true'
+        run: node scripts/run-exact-tests.mjs run tests/boundary/filesystem/workflowActionPinning.test.ts
       - name: Record static task source and execution identity
         continue-on-error: true
         if: needs.changes.outputs.code == 'true'
@@ -741,6 +747,7 @@ jobs:
           tests/boundary/filesystem/javascriptApplicationWorkflows.test.ts
           tests/boundary/filesystem/webScriptManifestPaths.test.ts
           tests/boundary/process/ProcessEvidence.host.test.ts
+          tests/boundary/process/packageNpmInvocation.test.ts
           tests/boundary/process/windowsOutputCapture.test.ts
           tests/acceptance/webModuleCli.test.ts
           tests/acceptance/webSourceLocationCli.test.ts
@@ -898,6 +905,11 @@ jobs:
     if: needs.changes.outputs.nativeaot == 'true'
     uses: ./.github/workflows/nativeaot-fixtures.yml
 
+  ghidra_bridge:
+    needs: [changes, static]
+    if: needs.changes.outputs.ghidra_bridge == 'true'
+    uses: ./.github/workflows/ghidra-bridge-sources.yml
+
   website:
     needs: [changes, static]
     if: needs.changes.outputs.website == 'true'
@@ -933,6 +945,7 @@ jobs:
         web_runtime,
         web_source_map,
         nativeaot,
+        ghidra_bridge,
         website,
       ]
     runs-on: ubuntu-24.04

.github/workflows/ghidra-bridge-sources.yml

@@ -0,0 +1,48 @@
+name: Ghidra bridge source compilation
+
+on:
+  workflow_dispatch:
+  workflow_call:
+
+permissions:
+  contents: read
+
+env:
+  HUSKY: "0"
+
+jobs:
+  ghidra-bridge-sources:
+    name: Compile packaged Ghidra bridge sources
+    runs-on: ubuntu-24.04
+    timeout-minutes: 15
+    env:
+      GHIDRA_RELEASE_URL: https://github.com/NationalSecurityAgency/ghidra/releases/download/Ghidra_12.1.4_build/ghidra_12.1.4_PUBLIC_20260921.zip
+      GHIDRA_RELEASE_SHA256: ddac49f903da9d5bac833e5cc79395098b9c33cfd3279be5f31bd00387d2d4db
+    steps:
+      - uses: actions/checkout@11d5960a326750d5838078e36cf38b85af677262 # v4.4.0
+      - uses: actions/setup-node@49933ea5288caeca8642d1e84afbd3f7d6820020 # v4.4.0
+        with:
+          node-version-file: .nvmrc
+      - name: Validate the SDK pin against the provider's verified release
+        run: node scripts/verify-ghidra-bridge-sources.mjs --check-sdk-pin >> "$GITHUB_ENV"
+      # Only the jars and application.properties are extracted: the check
+      # compiles against them and never launches Ghidra or imports a target.
+      - name: Acquire checksum-verified Ghidra jars in isolated runner storage
+        shell: bash
+        run: |
+          ghidra_root="${RUNNER_TEMP}/rea-ghidra"
+          mkdir -p "${ghidra_root}"
+          curl -fsSL --retry 3 --write-out 'ghidra_download_bytes=%{size_download} ghidra_download_seconds=%{time_total}\n' -o "${ghidra_root}/ghidra.zip" "${GHIDRA_RELEASE_URL}"
+          echo "${GHIDRA_RELEASE_SHA256}  ${ghidra_root}/ghidra.zip" | sha256sum -c -
+          unzip -q "${ghidra_root}/ghidra.zip" '*/Ghidra/*.jar' '*/Ghidra/application.properties' -d "${ghidra_root}"
+          rm "${ghidra_root}/ghidra.zip"
+          properties="$(find "${ghidra_root}" -path '*/Ghidra/application.properties' -print -quit)"
+          test -n "${properties}"
+          echo "GHIDRA_INSTALL_DIR=$(dirname "$(dirname "${properties}")")" >> "${GITHUB_ENV}"
+      # JAVA_HOME_<major>_X64 comes from the runner image, not the env context.
+      - name: Compile bridge sources with the verified runner JDK
+        shell: bash
+        run: |
+          java_home_variable="JAVA_HOME_${GHIDRA_JAVA_MAJOR}_X64"
+          test -n "${!java_home_variable}"
+          JAVA_HOME="${!java_home_variable}" npm run verify:ghidra:bridge-sources -- --verified-sdk

.github/workflows/real-cutter.yml

@@ -22,7 +22,7 @@ jobs:
           sudo apt-get update
           sudo apt-get install --no-install-recommends -y g++-mingw-w64-x86-64
       - run: npm run build:windows-native
-      - uses: actions/upload-artifact@v4
+      - uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4.6.2
         with:
           name: cutter-windows-native
           path: native/windows/build/
@@ -53,7 +53,7 @@ jobs:
         uses: ./.github/actions/npm-cache
       - run: npm ci
       - if: matrix.platform == 'Windows'
-        uses: actions/download-artifact@v4
+        uses: actions/download-artifact@d3f86a106a0bac45b974a628896c90dbdf5c8093 # v4.3.0
         with:
           name: cutter-windows-native
           path: native/windows/build/
@@ -67,7 +67,7 @@ jobs:
         env:
           EXPECTED_PLATFORM: ${{ matrix.platform }}
         run: npm run verify:cutter
-      - uses: actions/upload-artifact@v4
+      - uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4.6.2
         if: always()
         with:
           name: cutter-upstream-${{ matrix.platform }}

AGENTS.md

@@ -20,6 +20,8 @@ Give each interpretation and resource a clear owner. Consumers must use that own
 
 Keep proof of ownership separate from validity, editability, and readiness. Malformed owned data may still be safely removable. Unresolved identity or applicability must remain unknown rather than be treated as absent, foreign, or aligned.
 
+For lifecycle changes, follow [ownership and completion invariants](docs/tool-design.md#own-the-lifecycle) and select the applicable [lifecycle regressions](docs/testing.md#lifecycle-regressions). Include late handshakes, cancellation during final cleanup, mixed child outcomes, and cleanup retry where those stages exist; active-execution cancellation alone does not cover finalization.
+
 Assess brittleness in touched code during implementation, debugging, review, and maintenance. Fix the producing representation or ownership rule before adding downstream repairs. Normalize once at the owning boundary using format-aware parsers. Remove superseded representations, aliases, wrappers, and redundant tests after verifying affected consumers. Introduce abstractions only when they remove observed complexity.
 
 When fixing a bug, inspect the producing boundary and affected consumers for duplicated interpretation or unclear ownership. Complete any directly related refactor in the same change and remove superseded code. Passing regression tests alone does not establish that the brittleness is resolved. If related cleanup remains, explain what remains and why.

README.md

@@ -205,7 +205,8 @@ or update either provider.
 
 Native formats and host support vary by provider. See
 [Hopper and Ghidra setup](docs/installation.md#hopper), the
-[IDA guide](docs/ida-provider.md) (including an optional native MCP bridge), and
+[IDA guide](docs/ida-provider.md) (including an optional native MCP bridge), the
+[Kuna decompiler guide](docs/kuna-provider.md), and
 [experimental Windows Ghidra support](docs/windows-ghidra-p0.md).
 Ghidra also supports [16-bit DOS analysis](docs/ghidra-dos.md).
 For large binaries, raise the provider's startup deadline with
@@ -344,7 +345,7 @@ Report vulnerabilities through [SECURITY.md](SECURITY.md).
 
 ## Star history
 
-🎉 **200,000 GitHub stars — thank you!**
+🎉 **250,000 GitHub stars — thank you!**
 
 Thanks to everyone using REA, reporting bugs, requesting features, testing builds, and contributing fixes.
 

THIRD_PARTY_NOTICES.md

@@ -58,6 +58,7 @@ pin inspection sources; they are not included in the npm package.
 | ghidra-nativeaot               | [MIT](https://github.com/Washi1337/ghidra-nativeaot/blob/effeb734fc570c32650f88b159608979dc7b423e/LICENSE.md)                                                                                                                                            | Pinned inspection source; derived parser attribution above.                                                                                                                                                                 |
 | JADX                           | [Apache-2.0](https://github.com/skylot/jadx/blob/v1.5.6/LICENSE)                                                                                                                                                                                         | The audited jadx-headless-mcp source uses JADX 1.5.6 for Android decompilation; the engine JAR is separately supplied.                                                                                                      |
 | jadx-headless-mcp              | [Apache-2.0](https://github.com/1013503897/jadx-headless-mcp/blob/5844800a486d2248fa949b74c9896285f5b54de2/LICENSE)                                                                                                                                      | v0.7.1; pinned submodule and user-supplied JAR.                                                                                                                                                                             |
+| Kuna                           | [Apache-2.0](https://github.com/Noelo-Lab/kuna/blob/v1.833/LICENSE), with its [NOTICE](https://github.com/Noelo-Lab/kuna/blob/v1.833/NOTICE)                                                                                                             | Bring your own `kuna`; REA verified 1.833 and never downloads, installs or bundles it. REA invokes it as a separate process.                                                                                                |
 | Apktool                        | [Apache-2.0](https://github.com/iBotPeaches/Apktool/blob/master/LICENSE.md)                                                                                                                                                                              | User-supplied Android resource decoder.                                                                                                                                                                                     |
 | Binwalk                        | [MIT](https://github.com/ReFirmLabs/binwalk/blob/4fdab3d464d97b68e0af9088df3f9e2e1545b21c/LICENSE)                                                                                                                                                       | v3.1.0; pinned submodule and user-supplied firmware engine.                                                                                                                                                                 |
 | Unblob                         | [MIT](https://github.com/onekey-sec/unblob/blob/1fcc7a0a584a70a96c31f5a276c20944d199a089/LICENSE)                                                                                                                                                        | 26.6.4; pinned submodule and user-supplied firmware engine.                                                                                                                                                                 |

bridge/hopper_bridge.py

@@ -1057,6 +1057,10 @@ def _serve_connection(connection):
                 ):
                     raise PermissionError("Invalid bridge capability")
                 authenticated = True
+                # The connection grace period only bounds unauthenticated idle
+                # clients. Native analysis and a connected owner's idle time
+                # remain unbounded after capability validation.
+                connection.settimeout(None)
                 _write_message(file, {"id": request_id, "event": {
                     "type": "progress",
                     "phase": "hopper_bridge",
@@ -1148,13 +1152,17 @@ def _run():
     server.bind(REA_SOCKET)
     os.chmod(REA_SOCKET, 0o600)
     server.listen(1)
+    # A bridge can start after its caller's startup/cleanup deadline. Bound
+    # the first accept as well as reconnects so an orphan cannot pin Hopper's
+    # dedicated Python thread and prevent later documents from starting REA.
+    server.settimeout(REA_RECONNECT_TIMEOUT_SECONDS)
     try:
         while True:
             try:
                 connection, _ = server.accept()
             except socket.timeout:
                 break
-            connection.settimeout(None)
+            connection.settimeout(REA_RECONNECT_TIMEOUT_SECONDS)
             if _serve_connection(connection):
                 break
             # Match REA's cleanup handshake budget without pinning Hopper's

bridge/native/rea_lldb_tracer.py

@@ -555,6 +555,38 @@ def _failed_launch_description(process):
     return None
 
 
+def _retire_failed_launch(process, config, listener, stdout, stderr):
+    """Reconcile an SBProcess returned alongside a failed Launch error."""
+    if not process.IsValid():
+        stdout.close()
+        stderr.close()
+        return False
+
+    pid = process.GetProcessID()
+    if pid > 0:
+        # Even an already-exited process needs its PID published so the host
+        # can distinguish a completed launch failure from an unknown target.
+        with open(config["pid_path"], "w") as handle:
+            handle.write(str(pid))
+
+    if process.GetState() != lldb.eStateExited:
+        stdout.mark_launched()
+        stderr.mark_launched()
+        if pid > 0:
+            # Hold a surviving target until the host has captured its start
+            # identity. If that proof is unavailable, Kill still addresses
+            # this exact SBProcess created by the current launch request.
+            _wait_for_identity_ack(config)
+        process.Kill()
+        _wait_for_exit(listener, process, 5)
+
+    exit_observed = process.GetState() == lldb.eStateExited
+    expected_exit = exit_observed
+    stdout.close(expected_exit=expected_exit)
+    stderr.close(expected_exit=expected_exit)
+    return exit_observed
+
+
 def trace(debugger, config):
     global _TRACE_CONTEXT
     debugger.SetAsync(True)
@@ -630,8 +662,9 @@ def trace(debugger, config):
             )
         process = target.Launch(_launch_info(target, config, listener), error)
         if not error.Success() or not process.IsValid():
-            stdout.close()
-            stderr.close()
+            exit_observed = _retire_failed_launch(
+                process, config, listener, stdout, stderr
+            )
             return {
                 "status": "launch-error",
                 "error": error.GetCString(),
@@ -640,15 +673,26 @@ def trace(debugger, config):
                     if process.IsValid()
                     else {}
                 ),
+                **(
+                    {"target_process_exit_observed": True}
+                    if exit_observed
+                    else {}
+                ),
             }
         launch_failure = _failed_launch_description(process)
         if launch_failure is not None:
-            stdout.close(expected_exit=True)
-            stderr.close(expected_exit=True)
+            exit_observed = _retire_failed_launch(
+                process, config, listener, stdout, stderr
+            )
             return {
                 "status": "launch-error",
                 "error": launch_failure,
                 "target_process_created": True,
+                **(
+                    {"target_process_exit_observed": True}
+                    if exit_observed
+                    else {}
+                ),
             }
         stdout.mark_launched()
         stderr.mark_launched()

bridge/native/test_rea_lldb_tracer.py

@@ -15,6 +15,8 @@ fake_lldb = sys.modules.setdefault("lldb", types.ModuleType("lldb"))
 fake_lldb.LLDB_INVALID_ADDRESS = -1
 fake_lldb.eLaunchFlagDebug = 2
 fake_lldb.eLaunchFlagStopAtEntry = 4
+fake_lldb.eStateExited = 10
+fake_lldb.SBEvent = lambda: object()
 spec = importlib.util.spec_from_file_location(
     "rea_lldb_tracer", os.path.join(os.path.dirname(__file__), "rea_lldb_tracer.py")
 )
@@ -23,6 +25,136 @@ spec.loader.exec_module(tracer)
 
 
 class NativeTraceBudgetTests(unittest.TestCase):
+    def test_failed_launch_does_not_infer_exit_from_a_failed_kill(self):
+        class Process:
+            def IsValid(self):
+                return True
+
+            def GetProcessID(self):
+                return 12345
+
+            def GetState(self):
+                return 6
+
+            def Kill(self):
+                return None
+
+        class Sink:
+            close_exit = None
+
+            def mark_launched(self):
+                pass
+
+            def close(self, expected_exit=False):
+                self.close_exit = expected_exit
+
+        with tempfile.TemporaryDirectory() as directory:
+            stdout, stderr = Sink(), Sink()
+            with patch.object(tracer, "_wait_for_identity_ack", return_value="unavailable"), patch.object(
+                tracer, "_wait_for_exit", return_value=None
+            ):
+                exited = tracer._retire_failed_launch(
+                    Process(), {"pid_path": os.path.join(directory, "pid")}, object(), stdout, stderr
+                )
+        self.assertFalse(exited)
+        self.assertFalse(stdout.close_exit)
+        self.assertFalse(stderr.close_exit)
+
+    def test_failed_launch_publishes_pid_and_retires_live_process(self):
+        class Process:
+            state = 6
+            killed = False
+
+            def IsValid(self):
+                return True
+
+            def GetProcessID(self):
+                return 12345
+
+            def GetState(self):
+                return self.state
+
+            def Kill(self):
+                self.killed = True
+                self.state = tracer.lldb.eStateExited
+
+        class Sink:
+            launched = False
+            close_exit = None
+
+            def mark_launched(self):
+                self.launched = True
+
+            def close(self, expected_exit=False):
+                self.close_exit = expected_exit
+
+        with tempfile.TemporaryDirectory() as directory:
+            pid_path = os.path.join(directory, "pid")
+            process = Process()
+            stdout, stderr = Sink(), Sink()
+
+            def acknowledge(config):
+                with open(config["pid_path"], "r") as pid_file:
+                    self.assertEqual(pid_file.read(), "12345")
+                return "readable"
+
+            with patch.object(tracer, "_wait_for_identity_ack", acknowledge):
+                exited = tracer._retire_failed_launch(
+                    process,
+                    {"pid_path": pid_path},
+                    object(),
+                    stdout,
+                    stderr,
+                )
+
+        self.assertTrue(exited)
+        self.assertTrue(process.killed)
+        self.assertTrue(stdout.launched)
+        self.assertTrue(stderr.launched)
+        self.assertTrue(stdout.close_exit)
+        self.assertTrue(stderr.close_exit)
+
+    def test_failed_launch_records_when_valid_process_is_already_exited(self):
+        class Process:
+            def IsValid(self):
+                return True
+
+            def GetProcessID(self):
+                return 12345
+
+            def GetState(self):
+                return tracer.lldb.eStateExited
+
+            def Kill(self):
+                raise AssertionError("an exited process must not be killed")
+
+        class Sink:
+            close_exit = None
+
+            def mark_launched(self):
+                raise AssertionError("an exited process has no output to drain")
+
+            def close(self, expected_exit=False):
+                self.close_exit = expected_exit
+
+        with tempfile.TemporaryDirectory() as directory:
+            pid_path = os.path.join(directory, "pid")
+            stdout, stderr = Sink(), Sink()
+            with patch.object(
+                tracer,
+                "_wait_for_identity_ack",
+                side_effect=AssertionError("exit is already observed"),
+            ):
+                exited = tracer._retire_failed_launch(
+                    Process(), {"pid_path": pid_path}, object(), stdout, stderr
+                )
+            with open(pid_path, "r") as pid_file:
+                self.assertEqual(pid_file.read(), "12345")
+
+        self.assertTrue(exited)
+        self.assertTrue(stdout.close_exit)
+        self.assertTrue(stderr.close_exit)
+
     def test_pre_target_integrity_failure_reports_no_process_created(self):
         original_debugger = getattr(tracer.lldb, "SBDebugger", None)
 

docs/.vitepress/config.ts

@@ -60,6 +60,7 @@ export default defineConfig({
         collapsed: true,
         items: [
           { text: "IDA", link: "/ida-provider" },
+          { text: "Kuna", link: "/kuna-provider" },
           { text: "Ghidra on Windows", link: "/windows-ghidra-p0" },
           { text: "Ghidra DOS analysis", link: "/ghidra-dos" },
           { text: "Ghidra raw binary", link: "/raw-binary" },

docs/adb-device-analysis.md

@@ -108,6 +108,13 @@ The `replaced` result is `null`: the successful install response does not say
 whether the package was installed fresh or replaced, and REA does not query the
 device first. The install verdict and host APK digest remain available.
 
+Push and install consume a private snapshot of the selected local input. The
+reported size and SHA-256 identify those consumed bytes; `local_path` and
+`apk_path` retain the original input path, whose contents may change later.
+Snapshot creation needs temporary disk space proportional to the input size.
+Failed snapshot cleanup remains owned and is retried before later operations
+and during shutdown.
+
 ## Interpret the result
 
 - **Directory names retain whitespace and newlines.** Symlink targets do too,
@@ -168,6 +175,9 @@ device first. The install verdict and host APK digest remain available.
 - Cancellation during a package pull preserves caller-selected output. The
   error includes partial observations identifying completed artifacts and the
   interrupted local path; remaining APKs were not attempted.
+- Screen capture checks cancellation before publishing its exclusive output.
+  If cancellation arrives after publication, the error retains the committed
+  file's path, digest and dimensions; the file remains available as evidence.
 - Bugreports collect into isolated staging and publish with exclusive creation;
   filename collisions select the next numeric suffix. If publication fails, the
   completed archive remains in staging. The error's partial observation gives

docs/binary-diagnostics.md

@@ -147,6 +147,24 @@ received signal alone does not establish its exact cause. Dynamic tags require
 a complete DT_NULL inside PT_DYNAMIC; interpreter names and ranges end at the
 first NUL, with any padding still represented by the original segment range.
 
+## Ghidra process failures
+
+When output from the failing Ghidra startup or operation reports
+`java.lang.OutOfMemoryError`, REA returns `resource_constraint`. Java heap or GC
+overhead exhaustion includes heap-specific recovery advice. Metaspace and array
+size failures do not recommend increasing the heap; native-thread exhaustion is
+reported as a capacity constraint with unknown limits.
+
+`configured_jvm_max_heap` records the explicit `-Xmx` value selected by REA's
+launch builder. It is a configuration fact, not measured memory use or proof of
+the effective JVM limit; JVM environment options may affect that limit.
+When a Windows script wrapper selects the JVM, this value remains unknown.
+Historical stdout and stderr remain available, but classification uses only the
+failing operation's `failure_output`. Output draining waits at most one second,
+subject to earlier cancellation or the startup deadline. `output_incomplete`
+marks a snapshot taken before stream closure; cleanup ownership remains with
+the session until release succeeds.
+
 ## Portable PE resources
 
 `inspect_pe_resources` / `inspect-pe-resources` inspects an explicit PE32 or

docs/browser-observation.md

@@ -66,14 +66,25 @@ Electron `file://` pages use a separate provider and target boundary; see [elect
 
 ## Provider authorities
 
-CDP page capture, V8 Inspector observation, and Playwright scenarios stay
-separate providers because their authorities differ: CDP and V8 attach to an
-already-running target and never launch, drive, evaluate, or mutate it, while
-Playwright providers own and drive the runtime (launch browsers, run actions,
-capture step snapshots). Merging them would mix attach-only and owned-process
-lifecycles in one contract. What they share instead is one exclusion
-vocabulary (`BrowserExclusionReason`): every denial reason in every stack maps
-onto it, even where a wire schema keeps a narrower historical bucket.
+CDP page capture and V8 Inspector observation attach to an already-running
+target without launching it, driving actions, or evaluating target code.
+Playwright scenarios drive actions and capture step snapshots in either a
+provider-owned browser or a caller-selected external browser. Separate providers
+preserve these authority and lifecycle differences. They share one exclusion
+vocabulary (`BrowserExclusionReason`): every denial reason maps onto it, even
+where a wire schema keeps a narrower historical bucket.
+
+CDP attachment is not a guarantee that the browser preserves all state.
+Chromium can clear another client's `navigator.platform` emulation when REA's
+page session disconnects, even if that session sent no emulation commands.
+This affects page inspection, bundle capture, session observation, screenshots,
+WebMCP discovery, and Electron page inspection; target listing alone does not
+attach a page session. REA cannot read the other client's complete
+user-agent/client-hint configuration and does not replay guessed settings.
+Use a disposable browser when preserving externally configured emulation is
+required, or have the owning client reapply its complete configuration afterward.
+See [the reproduction and Chromium ownership limitation](https://github.com/morluto/rea/issues/4089).
+
 Existing static application Evidence and passive web/Electron captures can be
 combined later through
 [JavaScript static/runtime reconciliation](javascript-runtime-reconciliation.md).
@@ -343,8 +354,10 @@ an invalid supplied size remains unknown with incomplete resource coverage while
 the resource's other facts are retained. Zero and nonnegative fractional sizes
 are preserved.
 
-CDP discovery, connection, and commands wait for the peer or caller cancellation;
-a stalled peer can leave an operation waiting. WebSocket messages remain subject
+CDP discovery has a 10-second deadline. Each WebSocket handshake and command
+response has its own 20-second deadline, separate from the caller-selected
+observation window. A stalled handshake or command reports `provider_timeout`;
+caller cancellation still reports cancellation. WebSocket messages remain subject
 to the [64 MiB limit](#request-boundary). REA does not impose a message-count or
 pending-command quota. On cancellation, it closes only its own WebSocket,
 waiting up to one second before terminating it. Malformed unsolicited events

docs/browser-scenario-contract.md

@@ -144,6 +144,24 @@ distinction: it closes and deletes provider-owned profiles, but only disconnects
 from an external CDP browser. Real-browser verification checks both outcomes and
 confirms that an attached external browser remains alive.
 
+Keeping that browser alive does not guarantee its emulation state survives.
+Chromium can clear caller-owned `navigator.platform` emulation when a page CDP
+session disconnects, including sessions that sent no emulation commands. REA
+cannot snapshot the other client's complete user-agent/client-hint settings and
+does not replay a partial reconstruction. Use launch mode to keep this effect
+inside a provider-owned browser, or have the external owning client reapply its
+complete configuration after capture. Connected captures report this limitation;
+see [issue #4089](https://github.com/morluto/rea/issues/4089).
+
+`environment.locale` sets Intl formatting in both modes. Launch also sets
+`navigator.language`, `navigator.languages`, and the request `Accept-Language`.
+Connect preserves the caller's page and request languages; every attached
+capture identifies this locale limitation, including retained partial
+observations. Use launch mode when the application must select the requested
+language through navigator values or request headers. Artifact completeness
+covers the selected observations, not equivalence of launch and connect
+language environments.
+
 HTTP(S) URLs may include ordinary query values and fragments directly. Use
 ordered structured query entries when a value needs to reference a declared
 environment-backed secret. Form, storage, and cookie values are either literal

docs/cli.md

@@ -95,8 +95,9 @@ dependency edges, and the section's file range. A missing `.dep-v0` section is
 null metadata, not proof that the binary has no dependencies. See
 [Rust binaries](rust-binaries.md).
 
-Configure [Hopper or Ghidra](installation.md#hopper), or the
-[IDA adapter](ida-provider.md), before analyzing a native target. Substitute
+Configure [Hopper or Ghidra](installation.md#hopper), the
+[IDA adapter](ida-provider.md), or [Kuna](kuna-provider.md) before analyzing a
+native target. Substitute
 your target, search text and function name or address in these examples:
 
 ```bash
@@ -167,6 +168,12 @@ be verified. See [MCP contracts](mcp-contracts.md) for session lifecycle and
 [Ghidra first-query deadlines](mcp-contracts.md#ghidra-first-query-deadlines-and-recovery)
 for import, client timeouts and recovery.
 
+If one-shot analysis fails and closing its session also fails, the original
+error details remain inline and `details.cleanup_error` describes the later
+close failure. Each diagnostic retains its own reported resources; the later
+failure does not establish whether an earlier resource was released. Captured
+output and partial observations remain in the original details.
+
 ## Save and reuse analysis snapshots
 
 A snapshot retains successful analysis results for later queries. REA reuses

docs/electron-observation.md

@@ -2,6 +2,12 @@
 
 REA can attach to an Electron/Chromium CDP endpoint and inspect existing `file://` renderer pages without evaluating JavaScript or invoking Electron APIs. The request supplies the endpoint directly; REA accepts only literal-port loopback HTTP endpoints. Selecting an endpoint exposes every Electron target it serves, including local file paths and page metadata.
 
+Disconnecting the inspection's page session can clear caller-owned
+`navigator.platform` emulation inside Chromium. REA cannot restore another
+client's complete user-agent/client-hint configuration. Use a disposable target
+or let the owning client reapply its configuration; see the shared
+[CDP state limitation](browser-observation.md#provider-authorities).
+
 This passive runtime surface is distinct from the target-free static
 [`analyze_javascript_application`](javascript-artifact-reconstruction.md)
 workflow. Static analysis reads a local ASAR or extracted directory without
@@ -116,6 +122,12 @@ If cleanup cannot be verified, the error retains collected observations in
 `details.partial_observation` and marks their application cleanup `unverified`.
 Cleanup uncertainty does not authorize signaling unrelated processes.
 
+### Target fuse prerequisites
+
+Active scenario capture requires that the target Electron executable honors Node CLI inspector flags (`--inspect` or `--remote-debugging-port`) so that Playwright can establish its DevTools protocol session.
+
+When an Electron executable explicitly disables the `EnableNodeCliInspectArguments` fuse (or removes the capability), `capture-electron-scenario` fails immediately before process launch with an `AnalysisUnsupportedTargetError` ("unsupported_target") rather than waiting for the startup deadline. For targets with disabled inspector fuses, use passive CDP inspection via `inspect_electron_page` or run against an unrestricted development binary with the extracted `app.asar`.
+
 The real verifier exercises compiled CLI and stdio MCP discovery, page/script
 inspection, static/runtime reconciliation, active observation, and CLI SIGTERM
 cleanup before checking the IPC, renderer crash/reload, utility/child-process,

docs/ghidra-nativeaot.md

@@ -69,6 +69,12 @@ Pass a returned MethodTable address to `inspect_native_data_type`. A direct PE
 query also derives the snapshot metadata from the authenticated Ghidra session
 handshake, so it does not require a preceding load-image call. In a persistent
 MCP session, load-image followed by type inspection reuses that same derivation.
+The bridge resolves the caller's address spelling once; metadata lookup uses
+its canonical default-space identity. Bare hexadecimal, prefixed hexadecimal,
+and encoded default-space selectors therefore select the same MethodTable.
+Other address spaces and unresolved identities are not matched by offset alone.
+Metadata already supplied by the engine is retained without reading or deriving
+an additional source snapshot.
 The database result remains `unavailable` when Ghidra has no DataType at that
 address; its `metadata_recovery.source` is `read-only-derived-overlay` and
 carries related type, interface, slot, and target observations. Generated

docs/ghidra-provider.md

@@ -139,6 +139,12 @@ overview merely because its source artifact is unchanged.
 
 ## Admitted function-analysis semantics
 
+Concurrent callers share one startup while retaining independent cancellation.
+Only the final departing caller abandons startup and waits for its cleanup.
+New attempts wait for that cleanup; failed cleanup retains its original error
+and owned directories until a successful close. Closing the provider reports a
+provider-state failure to live callers rather than cancelling their signals.
+
 | Concern                 | Ghidra contract                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                            |
 | ----------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
 | Decompiler lifetime     | One persistent `DecompInterface` is opened for the imported Program and disposed during bridge shutdown. Native decompilation has no fixed per-function deadline. External functions or functions without bodies return `null`; cancellation and native failure remain distinct.                                                                                                                                                                                                                                                                                                                                           |
@@ -193,7 +199,11 @@ these tools before compilation. Keeping this matrix separate lets Linux
 host/provider acceptance run with only the host compiler.
 
 `npm run verify:ghidra:windows` checks source-owned native x86-64 PE applications;
-`-- --x86` selects x86 and `-- --dll` selects DLLs, and the flags can be combined.
+`-- --x86`, `-- --arm`, or `-- --arm64` selects another target architecture;
+`-- --dll` selects DLLs. Choose one architecture and optionally combine it with
+`--dll`. These static fixtures contain source-owned encoded instructions; they
+are not compiler-produced programs. See the [Windows verification scope](windows-ghidra-p0.md#native-authority-and-target-identity)
+for real-provider coverage and exclusions.
 These lanes check all 26 read-only operations, target/snapshot/import SHA-256 linkage,
 and project, endpoint, process, and runtime cleanup. The independent native
 lane checks DACLs, handle admission, cancellation, and Job Object lifecycle.

docs/go-binaries.md

@@ -39,8 +39,8 @@ container format, architecture, word size and byte order. A recognized
 - `module.dependencies`, with the same fields and their original order;
 - `module.settings`, retaining ordered key/value entries and duplicate keys;
 - `module.unparsed_lines`, `module.unparsed_line_bytes_base64` and
-  `module.complete`, so unfamiliar, malformed, conflicting or undecodable records
-  do not disappear behind a complete result.
+  `module.complete`, which says whether every present module-text line was
+  parsed. It does not establish that the dependency inventory is complete.
 
 Missing checksums are `null`; empty fields and development/local replacement
 versions remain unchanged. Build settings such as `GOOS`, `GOARCH` or `vcs.revision`
@@ -50,6 +50,13 @@ base64 while other decodable module records are still reported; the reader does
 not insert replacement characters or reject the entire artifact for those
 strings.
 
+When no main-module (`mod`) record was decoded, `module.main` is `null` and
+dependency coverage is unknown. Successfully decoded dependency records remain
+visible. GOPATH-mode builds may contain only a package `path` and build settings;
+their empty dependency array does not establish the absence of dependencies.
+The result's limitations state this explicitly. A module-mode build with a main
+module record and no dependency records has an observed empty dependency list.
+
 ## Supported boundary
 
 The built-in TypeScript reader supports ELF, PE and thin Mach-O images with Go's

docs/index.md

@@ -50,7 +50,7 @@ For runtime instrumentation, see the [Frida guide](./frida-instrumentation.md).
 | Task                                                | Guides                                                                                                                                                                                                                                                                                                                      |
 | --------------------------------------------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
 | Inspect native functions, references, or types      | [Native binaries](native-investigation.md), [Go metadata](go-binaries.md), [Rust dependencies](rust-binaries.md), [offline ELF layout](binary-diagnostics.md), [recorded crashes](recorded-crashes.md)                                                                                                                      |
-| Select a native provider or import format           | [Ghidra semantics](ghidra-provider.md), [Windows](windows-ghidra-p0.md), [DOS](ghidra-dos.md), [raw bytes](raw-binary.md), [MIPS](ghidra-mips.md), [PSP](ghidra-psp.md), [NativeAOT](ghidra-nativeaot.md), [IDA](ida-provider.md), [JEB](jeb-analysis.md)                                                                   |
+| Select a native provider or import format           | [Ghidra semantics](ghidra-provider.md), [Windows](windows-ghidra-p0.md), [DOS](ghidra-dos.md), [raw bytes](raw-binary.md), [MIPS](ghidra-mips.md), [PSP](ghidra-psp.md), [NativeAOT](ghidra-nativeaot.md), [IDA](ida-provider.md), [Kuna](kuna-provider.md), [JEB](jeb-analysis.md)                                         |
 | Analyze JavaScript or Electron artifacts            | [Static analysis](javascript-artifact-reconstruction.md), [tracing and comparisons](javascript-application-workflows.md), [application graph](javascript-application-graph.md), [source recovery](javascript-recovery.md)                                                                                                   |
 | Compare static JavaScript with runtime observations | [Runtime attribution](javascript-runtime-observation.md), [reconciliation](javascript-runtime-reconciliation.md), [Electron observation](electron-observation.md), [Electron fuses](electron-fuses.md)                                                                                                                      |
 | Inspect assemblies, packages, or firmware           | [.NET](managed-code-analysis.md), [Android](android-analysis.md), [Apktool resources](apktool-resource-analysis.md), [ADB acquisition](adb-device-analysis.md), [Apple](apple-application-analysis.md), [Harmony](harmony-analysis.md), [firmware](firmware-analysis.md), [WASM](wasm-artifacts.md), [EVM](evm-bytecode.md) |

docs/installation.md

@@ -3,13 +3,13 @@
 Run REA, connect your selected agents, and check the tools needed for your task.
 Setup previews configuration changes and asks for approval before applying them.
 
-| Task                              | Section                                                                   |
-| --------------------------------- | ------------------------------------------------------------------------- |
-| Connect an agent                  | [Start setup](#start-setup), then [review changes](#review-setup-changes) |
-| Diagnose one task                 | [Scoped readiness checks](#check-readiness-for-your-task)                 |
-| Configure an analysis engine      | [Hopper](#hopper), [Ghidra](#ghidra), or the [IDA guide](ida-provider.md) |
-| Update or remove REA              | [Maintenance](#diagnose-update-and-remove)                                |
-| Resolve a client-specific problem | [Client configuration](client-configuration.md)                           |
+| Task                              | Section                                                                                             |
+| --------------------------------- | --------------------------------------------------------------------------------------------------- |
+| Connect an agent                  | [Start setup](#start-setup), then [review changes](#review-setup-changes)                           |
+| Diagnose one task                 | [Scoped readiness checks](#check-readiness-for-your-task)                                           |
+| Configure an analysis engine      | [Hopper](#hopper), [Ghidra](#ghidra), the [IDA guide](ida-provider.md), or [Kuna](kuna-provider.md) |
+| Update or remove REA              | [Maintenance](#diagnose-update-and-remove)                                                          |
+| Resolve a client-specific problem | [Client configuration](client-configuration.md)                                                     |
 
 ## Start setup
 
@@ -179,6 +179,8 @@ as well, so it is broader than this Claude-only migration.
 history under `CLAUDE_CONFIG_DIR`. Setup follows the same
 variable: it writes the registration to `$CLAUDE_CONFIG_DIR/.claude.json` and
 the skill to `$CLAUDE_CONFIG_DIR/skills`.
+For REA setup, an empty `CLAUDE_CONFIG_DIR` is treated as unset: registration
+goes to `~/.claude.json` and the skill goes to `~/.claude/skills`.
 
 ```bash
 CLAUDE_CONFIG_DIR=~/.claude-rea npx rea-agents setup --client claude_code
@@ -398,8 +400,9 @@ running server or its registration.
 The npm release checked on **2026-10-11** was **6.4.0**, published from
 [`a56f7bb6`](https://github.com/morluto/rea/commit/a56f7bb668ab23fec8d33f7af353ac7ed2fe1895).
 The actual registry package accepted dry-run setup plans for the 17 client IDs
-in that release, including Qwen Code, Pi and Hermes. DeepSeek Harness (`dsh`)
-is included from 6.5.0 and was absent from the checked npm 6.4.0 package.
+in that release, including Qwen Code, Pi and Hermes. Seven client IDs are
+included from 6.5.0 and were absent from the checked npm 6.4.0 package: `droid`,
+`dsh`, `kimi`, `qoder`, `trae`, `workbuddy` and `zcode`.
 The 6.4.0 registry check does not establish 6.5.0 publication or live DeepSeek
 API integration.
 OpenCode's plan selected
@@ -482,8 +485,8 @@ hosted connector; see [its configuration](client-configuration.md#other-client-s
 | ZCode              | `zcode`          |
 | DeepSeek Harness   | `dsh`            |
 
-`dsh`, `kimi`, and `workbuddy` require a release containing them. The registry
-check above does not cover them.
+`droid`, `dsh`, `kimi`, `qoder`, `trae`, `workbuddy`, and `zcode` require
+`rea-agents@6.5.0` or later. The 6.4.0 registry check above does not cover them.
 
 See [client configuration and troubleshooting](client-configuration.md) for
 configuration paths, profiles, trust policies, and schema-limit workarounds.
@@ -634,6 +637,11 @@ bridge readiness. For large targets, set `REA_HOPPER_STARTUP_TIMEOUT_MS` to a
 decimal integer from 1 to 2,147,483,647 milliseconds in the environment that
 starts REA. CLI and MCP use the same setting; invalid values fail configuration
 validation. An MCP client's own request timeout must also allow enough time.
+`open_binary` validates and selects a target without waiting for native provider
+readiness. During Hopper startup, `binary_session.provider_operation_health`
+reports `busy` at stage `launch`; `analysis_activity` describes bridge requests
+and can remain idle until the socket is ready. Cleanup work reports `busy` at
+stage `connection`.
 
 If macOS denies Apple Events, allow the application that starts REA under
 **System Settings → Privacy & Security → Automation**, then retry. REA releases
@@ -643,9 +651,12 @@ Arguments containing quotes, backslashes, or line breaks remain unchanged but
 cannot establish that a failed handoff opened no document. Other failed handoffs
 retain ownership until document closure is confirmed. If a timed-out bridge
 becomes ready later, `close_binary` connects to that run, authenticates its
-identity, and requests document shutdown. After a connection drops, the bridge
-allows 30 seconds for an authenticated cleanup reconnect, then releases Hopper's
-Python execution thread. Expiry does not establish document closure, so REA
+identity, and requests document shutdown. Once its listener is ready, the bridge
+allows 30 seconds for the first connection and capability validation. After a
+connection drops, it allows the same grace period for an authenticated cleanup
+reconnect, then releases Hopper's Python execution thread. Initial analysis is
+outside this listener grace period, and authenticated native work is not bounded
+by it. Expiry does not establish document closure, so REA
 retains unconfirmed ownership. Retry close while the bridge remains available.
 
 If a Linux launcher exists but cannot start, inspect its shared libraries:
@@ -735,9 +746,17 @@ Once document and process closure are confirmed, the lease can be released even
 if temporary-file cleanup fails. Another client can then launch, but the owning
 client must finish its retained cleanup before starting again.
 An unconfirmed external document can be retried while its authenticated bridge
-remains connected. If that bridge has disconnected, another close cannot confirm
-the document: inspect and close the reported document in Hopper before ending
-the owning REA connection.
+remains available. The same authentication and process-identity capture apply
+when the first successful handshake occurs during cleanup after a startup
+timeout. REA can confirm closure if that exact authenticated process exits,
+including while a shutdown request is in flight. A helper exit alone is
+insufficient, and an application that never authenticated has no verified
+process identity. If no matching bridge becomes available, inspect and close
+the reported document in Hopper before ending the owning REA connection.
+Export the Evidence bundle before `close_binary` clears session records.
+`binary_session.cleanup_pending` remains true while provider cleanup is retained,
+and provider health remains visible even with `open: false`; that flag alone
+does not establish successful cleanup.
 
 ### Hopper in CI
 
@@ -796,7 +815,7 @@ and `set_addresses_names` name a function entry or any other mapped address
 session metadata edits leave executable bytes unchanged and are discarded on close. GUI controls
 require Hopper; Windows P0 remains read-only.
 
-Windows P0 admits native x86 and x86-64 PE applications and DLLs, plus explicitly
+Windows P0 admits native x86, x86-64, ARMNT (Thumb-2) and ARM64 PE applications and DLLs, plus explicitly
 selected [raw NMOS 6502 images](raw-binary.md#nmos-6502) with a caller-supplied
 entry at file offset 0, on fixed local NTFS volumes. Other raw CPU families remain
 unsupported on Windows.

docs/javascript-application-graph.md

@@ -68,6 +68,17 @@ The node vocabulary covers:
 - runtime script instance;
 - an explicit `unknown` kind when classification is not justified.
 
+Static network endpoints can have `resolution: "partial"` when the analyzer
+recognizes a request call but cannot resolve its target to one exact value.
+The observation's `value` then preserves the source expression, and its
+evidence points to the call site. Partial and exact endpoints have distinct
+identities even when their displayed values match. Partial identities also retain
+the source site: matching expression text at different sites does not establish
+the same runtime target. This records observed source syntax; it does not
+establish that the call ran or contacted that endpoint.
+Graph coverage describes the analyzed artifact scope and resource limits, not
+runtime execution. A partial target alone does not imply omitted source.
+
 The directed relation vocabulary is `contains`, `loads`, `imports`, `maps_to`,
 `exposes`, `sends`, `invokes`, `handles`, `calls`, `persists_to`, `observed_as`,
 and `changed_from`. Provider names are deliberately absent from both

docs/javascript-application-workflows.md

@@ -6,7 +6,7 @@ are `inspect_analysis_view`, `trace_application_feature`,
 `trace_javascript_semantics`, `compare_application_versions`,
 `compare_source_to_bundle`, and `compare_javascript_export_shapes`; their CLI
 equivalents use the same names with hyphens. `inspect_analysis_view` projects a
-summary, module page, or one module from already completed application
+summary, module page, typed module references, or one module from already completed application
 Evidence; it does not walk relationships.
 
 `inspect_analysis_view` also projects retained `inspect_binary_layout`
@@ -40,6 +40,46 @@ combine its fields with the fact's location. Semantic trace results include the
 canonical context subset referenced by their returned facts, so those facts
 remain self-contained in the response.
 
+## Retained module facts
+
+For canonical identities and retained source-range counts, query the existing
+`inspect_analysis_view` tool with a typed module-reference projection:
+
+```json
+{
+  "source": {
+    "kind": "retained-evidence",
+    "evidence_id": "ev_ACTUAL_PARENT_ID"
+  },
+  "view": {
+    "kind": "javascript",
+    "collection": "modules",
+    "projection": "references",
+    "where": { "path": "main.js", "kind": "javascript-module" },
+    "offset": 0,
+    "limit": 16
+  }
+}
+```
+
+Replace the parent ID with the actual analysis summary's `parent_evidence_id`.
+Filters match exact retained paths and kinds. Omission includes all values;
+`path: null` selects an unavailable path and `path: ""` selects a recorded empty
+path. Multiple identities can share a path. The default offset is 0 and limit
+is 64; follow `coverage.next_offset` for later matching rows in retained order.
+
+Each row contains its `node_id`, kind, path, canonical identity, retained
+source-range observation count, and distinct supporting authority/state/confidence
+combinations in `basis`. These counts do not prove unique reference sites or
+runtime loading. Use the existing module item view with the row's `node_id` for
+exact source ranges and complete observations. The result retains parent/query
+Evidence identity, artifact identity, limitations and unknowns. `source_coverage`
+keeps selected/unselected sources, graph coverage and unknown counts separate
+from page exhaustion; an exhausted or empty page cannot prove absence from an
+incompletely analyzed application. CLI `inspect-analysis-view` reads the same
+request from a JSON file; portable input uses `source.kind: "inline"` with the
+complete parent Evidence.
+
 ## Feature tracing
 
 Select one literal seed kind: node ID, route, string, API, IPC channel, module,
@@ -55,6 +95,14 @@ matches exactly. Otherwise the result recommends provider-neutral follow-up
 tools and reports `requires-provider-analysis`; it never starts or switches a
 provider implicitly.
 
+A handoff requires the add-on's exact content identity. Unavailable unpacked
+companions remain visible in the trace graph with their canonical path,
+availability and container-backed Evidence. They yield no actionable handoff
+until their bytes are available and the application is reanalyzed; the archive
+digest does not identify the missing add-on. An empty handoff list does not prove
+that the trace contains no native components. Source partial coverage remains
+visible independently of handoff eligibility.
+
 ## Semantic relation tracing
 
 `trace_javascript_semantics` queries the authenticated companion graph without
@@ -77,6 +125,17 @@ registrations/removals and dispatch, Node timers and cancellation handles, async
 defaults, request construction and response consumers, parse/coercion/
 validation boundaries, and built-in resource acquisition/release.
 
+Local callback candidates do not establish a complete target set when a
+conditional write may leave a parameter, catch binding, or uninitialized
+variable unchanged. These calls retain candidate edges and dynamic uncertainty;
+default traces remain ambiguous with partial coverage. Opt into
+`include_ambiguous_dynamic_edges` to traverse the candidates. Proven preceding
+unconditional assignments, including writes dominating a read in the same
+conditional block, can resolve a target; later writes do not replace an
+earlier read or captured alias. Unproved ordering across branches or function
+activations remains conservative. Returned-reference effects use the same
+read-position facts as call resolution.
+
 Mutation tracking covers explicit member assignments, updates, deletion, and
 loop assignment targets, including the supported local aliases and shared
 nested references. Calls conservatively invalidate exposed mutable references;
@@ -126,8 +185,13 @@ graph and read literal values from `properties.value`, rather than decoding
 role keys or labels. Previously stored graphs and their original IDs remain
 valid inputs.
 
-Endpoint nodes keep the exact endpoint in each observation's `properties.value`;
-semantic request and response nodes keep it in `properties.endpoint`. Endpoint
+Endpoint observations keep their value in `properties.value`. When
+`properties.resolution` is `partial`, that value is the unresolved source
+expression, with an identity distinct from an exact endpoint of the same text.
+Semantic request and response nodes keep their endpoint in `properties.endpoint`. Endpoint
+lookup identities are preserved independently of the primitive-value lattice's
+byte budget. An oversized normalized literal can become unknown while the same
+string remains an exact endpoint for tracing and comparison. Endpoint
 identity keys use `value-sha256:<digest>` over the canonical JSON string when
 that is shorter, or when the original value starts with the reserved digest
 prefix. Other short endpoint keys remain unchanged. Endpoint labels use the
@@ -208,6 +272,13 @@ Direct return expressions, including expression-bodied arrows, are evaluated
 through an execution-free value lattice. Literal object fields and direct
 return sites are represented in the result. Calls, dynamic spreads, computed
 keys, and parser recovery remain partial or unknown.
+Rebound values are evaluated at their read position: saving an incoming parameter
+before a later assignment preserves caller-dependent uncertainty, while a proven
+preceding assignment can supply a constant. Unresolved ordering remains unknown.
+Supported reads within object and array literals follow left-to-right evaluation;
+an unproved relative order remains unknown. A later write in a separate conditional
+block can still leave an earlier capture unknown because ordering across those
+blocks is not proven.
 Objects and arrays passed to calls, constructors, or tagged templates, used
 as method receivers, or stored through property targets are not assumed
 unchanged afterward. Property stores conservatively retain reference escape

docs/javascript-artifact-reconstruction.md

@@ -182,6 +182,18 @@ partial Evidence. Cancellation after source analysis begins can also retain
 completed facts, including when interrupted during cleanup, graph construction,
 or Evidence sealing. Unexamined relationships remain unknown.
 
+Per-file gaps identify the applicable processing phase: JavaScript source analysis
+uses AST authority, while package/JSON, HTML-reference and source-map processing
+use artifact-byte authority. File-analysis interruption gaps carry `analysis_status`:
+`interrupted` means the phase was entered (including source-worker startup),
+and `not-started` means interruption prevented that phase from beginning.
+Neither state establishes that parsing completed. Remaining parsers are not run
+after interruption just to populate a result.
+
+Native add-ons remain metadata-only and acquire no JavaScript parse or analysis
+gap. This also applies when their paths are outside `source_paths`; their inventory
+and caller selection remain available without implying applicable AST work.
+
 MCP returns a `partial_observation` retained reference after recording succeeds;
 open it with `inspect_analysis_view` on the same connection. If the SDK stopped
 waiting before receiving the reply, follow [MCP cancellation recovery](mcp-contracts.md#progress-and-cancellation)
@@ -276,8 +288,13 @@ RPC client syntax also contributes network endpoint candidates, in both
   named operation found at the start of a string or cooked template literal —
   the name must be followed by variable definitions (`($`), a directive, or a
   selection set starting with a field name — or in a precompiled
-  `OperationDefinition` document object. A graphql-tag `loc.source.body` copy is
-  not reported twice. Anonymous operations are not reported.
+  `OperationDefinition` document object. A standalone Source object's `body`
+  remains an operation candidate even when it has `locationOffset` metadata.
+  A graphql-tag `loc.source.body` copy is suppressed only when that source
+  belongs to a recovered document with the same operation and its containing
+  document, location and source have proven private metadata ownership. Writes,
+  escapes and classic-script global bindings retain the source observation.
+  Anonymous operations are not reported.
 
 Literal link options and precompiled document fields follow object property
 order: the final definition owns the key. A later spread, dynamic computed key,
@@ -294,13 +311,25 @@ suffix. Utility calls such as `useUtils().invoice.list.fetch()`, `useQueries`,
 and links whose callee was renamed by a minifier are not recognized.
 
 Each recovered bundle module retains the exact factory-source digest. A complete
-bounded AST also receives a `babel-ast-v2` structural fingerprint that ignores
-ordinary identifier names while retaining syntax, literals, operators, object
-keys, and member properties. Computed member identifiers retain their access
-syntax without treating a local variable's spelling as a fixed property; template
-literal contents remain part of the fingerprint. Version comparison keeps the
-algorithm in its matching key, so retained `babel-ast-v1` fingerprints cannot match
-these new fingerprints. Re-analyze both snapshots to compare their new structure.
+bounded AST also receives a `babel-ast-v3` structural fingerprint. It ignores
+ordinary identifier spellings while retaining literal and directive values,
+operators, fixed property names, and selected syntax flags such as computed
+keys, declaration kind, function async/generator state, update-expression
+prefix, method kind, and class static state. Local names used in computed keys
+remain rename-resistant; private member names are also ignored while their
+method kind and static state remain represented.
+
+This fingerprint is structural evidence for candidate matching, not proof that
+two modules behave equivalently. Its scope follows the AST facts listed above;
+unrepresented syntax can still collide. The exact source digest remains
+available to show byte-level change, and comparison reports content changes even
+when structural matching succeeds.
+
+Version comparison includes the algorithm in its structural matching key.
+Retained `babel-ast-v1` results can still match other v1 results, and v2 results
+can still match v2 results, but v1/v2 results cannot structurally match v3
+results. Other matching tiers remain available across algorithm versions.
+Re-analyze both snapshots to compare using the v3 rules.
 If fingerprint construction reaches its structural
 bound, the fingerprint is unavailable and its status is `truncated`; a bounded
 prefix is never presented as a complete stable fingerprint.
@@ -397,10 +426,13 @@ For `.node` bindings, member names mean “requested by JavaScript syntax.” A
 resolved add-on path does not convert them into verified native exports. Native
 symbol verification remains a separate deep-analysis claim.
 
-Endpoint observations preserve useful local diagnostics, including query values
-and fragments, while removing only URL username/password credentials. Artifact
-paths, digests, parse locations, and analysis metadata remain actionable
-because REA is local-only.
+Exact endpoint values preserve useful local diagnostics, including query values
+and fragments, while removing URL username/password credentials when the value
+parses as a URL. When a recognized request target cannot be resolved exactly,
+the endpoint's partial resolution retains the original source expression with
+its call-site provenance. This does not establish that the request executed.
+Artifact paths, digests, parse locations, and analysis metadata remain
+actionable because REA is local-only.
 
 ## Safety boundary
 
@@ -432,7 +464,18 @@ retained candidate strings and temporary canonical-key serialization. This is
 an in-memory semantic allocation bound independent of CLI or MCP transport
 budgets. Source literal bytes remain in the parsed source and artifact evidence;
 when a normalized semantic value exceeds the bound, its value is unknown and
-coverage is partial. Expression
+coverage is partial. Direct string literals, cooked templates without
+substitutions, and derived strings all use this normalization boundary. A
+single exact primitive is validated without allocating deduplication or sort
+state or serializing a canonical key. Configuration defaults use the same
+normalizer; limited defaults retain `value_resource_limit` and an anchored
+resource-limit unknown. Known null, false, zero, and empty-string defaults stay
+exact. This primitive-value bound does not truncate lookup identities such as
+source paths, field names or request endpoints. Full endpoint properties remain
+available for endpoint queries, tracing and comparison. Each analysis records encountered limit classifications independently of
+later projections: a selected healthy sibling can remain exact while whole-file
+coverage reports the encountered bound. Projection preserves limiting values
+from the shared property-slot owner. Expression
 evaluation and provenance walks stop after 256 nested levels. Union products
 and string lengths are checked before allocating combinations or concatenated
 strings. When a boundary is reached, the value stays unknown, semantic coverage

docs/javascript-runtime-observation.md

@@ -95,6 +95,11 @@ pauses or resumes execution, calls `Runtime.runIfWaitingForDebugger`, launches
 the target, instruments JavaScript, or invokes Electron APIs. Closing a call
 closes only REA's WebSocket.
 
+Each WebSocket handshake and command response has its own 20-second deadline,
+separate from the caller-selected observation window. A stalled handshake or
+command reports `provider_timeout` and retains observations already received;
+caller cancellation still reports cancellation.
+
 The direct observations are:
 
 - `Debugger.scriptParsed`, establishing that a bounded script location was

docs/javascript-runtime-reconciliation.md

@@ -91,17 +91,20 @@ nodes. REA evaluates compatible static candidates conservatively:
 
 1. A captured-source SHA-256 plus one matching mapped location yields
    a content-and-location match.
-2. One unique exact content digest can match even when no location mapping is
-   available. Whole-asset digests and reconstructed module-factory source
+2. When no static candidate exists at a mapped location, one unique exact
+   content digest can match. Whole-asset digests and reconstructed module-factory source
    digests remain distinguishable in the reported basis.
 3. Without captured bytes, one unique mapped artifact path can produce a
    medium-confidence location match.
 4. Multiple candidates remain `ambiguous`; no `observed_as` edge is emitted.
+   With captured source and mapped candidates, only mapped content matches
+   qualify; unrelated content matches elsewhere are excluded.
    `candidate_static_count` reports the full qualified layer/node count, and
    `candidate_static_nodes` includes every unique candidate reference.
 5. If captured bytes disagree with the static candidate at the mapped path,
-   the result is an explicit mismatch. A path never overrides a digest
-   disagreement.
+   the result is an explicit mismatch. An equal digest elsewhere cannot hide
+   that disagreement. If a mapped candidate's digest is unavailable and no
+   mapped content match is known, the comparison remains `unknown`.
 6. Missing authorization mappings and incomplete or truncated input coverage
    remain unknown or truncated rather than becoming absence claims.
 

docs/kuna-provider.md

@@ -0,0 +1,113 @@
+# Kuna provider
+
+[Kuna](https://github.com/Noelo-Lab/kuna) is an Apache-2.0 decompiler distributed as one `kuna` executable. REA uses a `kuna` you supply. It does not download, install, update or bundle Kuna. REA verified this provider with Kuna 1.833; it accepts other releases and reports them as unverified in `rea doctor` and in each analysis profile.
+
+Kuna is static: it reads the selected image and never executes it.
+
+## Configure
+
+```bash
+export REA_KUNA_COMMAND=/absolute/path/to/kuna
+rea doctor --provider kuna
+```
+
+Select it explicitly with `--provider kuna`, MCP's `open_binary.provider_id: "kuna"`, or `REA_ANALYSIS_PROVIDER=kuna`. With Kuna and another provider configured for the same target, `auto` reports `ambiguous` instead of choosing one.
+
+| Setting                       | Default                                                                                                                                 | Meaning                                                                                                                                                                                  |
+| ----------------------------- | --------------------------------------------------------------------------------------------------------------------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
+| `REA_KUNA_COMMAND`            | none (required)                                                                                                                         | Absolute path of the `kuna` executable.                                                                                                                                                  |
+| `REA_KUNA_CACHE_DIR`          | `~/Library/Caches/rea/kuna` (macOS), `$XDG_CACHE_HOME/rea/kuna` (Linux; `~/.cache/rea/kuna` when `XDG_CACHE_HOME` is unset or relative) | Absolute directory REA owns for project exports. Kuna's own spec cache also lives here, not in your Kuna cache.                                                                          |
+| `REA_KUNA_CACHE_MAX_BYTES`    | 4 GiB                                                                                                                                   | Size cap; least recently used exports are evicted above it.                                                                                                                              |
+| `REA_KUNA_PROJECT_TIMEOUT_MS` | 30 minutes                                                                                                                              | Deadline for whole-program inventory/export, reference and string scans.                                                                                                                 |
+| `REA_KUNA_TIMEOUT_MS`         | 120 s                                                                                                                                   | Deadline for one single-function decompile, and for the inventory run REA makes when the export ends without one.                                                                        |
+| `REA_KUNA_JOBS`               | 1                                                                                                                                       | `decompile-project --jobs`. 1 is Kuna's reference answer; more jobs finish sooner but lose callee-first prototypes, and change about 40–60% of function bodies on the measured binaries. |
+
+Windows is not supported yet. Supported targets are ELF and thin Mach-O executables for x86-64, AArch64 and ARM; PE and universal Mach-O are rejected with a typed reason.
+
+## What Kuna answers
+
+| Tool                                                        | Source                                                                                                                           |
+| ----------------------------------------------------------- | -------------------------------------------------------------------------------------------------------------------------------- |
+| `list_procedures`, `search_procedures`, `procedure_address` | Kuna's callable inventory (`kuna functions --json`), including import slots it does not decompile                                |
+| `address_name`                                              | Function entries, then the data labels Kuna's export assigns                                                                     |
+| `procedure_pseudo_code`, `batch_decompile`                  | The project export, or a single-function run while the export builds                                                             |
+| `procedure_assembly`                                        | The project export's labeled listing                                                                                             |
+| `procedure_callers`, `procedure_callees`                    | Call-kind edges from a lazy, body-free Kuna graph; synthetic import nodes are accepted by address                                |
+| `list_strings`, `search_strings`                            | Typed string candidates from one cached, reference-free scan; exact-address, literal and ECMAScript regex selection              |
+| `xrefs`                                                     | Exact-address incoming reference sites, with forwarding aliases filtered explicitly                                              |
+| `analyze_function`                                          | Pseudocode and assembly from the export, plus call relationships; typed references, comments, strings and CFG remain unavailable |
+
+Typed procedure references, basic blocks, raw-memory tools, annotations and type recovery are not available from this provider. Use another provider for those operations.
+
+## How answers are produced
+
+The first text or assembly request in a session looks for this target's export in the cache. `list_documents` reads the selected path without initializing analysis. Inventory, function lookup, string queries, call relationships and xrefs do not start a C export. A data-name lookup starts it only when the address is absent from the function inventory. The export is keyed by the target's SHA-256, the Kuna executable's SHA-256, its version and `--jobs`.
+
+- **Warm cache:** analysis reads reuse the verified export. A fresh server verifies its file digests once, even for concurrent first requests; that cost scales with export size.
+- **Cold cache:**
+  - REA starts `kuna decompile-project` in the background on a private copy of the target.
+  - Meanwhile, pseudocode comes from one `kuna decompile --addr` run. Name and address lookups, and listings, use the inventory the export takes with `kuna functions --json` before it decompiles. The session runs `kuna functions` itself, under `REA_KUNA_TIMEOUT_MS`, only when no export is building (an inventory-only request does not start one) or the export ends without an inventory.
+  - Requests for the same function share one run; it stops when its last waiting request is cancelled. Each run analyzes the whole image again in its own process, so one REA process runs at most as many at once as the host has cores beyond `REA_KUNA_JOBS` (at least one). Further requests wait in order, and a waiting request is answered from the export if it finishes first. A run whose stop REA cannot prove keeps its slot until `close_binary` proves it stopped.
+  - The session keeps each completed single-function text, up to 16 MiB in all with the least recently used dropped first, and answers repeats from it until the export can. After a failed export that lasts for the session.
+  - Assembly and data names wait for the export; progress notifications report the wait every ten seconds. A session does not rebuild a failed export: requests that need it fail with `retryable: false`, the export's failure, its captured Kuna output and what to change before reopening the target. A setting such as `REA_KUNA_PROJECT_TIMEOUT_MS` or `REA_KUNA_JOBS` is read when REA starts, so changing it also needs a restart. Pseudocode keeps using single-function runs.
+  - A single-function answer says so in its limitations. It lacks the export's whole-program prototypes, so a later request can return different text for the same function.
+
+Pseudocode, assembly, `analyze_function` and data-name answers record where their text came from in `raw_result.source`: `single-function`, or `project-export` with the cache directory. `procedure_pseudo_code` and `procedure_assembly` answers from the export also name the file, byte span and the file's SHA-256. Combined `analyze_function` answers retain both facet records in `raw_result.source.files`, each with its file, digest and byte span. Both texts come from one export; losing that project retries the complete report. Listings and procedure lookups carry no `raw_result`. Agents can open or search the export's `.c` and `.asm` there directly when that is cheaper than calling a tool per function. Text read that way is target-derived, like any decompiler output, but it bypasses the untrusted-content marker that REA's tool results carry.
+
+The background export belongs to the session. `close_binary`, or the end of a CLI command, cancels it unless another session still waits for it. A one-shot CLI decompile therefore answers from a single-function run; any request that needs the export, such as `procedure_assembly` or `analyze_function`, keeps it.
+
+The assembly label index is scanned once per export reader. Cancelling a request stops its wait without interrupting other readers; closing the session aborts and drains the shared scan before releasing its files.
+
+Regex procedure and string search uses ECMAScript Unicode matching in a cancellable worker with a five-second matching deadline. Literal matching remains the default. Cancellation or a matching deadline leaves the session available.
+
+Single-function subprocess output and each cached source facet share a 16 MiB buffered-text capacity. Cached spans are checked before allocation; larger text remains in the local export and the error identifies the file and range. `analyze_function` initially returns the complete dossier, which can still be large; later retained views do not remove that first transfer. Prefer the individual text tools for a focused question.
+
+## Strings
+
+`list_strings` and `search_strings` share one lazy `kuna strings --json --encoding all --min-length 5 --termination any --no-xrefs` scan of the digest-checked private input. Skipping references avoids program lifting, a call graph and a C export. REA validates the input, scan policy, every row and the total before caching any facts. Encoding, character count, byte length and terminator facts must agree; supplementary UTF-8 characters count as Unicode scalars. Malformed or oversized output is an error, not an empty search. JSON admits at most 16 MiB before parsing. String scans use the whole-program `REA_KUNA_PROJECT_TIMEOUT_MS` deadline; `REA_KUNA_TIMEOUT_MS` applies to single-function decompilation. The parsed inventory belongs to one selected target/executable profile and expires on session close; deterministic output-buffer failures are retained until close, while other failures are not cached.
+
+Results keep exact u64 virtual addresses, decoded text, native encoding, terminator status and byte length including any terminator. Kuna scans mapped sections or segments for runs of at least five characters; unterminated fragments can include instruction bytes. `all` includes ASCII, UTF-8 and the engine's Latin-range UTF-16LE reading. Other text, including short strings, non-Latin wide text and UTF-16BE, can remain undiscovered. An empty result does not prove absence.
+
+Use `search_strings` for the matching rows or `list_strings` with an exact address for that location. Literal and regex matching run in REA, preserving the existing case and ECMAScript Unicode semantics instead of substituting Kuna's different regex dialect. Procedure and string regex queries share one worker slot per session. Input projections are materialized only after admission, so queued queries retain no per-query inventory arrays. Close cancels and drains matching. Native metadata queries also share one admission slot, including references; a failed stop retains admission until cleanup is proven.
+
+`raw_result.strings` reports the source JSON digest/size, scanned scope, extraction policy, complete inventory count and selected count without copying unrequested text into the answer. Omit the address from `list_strings` to retrieve the complete inventory as Evidence; normal session retention and bundle export apply, including transport-capacity errors for oversized inline replies. These operations preserve their complete-result contracts and do not add paging. Function-referenced string facets in `analyze_function` remain unavailable; call `xrefs` explicitly when references to one returned string are needed.
+
+## Calls and references
+
+The first call query runs `kuna decompile-graph --functions ''`, admitting no C or assembly bodies. One validated graph and forward/reverse indexes serve later call queries and existing `get_call_graph` / `trace_call_path` workflows. Numeric u64 addresses are parsed from their original JSON tokens. Only `call` edges become callers/callees; jumps and data references retain their distinct meaning. Graph counts, edge endpoints and ordering are validated before any relationship is returned. A forwarding address may be an instruction inside a function, so it is retained separately and never promoted to a function node or call edge. Reported indirect calls make affected relationship coverage partial; no target is invented for them.
+
+`xrefs` runs native `xrefs --to ADDRESS --json` on demand. Kuna unifies forwarding aliases and suppresses edges internal to their class; REA filters to the requested address and records the excluded-row count. Duplicate source sites appear once. An empty result cannot establish the absence of references. Addresses are explicit u64 values or exact, unambiguous procedure names; Kuna has no cursor.
+
+Graph JSON admits at most 32 MiB before parsing; xref JSON at most 16 MiB. A large image can exceed the graph budget: an arm64 Firefox `XUL` (404k functions) does, while a 14 MB Sublime Text binary produced 31.8 MB. Output beyond a buffer is reported as a `memory` resource constraint with `retryable: false`, because the same target and executable produce the same output. The session remembers a graph that overflowed, so later call queries fail at once instead of repeating the whole-program walk. The graph is cached for the selected target and executable profile within one session. Xref reuse is an LRU charged by source bytes plus metadata, limited to 16 MiB and 128 entries; at most eight distinct native xref queries may be pending. Reference commands run one at a time per session and share the provider's foreground budget with single-function decompiles. An unproven process stop retains its admission slots until cleanup succeeds. Concurrent callers share work, individual cancellation releases only that waiter, and the last departing waiter cancels its producer and receives any unproven cleanup. Session close aborts and drains producers, releases parsed caches and retries any retained process/file cleanup.
+
+Evidence includes only selected addresses or dossier neighbors, the producing JSON's SHA-256/byte count, coverage and relevant function facts; unrequested program-wide JSON and instruction text are omitted. Reference caches are in memory and expire on close: a fresh CLI process or MCP server rebuilds them. A first call query therefore costs a whole-program reference walk even when a C export is warm. These operations do not establish full Ghidra/IDA feature parity.
+
+## Provenance and limitations
+
+- Addresses and bytes are Kuna's observations of the image. Procedure entries and extents are its derivations: extents are upper bounds clipped at the next entry, and Kuna can split one function at an entry its own analysis infers. Pseudocode is its reconstruction, never original source.
+- Pseudocode and assembly are exact Kuna text; text from Kuna and from another provider is not comparable as equal or different.
+- Results are re-read rather than replayed from REA's analysis snapshots, because a cold answer can differ from the export's.
+
+## Cache ownership
+
+REA writes Kuna's exports only inside `REA_KUNA_CACHE_DIR`:
+
+- `projects/<key>/` holds published, immutable exports with a manifest of file digests.
+- `staging/` holds builds whose processes may still be live.
+- `trash/` holds retired builds and evicted exports awaiting removal.
+- `home/` is Kuna's HOME.
+
+Builds publish by one rename, so a reader sees a complete export or none. A session keeps the export it uses from eviction only within its own REA process. If another REA process sharing the cache evicts it, the session rebuilds it; meanwhile pseudocode falls back to single-function runs. A damaged or edited export is removed and rebuilt. Staging moves to `trash/` for removal only after REA proves its Kuna process stopped. Staging of a process it cannot prove stopped is never swept; REA keeps it owned, retries the stop on `close_binary`, and reports it until the stop is proven. Failed file removal also stays bound to its export and is retried by `close_binary`; closing does not report success while that cleanup remains. `rea doctor --provider kuna` reports the cache directory without creating it.
+
+## Verify with a real Kuna
+
+```bash
+REA_KUNA_COMMAND=/absolute/path/to/kuna npm run verify:kuna
+# Compare with Ghidra on caller-built binaries
+REA_KUNA_COMMAND=/absolute/path/to/kuna GHIDRA_INSTALL_DIR=/absolute/path/to/ghidra \
+  npm run verify:kuna -- --targets DIR --functions inflate,crc32 --compare-ghidra
+```
+
+The lane compiles a small source-owned fixture with the host C compiler and verifies the MCP workflow end to end. `--targets DIR --functions a,b` adds caller-built binaries. `--compare-ghidra` (with `GHIDRA_INSTALL_DIR`) runs those same functions through Ghidra; it requires `--targets` and `--functions` and refuses empty matrices or requested function names missing from every target. See [testing](testing.md#kuna-lane) for coverage.
+
+Failed runs also report process-lineage evidence and exit unsuccessfully. If provider or transport cleanup is unconfirmed, the report says `cleanup: "incomplete"` and records `retained_workspace`; that directory is preserved for investigation and cleanup retries.

docs/managed-code-analysis.md

@@ -352,7 +352,7 @@ install it or its toolchain. The [NativeAOT guide](ghidra-nativeaot.md) defines
 these routes, their layout/host restrictions, derived-memory provenance, and
 verification coverage.
 
-REA's experimental Windows Ghidra boundary admits native x86 and x86-64 PE
+REA's experimental Windows Ghidra boundary admits native x86, x86-64, ARMNT (Thumb-2) and ARM64 PE
 applications and DLLs for static analysis. For x64 PE NativeAOT, the read-only
 parser also runs there; the [NativeAOT guide](ghidra-nativeaot.md) lists what a
 Windows host covers.

docs/mcp-contracts.md

@@ -393,15 +393,26 @@ concrete version, selection source, and complete analysis profile through
 choices return typed selection details. A selected provider is never replaced
 automatically after a runtime failure.
 
-Every successful target transition allocates `binary_session.analysis_run.run_id`
-before any provider startup. `process_lineage` is `not_observed` until a dynamic
-provider starts, then becomes `snapshots` with every started provider's identity
+Each provider client receives `binary_session.analysis_run.run_id` before any
+provider startup, including clients whose startup fails. `process_lineage` is
+`not_observed` until a dynamic provider starts, then becomes `snapshots` with
+every started provider's identity
 and retained ownership observation. Each observation records `observed_at` and
 is `unavailable` with a reason when ownership could not be revalidated, or
 `verified` with launcher PID, parent PID, process group, and descendants observed
 at that bounded check. These are historical snapshots, not live process
 inventories, and do not claim that no short-lived descendant existed.
 
+`binary_session.cleanup_pending` reports whether the session retains a provider
+client, session file resource, or provider-candidate resource whose cleanup has
+not succeeded. It can be true with `open: false`; target retirement does not
+establish resource release. Health, activity, and lineage remain available
+under a retained client's original `analysis_run.run_id` where the provider
+supplies them. Successful client cleanup clears the retained run; a replacement
+client receives a fresh run ID. Follow the cleanup error's reported resources
+and recovery action before reopening; a missing health observation does not
+establish an idle provider.
+
 ## Progress and cancellation
 
 REA accepts ordinary `tools/call` progress tokens. The numeric `progress` counts
@@ -428,6 +439,10 @@ Native call tracing and process capture retain available observations in
 `details.partial_observation` on failure, including when cleanup succeeds.
 The observation reports its partial coverage; cleanup details describe host
 state separately from the execution failure.
+For these tools, cancellation observed during result finalization or owned
+cleanup also returns typed cancellation with the available observations.
+A normal target exit does not establish request completion. Incomplete cleanup
+still takes precedence and retains its resource diagnostics.
 A process capture that fails during a finalization interval keeps every signal
 attempt and its delivery result in `details.partial_observation`, with
 `elapsed_ms` and `exit_code` null when no exit was observed.
@@ -497,7 +512,10 @@ declared by their tools. Artifact extraction materializes the selected regular
 files into a fresh REA-chosen temporary directory and reports its path.
 
 `analyze_javascript_application` reads the supplied local directory or ASAR path
-and returns its result and Evidence inline.
+and returns its result and Evidence inline. Its owned analysis worker and private
+temporary source files are declared as `launchesProcess` and `writesFilesystem`,
+including for a single JavaScript source. This internal analysis does not execute
+the supplied application code.
 
 ## Ghidra first-query deadlines and recovery
 
@@ -691,6 +709,20 @@ after SDK conversion and check advertised validation and actual calls.
 Individual model APIs can impose additional nesting limits; complete producer
 captures can exceed ten structural levels.
 
+Boolean literal constraints in input schemas are relaxed only in this
+advertisement. For example, a boolean `const` is presented as `type: boolean`;
+an affected `not` is omitted, conditional predicates and their branches are
+omitted together, pure boolean-literal `oneOf` branches collapse to a boolean,
+and other affected `oneOf` branches are presented as `anyOf` to retain branch
+guidance without making overlapping widened branches reject inputs.
+This can make the advertised schema accept inputs the canonical contract
+rejects. The MCP handler still validates every call with the canonical
+contract. The Gemini CLI verifier checks the outgoing
+`parametersJsonSchema` and canonical contract examples against that declaration
+using a loopback model fixture; it does not establish live Google API or
+OpenCode client behavior. Set `REA_VERIFY_GEMINI_SCHEMA_PROFILE=compact` to
+verify the compact advertisement through the same client boundary.
+
 ### Compact input schema profile
 
 Some function-calling providers reject a request while any single tool's

docs/mcp-prompts.md

@@ -111,8 +111,12 @@ list does not establish that the target has no procedures.
 Suggestions are Unicode-normalized, case-insensitive prefix matches. The server
 deduplicates them and sorts by code point for deterministic results. Procedure
 completion derives candidates from the complete inventory, so a name is offered
-only when it identifies one address in that inventory. Completion returns all
-matching values; it does not expose paging controls.
+only when it identifies one address in that inventory. A completion reply
+carries at most 100 values, the MCP limit. `total` counts every match and
+`hasMore` is `true` when values were omitted. Completion has no paging
+controls. A longer prefix reaches an omitted value only when it excludes other
+matches; matching uses normalized, case-insensitive forms, so a value can stay
+unreachable when more than 100 matches share its normalized prefix.
 
 ## Lifecycle and safety
 

docs/native-investigation.md

@@ -270,7 +270,9 @@ Mach-O image. Architecture
 `file_offset` values are container-relative. Entrypoint, segment and section
 `file_offset` values retain their source meaning and are relative to the
 selected Mach-O slice. Raw load-command fields retain their original values.
-Evidence locations map the LC_MAIN entrypoint and nonempty file-backed segment
+A field otool prints more than once, such as each `LC_BUILD_VERSION` tool, is
+keyed `name #n` in print order. Each `LC_LINKER_OPTION` string is keyed
+`string #n` and checked against the declared `count`. Evidence locations map the LC_MAIN entrypoint and nonempty file-backed segment
 ranges to container offsets by adding the observed slice offset. Zero-fill
 segments do not produce file ranges. When the selected slice's container offset
 is unknown, those mapped locations are omitted with an explicit limitation;
@@ -449,6 +451,11 @@ The run ends when the process exits, `max_events` entries are recorded,
 Bounded termination reports partial coverage. REA checks process termination
 using a launch-time OS start identity; if ownership or cleanup cannot be
 verified, the result or error preserves that uncertainty. The result reports:
+Launch errors keep the fact that LLDB created a process separate from whether
+it remains alive. An exited `SBProcess` is reported as exited only after LLDB
+confirms that state; unresolved ownership remains retained for cleanup retry.
+Cancellation observed while output and Evidence are finalized returns partial
+cancellation Evidence, even if the target has already exited.
 
 - the outcome, exit status and captured stdout/stderr (1 MiB kept per stream
   while excess bytes are drained and counted without growing capture files).
@@ -479,6 +486,11 @@ with partial coverage. Available output prefixes remain inline; their byte
 counts are observed lower bounds when final drain counters are unavailable.
 The original failure classification and cleanup uncertainty remain separate.
 An interrupted journal row is ignored with an explicit limitation.
+If the trace completed before cancellation or runtime-directory cleanup failed,
+the retained observation also keeps its exit status, termination proof, launch
+identity, breakpoint results and final stream counters. Those fields remain
+absent when the interrupted producer did not report them; partial coverage alone
+does not mean that every retained fact is incomplete.
 
 Only entries are observed: return values, floating-point and stack arguments,
 and inlined or `objc_direct` calls are not. The target runs with the current
@@ -495,7 +507,7 @@ is present. A denied trace reports the `debugger-attach-denied` reason.
 Install a Ghidra 12.1.x release and the 64-bit full JDK it declares, then configure REA to
 use them. Ghidra analysis supports Linux x64/arm64 and macOS x64/arm64; each installation requires
 the matching native decompiler. Experimental Windows x64 P0 admits native
-x86 and x86-64 PE applications and DLLs, plus explicitly selected
+x86, x86-64, ARMNT (Thumb-2) and ARM64 PE applications and DLLs, plus explicitly selected
 [raw NMOS 6502 images](raw-binary.md#nmos-6502) with a caller-supplied load address,
 on fixed local NTFS volumes. Other raw CPU families remain unsupported on Windows.
 The boundary uses bundled Job Object ownership, protected runtime DACLs, and
@@ -575,6 +587,18 @@ Exact and ambiguous value counts describe emitted root/object value instances;
 a root and an object may contain the same value. Archive digests identify the
 original bytes.
 
+Interface Builder's XML and binary plist archives use the same number metadata
+and exact-decimal representation for emitted node attributes. Projection follows
+object and connection classification, so archive IDs, UID links and graph counts
+keep their meaning. Precision limitations name the source document. Unsafe
+integer-literal counts cover the archive, including fields not emitted as node
+attributes; exact and ambiguous projection counts cover emitted attribute values.
+Integral reals stay numbers, and ambiguous or incompletely observed values retain
+explicit precision limitations. Before plist decoding, the existing aggregate
+decode budget reserves estimated room for number metadata and attribute
+projection; these estimates are not a hard process-memory bound. Native
+NIBArchive decoding is unchanged by this plist-specific representation.
+
 Dates are reported as `{ "$plist_type": "date", "iso8601": "<timestamp>" }`. A
 date the decoder cannot represent, such as an XML date with a year beyond 9999
 or a negative year, is reported as `{ "$plist_type": "date", "iso8601": null }`

docs/process-capture.md

@@ -425,7 +425,9 @@ results for unaffected dimensions and mark affected dimensions unknown; an
 incomplete capture cannot locate the first divergence across all dimensions.
 Trace assertions use coverage for the sources they select, so an assertion
 about complete raw terminal output need not fail because rendered snapshots
-were omitted. Captures missing `truncation_details` are rejected on import;
+were omitted. A trace difference applies to the dimensions identified by its
+diagnostic; unrelated dimensions retain their own comparison, including unknown
+coverage. Captures missing `truncation_details` are rejected on import;
 preserve them as historical files and recapture for current comparisons.
 `limits.file_bytes` is already the cumulative whole-file hashing budget,
 independent of `limits.output_bytes`, which bounds retained terminal content.

docs/readme/README_ar.md

@@ -166,7 +166,7 @@ npx rea-agents@latest setup
 
 <a id="choosing-a-deep-analysis-provider"></a>
 
-تختلف صيغ ملفات الشيفرة الأصلية والأنظمة المضيفة المدعومة باختلاف الموفّر. راجع [إعداد Hopper وGhidra](../installation.md#hopper)، و[دليل IDA](../ida-provider.md)، و[دعم Ghidra التجريبي على Windows](../windows-ghidra-p0.md). يدعم Ghidra أيضًا [تحليل DOS ذي 16 بت](../ghidra-dos.md). لاختيار الموفّر، راجع [دليل CLI](../cli.md#choose-a-provider). تحقّق من [توفر الميزات في الإصدارات](../installation.md#released-package-and-main) للميزات المضافة بعد أحدث إصدار على npm.
+تختلف صيغ ملفات الشيفرة الأصلية والأنظمة المضيفة المدعومة باختلاف الموفّر. راجع [إعداد Hopper وGhidra](../installation.md#hopper)، و[دليل IDA](../ida-provider.md)، [Kuna](../kuna-provider.md)، و[دعم Ghidra التجريبي على Windows](../windows-ghidra-p0.md). يدعم Ghidra أيضًا [تحليل DOS ذي 16 بت](../ghidra-dos.md). لاختيار الموفّر، راجع [دليل CLI](../cli.md#choose-a-provider). تحقّق من [توفر الميزات في الإصدارات](../installation.md#released-package-and-main) للميزات المضافة بعد أحدث إصدار على npm.
 
 ## دراسات الحالة
 
@@ -274,7 +274,7 @@ npx rea-agents@latest setup
 
 ## سجل النجوم
 
-🎉 **200,000 نجمة على GitHub — شكرًا لكم!**
+🎉 **250,000 نجمة على GitHub — شكرًا لكم!**
 
 شكرًا لكل من يستخدم REA، ويبلّغ عن الأخطاء، ويقترح ميزات، ويختبر إصدارات البناء، ويساهم بالإصلاحات.
 

docs/readme/README_de.md

@@ -166,7 +166,7 @@ Statische JavaScript- und .NET-Untersuchungen lesen die bereitgestellten Dateien
 
 <a id="choosing-a-deep-analysis-provider"></a>
 
-Native Formate und unterstützte Hostsysteme unterscheiden sich je nach Provider. Siehe [Hopper- und Ghidra-Einrichtung](../installation.md#hopper), den [IDA-Leitfaden](../ida-provider.md) und die [experimentelle Ghidra-Unterstützung unter Windows](../windows-ghidra-p0.md). Ghidra unterstützt auch [16-Bit-DOS-Analyse](../ghidra-dos.md). Die Provider-Auswahl beschreibt der [CLI-Leitfaden](../cli.md#choose-a-provider). Prüfe die [Verfügbarkeit in Releases](../installation.md#released-package-and-main) für Funktionen, die seit der neuesten npm-Version hinzugekommen sind.
+Native Formate und unterstützte Hostsysteme unterscheiden sich je nach Provider. Siehe [Hopper- und Ghidra-Einrichtung](../installation.md#hopper), den [IDA-Leitfaden](../ida-provider.md), [Kuna](../kuna-provider.md) und die [experimentelle Ghidra-Unterstützung unter Windows](../windows-ghidra-p0.md). Ghidra unterstützt auch [16-Bit-DOS-Analyse](../ghidra-dos.md). Die Provider-Auswahl beschreibt der [CLI-Leitfaden](../cli.md#choose-a-provider). Prüfe die [Verfügbarkeit in Releases](../installation.md#released-package-and-main) für Funktionen, die seit der neuesten npm-Version hinzugekommen sind.
 
 ## Fallstudien
 
@@ -272,7 +272,7 @@ Melde Sicherheitslücken gemäß [SECURITY.md](../../SECURITY.md).
 
 ## Sternverlauf
 
-🎉 **200.000 GitHub-Sterne – vielen Dank!**
+🎉 **250.000 GitHub-Sterne – vielen Dank!**
 
 Danke an alle, die REA nutzen, Fehler melden, Funktionswünsche äußern, Builds testen und Fehlerbehebungen beitragen.
 

docs/readme/README_el.md

@@ -179,7 +179,7 @@ npx rea-agents@latest setup
 
 <a id="choosing-a-deep-analysis-provider"></a>
 
-Οι μορφές κώδικα μηχανής και τα υποστηριζόμενα συστήματα διαφέρουν ανά πάροχο ανάλυσης. Δες τη [ρύθμιση Hopper και Ghidra](../installation.md#hopper), τον [οδηγό IDA](../ida-provider.md) (που περιλαμβάνει προαιρετική γέφυρα MCP για κώδικα μηχανής) και την [πειραματική υποστήριξη Ghidra στα Windows](../windows-ghidra-p0.md). Το Ghidra υποστηρίζει επίσης [ανάλυση DOS 16 bit](../ghidra-dos.md). Για μεγάλα δυαδικά αρχεία, αύξησε το χρονικό όριο εκκίνησης του παρόχου με `REA_GHIDRA_STARTUP_TIMEOUT_MS` ή `REA_HOPPER_STARTUP_TIMEOUT_MS`. Για την επιλογή παρόχου, δες τον [οδηγό CLI](../cli.md#choose-a-provider). Έλεγξε τη [διαθεσιμότητα ανά έκδοση](../installation.md#released-package-and-main) για λειτουργίες που προστέθηκαν μετά την τελευταία έκδοση στο npm.
+Οι μορφές κώδικα μηχανής και τα υποστηριζόμενα συστήματα διαφέρουν ανά πάροχο ανάλυσης. Δες τη [ρύθμιση Hopper και Ghidra](../installation.md#hopper), τον [οδηγό IDA](../ida-provider.md) (που περιλαμβάνει προαιρετική γέφυρα MCP για κώδικα μηχανής), [Kuna](../kuna-provider.md) και την [πειραματική υποστήριξη Ghidra στα Windows](../windows-ghidra-p0.md). Το Ghidra υποστηρίζει επίσης [ανάλυση DOS 16 bit](../ghidra-dos.md). Για μεγάλα δυαδικά αρχεία, αύξησε το χρονικό όριο εκκίνησης του παρόχου με `REA_GHIDRA_STARTUP_TIMEOUT_MS` ή `REA_HOPPER_STARTUP_TIMEOUT_MS`. Για την επιλογή παρόχου, δες τον [οδηγό CLI](../cli.md#choose-a-provider). Έλεγξε τη [διαθεσιμότητα ανά έκδοση](../installation.md#released-package-and-main) για λειτουργίες που προστέθηκαν μετά την τελευταία έκδοση στο npm.
 
 ## Μελέτες περίπτωσης
 
@@ -285,7 +285,7 @@ npx rea-agents@latest setup
 
 ## Ιστορικό αστεριών
 
-🎉 **200.000 αστέρια στο GitHub — σας ευχαριστούμε!**
+🎉 **250.000 αστέρια στο GitHub — σας ευχαριστούμε!**
 
 Ευχαριστούμε όλους όσοι χρησιμοποιούν το REA, αναφέρουν σφάλματα, ζητούν νέες λειτουργίες, δοκιμάζουν builds και συνεισφέρουν διορθώσεις.
 

docs/readme/README_es.md

@@ -166,7 +166,7 @@ La inspección estática de JavaScript y .NET lee los archivos suministrados sin
 
 <a id="choosing-a-deep-analysis-provider"></a>
 
-Los formatos nativos y los sistemas anfitriones compatibles varían según el proveedor. Consulta [configuración de Hopper y Ghidra](../installation.md#hopper), la [guía de IDA](../ida-provider.md) y la [compatibilidad experimental de Ghidra en Windows](../windows-ghidra-p0.md). Ghidra también admite [análisis de DOS de 16 bits](../ghidra-dos.md). Para elegir un proveedor, consulta la [guía de CLI](../cli.md#choose-a-provider). Comprueba la [disponibilidad por versión](../installation.md#released-package-and-main) de las funciones añadidas después de la última versión publicada en npm.
+Los formatos nativos y los sistemas anfitriones compatibles varían según el proveedor. Consulta [configuración de Hopper y Ghidra](../installation.md#hopper), la [guía de IDA](../ida-provider.md), [Kuna](../kuna-provider.md) y la [compatibilidad experimental de Ghidra en Windows](../windows-ghidra-p0.md). Ghidra también admite [análisis de DOS de 16 bits](../ghidra-dos.md). Para elegir un proveedor, consulta la [guía de CLI](../cli.md#choose-a-provider). Comprueba la [disponibilidad por versión](../installation.md#released-package-and-main) de las funciones añadidas después de la última versión publicada en npm.
 
 ## Casos prácticos
 
@@ -272,7 +272,7 @@ Informa de vulnerabilidades siguiendo [SECURITY.md](../../SECURITY.md).
 
 ## Historial de estrellas
 
-🎉 **200.000 estrellas en GitHub: ¡gracias!**
+🎉 **250.000 estrellas en GitHub: ¡gracias!**
 
 Gracias a quienes usan REA, informan de errores, solicitan funciones, prueban compilaciones y contribuyen con correcciones.
 

docs/readme/README_fa.md

@@ -158,7 +158,7 @@ REA به Node.js 22.x (>=22.19)، 24.x (>=24.11) یا 26+، به‌همراه np
 
 <a id="choosing-a-deep-analysis-provider"></a>
 
-فرمت‌های بومی و پشتیبانی سیستم میزبان بر اساس ارائه‌دهنده متفاوت‌اند. [راه‌اندازی Hopper و Ghidra](../installation.md#hopper)، [راهنمای IDA](../ida-provider.md) و [پشتیبانی آزمایشی Ghidra در ویندوز](../windows-ghidra-p0.md) را ببینید. Ghidra همچنین از [تحلیل DOS شانزده‌بیتی](../ghidra-dos.md) پشتیبانی می‌کند. برای انتخاب ارائه‌دهنده، به [راهنمای CLI](../cli.md#choose-a-provider) مراجعه کنید. برای قابلیت‌هایی که پس از آخرین انتشار npm اضافه شده‌اند، [وضعیت انتشار](../installation.md#released-package-and-main) را بررسی کنید.
+فرمت‌های بومی و پشتیبانی سیستم میزبان بر اساس ارائه‌دهنده متفاوت‌اند. [راه‌اندازی Hopper و Ghidra](../installation.md#hopper)، [راهنمای IDA](../ida-provider.md)، [Kuna](../kuna-provider.md) و [پشتیبانی آزمایشی Ghidra در ویندوز](../windows-ghidra-p0.md) را ببینید. Ghidra همچنین از [تحلیل DOS شانزده‌بیتی](../ghidra-dos.md) پشتیبانی می‌کند. برای انتخاب ارائه‌دهنده، به [راهنمای CLI](../cli.md#choose-a-provider) مراجعه کنید. برای قابلیت‌هایی که پس از آخرین انتشار npm اضافه شده‌اند، [وضعیت انتشار](../installation.md#released-package-and-main) را بررسی کنید.
 
 ## نمونه‌های عملی
 
@@ -257,7 +257,7 @@ npx rea-agents@latest setup
 
 ## تاریخچهٔ ستاره‌ها
 
-🎉 **۲۰۰٬۰۰۰ ستارهٔ GitHub — سپاسگزاریم!**
+🎉 **۲۵۰٬۰۰۰ ستارهٔ GitHub — سپاسگزاریم!**
 
 از همهٔ کسانی که از REA استفاده می‌کنند، خطاها را گزارش می‌دهند، قابلیت‌های جدید پیشنهاد می‌کنند، buildها را آزمایش می‌کنند و در بهبود پروژه مشارکت دارند سپاسگزاریم.
 

docs/readme/README_fr.md

@@ -170,7 +170,7 @@ L'inspection statique JavaScript et .NET lit les fichiers fournis sans exécuter
 
 <a id="choosing-a-deep-analysis-provider"></a>
 
-Les formats natifs et les hôtes pris en charge varient selon le fournisseur. Consultez [la configuration de Hopper et Ghidra](../installation.md#hopper), le [guide IDA](../ida-provider.md) et [la prise en charge expérimentale de Ghidra sous Windows](../windows-ghidra-p0.md). Ghidra prend aussi en charge [l'analyse DOS 16 bits](../ghidra-dos.md). Pour les gros binaires, augmentez son délai de démarrage avec `REA_GHIDRA_STARTUP_TIMEOUT_MS`. Pour le choix du fournisseur, consultez le [guide CLI](../cli.md#choose-a-provider). Vérifiez la [disponibilité dans les versions publiées](../installation.md#released-package-and-main) pour les fonctionnalités ajoutées depuis la dernière version npm.
+Les formats natifs et les hôtes pris en charge varient selon le fournisseur. Consultez [la configuration de Hopper et Ghidra](../installation.md#hopper), le [guide IDA](../ida-provider.md), [Kuna](../kuna-provider.md) et [la prise en charge expérimentale de Ghidra sous Windows](../windows-ghidra-p0.md). Ghidra prend aussi en charge [l'analyse DOS 16 bits](../ghidra-dos.md). Pour les gros binaires, augmentez son délai de démarrage avec `REA_GHIDRA_STARTUP_TIMEOUT_MS`. Pour le choix du fournisseur, consultez le [guide CLI](../cli.md#choose-a-provider). Vérifiez la [disponibilité dans les versions publiées](../installation.md#released-package-and-main) pour les fonctionnalités ajoutées depuis la dernière version npm.
 
 ## Études de cas
 
@@ -276,7 +276,7 @@ Signalez les vulnérabilités via [SECURITY.md](../../SECURITY.md).
 
 ## Historique des étoiles
 
-🎉 **200 000 étoiles GitHub — merci !**
+🎉 **250 000 étoiles GitHub — merci !**
 
 Merci à toutes les personnes qui utilisent REA, signalent des bugs, proposent des fonctionnalités, testent les builds et contribuent des correctifs.
 

docs/readme/README_id.md

@@ -173,7 +173,7 @@ Pemeriksaan JavaScript dan .NET statis membaca berkas yang disediakan tanpa menj
 
 <a id="choosing-a-deep-analysis-provider"></a>
 
-Format native dan dukungan sistem host berbeda-beda menurut penyedia. Lihat [penyiapan Hopper dan Ghidra](../installation.md#hopper), [panduan IDA](../ida-provider.md), dan [dukungan eksperimental Ghidra di Windows](../windows-ghidra-p0.md).
+Format native dan dukungan sistem host berbeda-beda menurut penyedia. Lihat [penyiapan Hopper dan Ghidra](../installation.md#hopper), [panduan IDA](../ida-provider.md), [Kuna](../kuna-provider.md), dan [dukungan eksperimental Ghidra di Windows](../windows-ghidra-p0.md).
 
 Ghidra juga mendukung [analisis DOS 16-bit](../ghidra-dos.md).
 
@@ -297,7 +297,7 @@ Laporkan kerentanan melalui [SECURITY.md](../../SECURITY.md).
 
 ## Riwayat Bintang
 
-🎉 **200.000 bintang GitHub, terima kasih!**
+🎉 **250.000 bintang GitHub, terima kasih!**
 
 Terima kasih kepada semua orang yang menggunakan REA, melaporkan bug, meminta fitur, menguji build, dan berkontribusi dalam perbaikan.
 

docs/readme/README_it.md

@@ -170,7 +170,7 @@ L'ispezione statica di JavaScript e .NET legge i file forniti senza eseguire l'a
 
 <a id="choosing-a-deep-analysis-provider"></a>
 
-I formati nativi e il supporto host variano in base al provider. Vedi la [configurazione di Hopper e Ghidra](../installation.md#hopper), la [guida IDA](../ida-provider.md) e il [supporto sperimentale Windows Ghidra](../windows-ghidra-p0.md). Ghidra supporta anche l'[analisi DOS a 16 bit](../ghidra-dos.md). Per binari di grandi dimensioni, aumenta il timeout di avvio con `REA_GHIDRA_STARTUP_TIMEOUT_MS`. Per la selezione del provider, vedi la [guida CLI](../cli.md#choose-a-provider). Controlla la [disponibilità delle release](../installation.md#released-package-and-main) per le funzionalità aggiunte dall'ultima release npm.
+I formati nativi e il supporto host variano in base al provider. Vedi la [configurazione di Hopper e Ghidra](../installation.md#hopper), la [guida IDA](../ida-provider.md), [Kuna](../kuna-provider.md) e il [supporto sperimentale Windows Ghidra](../windows-ghidra-p0.md). Ghidra supporta anche l'[analisi DOS a 16 bit](../ghidra-dos.md). Per binari di grandi dimensioni, aumenta il timeout di avvio con `REA_GHIDRA_STARTUP_TIMEOUT_MS`. Per la selezione del provider, vedi la [guida CLI](../cli.md#choose-a-provider). Controlla la [disponibilità delle release](../installation.md#released-package-and-main) per le funzionalità aggiunte dall'ultima release npm.
 
 ## Showcase
 
@@ -276,7 +276,7 @@ Segnala vulnerabilità tramite [SECURITY.md](../../SECURITY.md).
 
 ## Cronologia delle stelle
 
-🎉 **200.000 stelle su GitHub — grazie!**
+🎉 **250.000 stelle su GitHub — grazie!**
 
 Grazie a tutti coloro che usano REA, segnalano bug, propongono funzionalità, testano le build e contribuiscono con correzioni.
 

docs/readme/README_ja.md

@@ -166,7 +166,7 @@ REA には Node.js 22.x（>=22.19）、24.x（>=24.11）、または 26+ と npm
 
 <a id="choosing-a-deep-analysis-provider"></a>
 
-対応するネイティブ形式とホストは、プロバイダーによって異なります。[Hopper と Ghidra のセットアップ](../installation.md#hopper)、[IDA ガイド](../ida-provider.md)、[Windows 向け Ghidra の実験的対応](../windows-ghidra-p0.md)を参照してください。Ghidra は [16 ビット DOS 分析](../ghidra-dos.md)にも対応しています。プロバイダーの選択は [CLI ガイド](../cli.md#choose-a-provider)を参照してください。最新の npm リリース以降に追加された機能については、[リリースでの利用可否](../installation.md#released-package-and-main)を確認してください。
+対応するネイティブ形式とホストは、プロバイダーによって異なります。[Hopper と Ghidra のセットアップ](../installation.md#hopper)、[IDA ガイド](../ida-provider.md)、[Kuna](../kuna-provider.md)、[Windows 向け Ghidra の実験的対応](../windows-ghidra-p0.md)を参照してください。Ghidra は [16 ビット DOS 分析](../ghidra-dos.md)にも対応しています。プロバイダーの選択は [CLI ガイド](../cli.md#choose-a-provider)を参照してください。最新の npm リリース以降に追加された機能については、[リリースでの利用可否](../installation.md#released-package-and-main)を確認してください。
 
 ## 事例
 
@@ -274,7 +274,7 @@ npx rea-agents@latest setup
 
 ## スター履歴
 
-🎉 **GitHub スター 200,000 件、ありがとうございます！**
+🎉 **GitHub スター 250,000 件、ありがとうございます！**
 
 REA の利用、バグ報告、機能の要望、ビルドのテスト、修正への貢献に感謝します。
 

docs/readme/README_ko.md

@@ -175,7 +175,7 @@ REA에는 Node.js 22.x(>=22.19), 24.x(>=24.11) 또는 26+와 npm이 필요합니
 
 <a id="choosing-a-deep-analysis-provider"></a>
 
-지원하는 네이티브 형식과 호스트는 제공자마다 다릅니다. [Hopper 및 Ghidra 설정](../installation.md#hopper), [IDA 가이드](../ida-provider.md)(선택 사항인 네이티브 MCP 브리지 포함), [실험적 Windows Ghidra 지원](../windows-ghidra-p0.md)을 참고하세요. Ghidra는 [16비트 DOS 분석](../ghidra-dos.md)도 지원합니다. 큰 바이너리를 분석할 때는 `REA_GHIDRA_STARTUP_TIMEOUT_MS` 또는 `REA_HOPPER_STARTUP_TIMEOUT_MS`로 제공자의 시작 제한 시간을 늘리세요. 제공자 선택은 [CLI 가이드](../cli.md#choose-a-provider)를 참고하세요. 최신 npm 릴리스 이후 추가된 기능은 [릴리스에서의 사용 가능 여부](../installation.md#released-package-and-main)를 확인하세요.
+지원하는 네이티브 형식과 호스트는 제공자마다 다릅니다. [Hopper 및 Ghidra 설정](../installation.md#hopper), [IDA 가이드](../ida-provider.md)(선택 사항인 네이티브 MCP 브리지 포함), [Kuna](../kuna-provider.md), [실험적 Windows Ghidra 지원](../windows-ghidra-p0.md)을 참고하세요. Ghidra는 [16비트 DOS 분석](../ghidra-dos.md)도 지원합니다. 큰 바이너리를 분석할 때는 `REA_GHIDRA_STARTUP_TIMEOUT_MS` 또는 `REA_HOPPER_STARTUP_TIMEOUT_MS`로 제공자의 시작 제한 시간을 늘리세요. 제공자 선택은 [CLI 가이드](../cli.md#choose-a-provider)를 참고하세요. 최신 npm 릴리스 이후 추가된 기능은 [릴리스에서의 사용 가능 여부](../installation.md#released-package-and-main)를 확인하세요.
 
 ## 사례
 
@@ -283,7 +283,7 @@ npx rea-agents@latest setup
 
 ## 스타 기록
 
-🎉 **GitHub 스타 200,000개, 감사합니다!**
+🎉 **GitHub 스타 250,000개, 감사합니다!**
 
 REA를 사용하고, 버그를 보고하고, 기능을 요청하고, 빌드를 테스트하고, 수정에 기여해 주신 모든 분께 감사드립니다.
 

docs/readme/README_pl.md

@@ -166,7 +166,7 @@ Statyczna analiza JavaScript i .NET odczytuje dostarczone pliki bez uruchamiania
 
 <a id="choosing-a-deep-analysis-provider"></a>
 
-Obsługiwane formaty natywne i platformy zależą od dostawcy analizy. Zobacz [konfigurację Hoppera i Ghidry](../installation.md#hopper), [przewodnik IDA](../ida-provider.md) oraz [eksperymentalną obsługę Ghidry w systemie Windows](../windows-ghidra-p0.md). Ghidra obsługuje też [analizę 16-bitowego DOS-a](../ghidra-dos.md). Wybór dostawcy opisuje [przewodnik CLI](../cli.md#choose-a-provider). Sprawdź [dostępność w wydaniach](../installation.md#released-package-and-main) dla funkcji dodanych po najnowszym wydaniu npm.
+Obsługiwane formaty natywne i platformy zależą od dostawcy analizy. Zobacz [konfigurację Hoppera i Ghidry](../installation.md#hopper), [przewodnik IDA](../ida-provider.md), [Kuna](../kuna-provider.md) oraz [eksperymentalną obsługę Ghidry w systemie Windows](../windows-ghidra-p0.md). Ghidra obsługuje też [analizę 16-bitowego DOS-a](../ghidra-dos.md). Wybór dostawcy opisuje [przewodnik CLI](../cli.md#choose-a-provider). Sprawdź [dostępność w wydaniach](../installation.md#released-package-and-main) dla funkcji dodanych po najnowszym wydaniu npm.
 
 ## Przykłady zastosowań
 
@@ -272,7 +272,7 @@ Zgłaszaj podatności zgodnie z [SECURITY.md](../../SECURITY.md).
 
 ## Historia gwiazdek
 
-🎉 **200 000 gwiazdek na GitHubie — dziękujemy!**
+🎉 **250 000 gwiazdek na GitHubie — dziękujemy!**
 
 Dziękujemy wszystkim, którzy używają REA, zgłaszają błędy, proponują nowe funkcje, testują kompilacje i przesyłają poprawki.
 

docs/readme/README_pt-BR.md

@@ -166,7 +166,7 @@ A inspeção estática de JavaScript e .NET lê os arquivos fornecidos sem execu
 
 <a id="choosing-a-deep-analysis-provider"></a>
 
-Os formatos nativos e os sistemas anfitriões compatíveis variam por provedor. Consulte [configuração do Hopper e Ghidra](../installation.md#hopper), o [guia do IDA](../ida-provider.md) e o [suporte experimental ao Ghidra no Windows](../windows-ghidra-p0.md). O Ghidra também oferece [análise de DOS de 16 bits](../ghidra-dos.md). Para selecionar um provedor, consulte o [guia da CLI](../cli.md#choose-a-provider). Verifique a [disponibilidade nas versões](../installation.md#released-package-and-main) das funcionalidades adicionadas após a versão mais recente no npm.
+Os formatos nativos e os sistemas anfitriões compatíveis variam por provedor. Consulte [configuração do Hopper e Ghidra](../installation.md#hopper), o [guia do IDA](../ida-provider.md), [Kuna](../kuna-provider.md) e o [suporte experimental ao Ghidra no Windows](../windows-ghidra-p0.md). O Ghidra também oferece [análise de DOS de 16 bits](../ghidra-dos.md). Para selecionar um provedor, consulte o [guia da CLI](../cli.md#choose-a-provider). Verifique a [disponibilidade nas versões](../installation.md#released-package-and-main) das funcionalidades adicionadas após a versão mais recente no npm.
 
 ## Casos de uso
 
@@ -272,7 +272,7 @@ Relate vulnerabilidades conforme [SECURITY.md](../../SECURITY.md).
 
 ## Histórico de estrelas
 
-🎉 **200.000 estrelas no GitHub — muito obrigado!**
+🎉 **250.000 estrelas no GitHub — muito obrigado!**
 
 Obrigado a todos que usam o REA, relatam bugs, sugerem funcionalidades, testam builds e contribuem com correções.
 

docs/readme/README_ru.md

@@ -166,7 +166,7 @@ REA требует Node.js 22.x (>=22.19), 24.x (>=24.11) или 26+, а так
 
 <a id="choosing-a-deep-analysis-provider"></a>
 
-Нативные форматы и поддерживаемые системы различаются у разных провайдеров. См. [настройку Hopper и Ghidra](../installation.md#hopper), [руководство по IDA](../ida-provider.md) и [экспериментальную поддержку Ghidra в Windows](../windows-ghidra-p0.md). Ghidra также поддерживает [анализ 16-битного DOS](../ghidra-dos.md). Выбор провайдера описан в [руководстве по CLI](../cli.md#choose-a-provider). Для функций, добавленных после последней версии в npm, проверьте [доступность в релизах](../installation.md#released-package-and-main).
+Нативные форматы и поддерживаемые системы различаются у разных провайдеров. См. [настройку Hopper и Ghidra](../installation.md#hopper), [руководство по IDA](../ida-provider.md), [Kuna](../kuna-provider.md) и [экспериментальную поддержку Ghidra в Windows](../windows-ghidra-p0.md). Ghidra также поддерживает [анализ 16-битного DOS](../ghidra-dos.md). Выбор провайдера описан в [руководстве по CLI](../cli.md#choose-a-provider). Для функций, добавленных после последней версии в npm, проверьте [доступность в релизах](../installation.md#released-package-and-main).
 
 ## Примеры
 
@@ -272,7 +272,7 @@ npx rea-agents@latest setup
 
 ## История звёзд
 
-🎉 **200 000 звёзд на GitHub — спасибо!**
+🎉 **250 000 звёзд на GitHub — спасибо!**
 
 Спасибо всем, кто использует REA, сообщает об ошибках, предлагает новые функции, тестирует сборки и помогает с исправлениями.
 

docs/readme/README_th.md

@@ -166,7 +166,7 @@ REA ต้องใช้ Node.js 22.x (>=22.19), 24.x (>=24.11) หรือ 26
 
 <a id="choosing-a-deep-analysis-provider"></a>
 
-รูปแบบไบนารีเนทีฟและแพลตฟอร์มที่รองรับแตกต่างกันตามผู้ให้บริการวิเคราะห์ ดู[การตั้งค่า Hopper และ Ghidra](../installation.md#hopper), [คู่มือ IDA](../ida-provider.md) และ[การรองรับ Ghidra บน Windows แบบทดลอง](../windows-ghidra-p0.md) Ghidra ยังรองรับ[การวิเคราะห์ DOS แบบ 16 บิต](../ghidra-dos.md) สำหรับการเลือกผู้ให้บริการ ดู[คู่มือ CLI](../cli.md#choose-a-provider) ตรวจสอบ[ความพร้อมใช้งานในแต่ละรุ่น](../installation.md#released-package-and-main) สำหรับฟีเจอร์ที่เพิ่มหลังรุ่นล่าสุดบน npm
+รูปแบบไบนารีเนทีฟและแพลตฟอร์มที่รองรับแตกต่างกันตามผู้ให้บริการวิเคราะห์ ดู[การตั้งค่า Hopper และ Ghidra](../installation.md#hopper), [คู่มือ IDA](../ida-provider.md) [Kuna](../kuna-provider.md) และ[การรองรับ Ghidra บน Windows แบบทดลอง](../windows-ghidra-p0.md) Ghidra ยังรองรับ[การวิเคราะห์ DOS แบบ 16 บิต](../ghidra-dos.md) สำหรับการเลือกผู้ให้บริการ ดู[คู่มือ CLI](../cli.md#choose-a-provider) ตรวจสอบ[ความพร้อมใช้งานในแต่ละรุ่น](../installation.md#released-package-and-main) สำหรับฟีเจอร์ที่เพิ่มหลังรุ่นล่าสุดบน npm
 
 ## ตัวอย่างการใช้งาน
 
@@ -272,7 +272,7 @@ npx rea-agents@latest setup
 
 ## ประวัติดาว
 
-🎉 **200,000 ดาวบน GitHub — ขอบคุณทุกคน!**
+🎉 **250,000 ดาวบน GitHub — ขอบคุณทุกคน!**
 
 ขอบคุณทุกคนที่ใช้ REA รายงานบั๊ก เสนอฟีเจอร์ ทดสอบบิลด์ และช่วยแก้ไข
 

docs/readme/README_tr.md

@@ -171,7 +171,7 @@ Statik JavaScript ve .NET incelemesi, uygulamayı çalıştırmadan sağlanan do
 
 <a id="choosing-a-deep-analysis-provider"></a>
 
-Yerel makine kodu içeren ikili dosya biçimleri ve ana bilgisayar desteği sağlayıcıya göre değişir. [Hopper ve Ghidra kurulumu](../installation.md#hopper), isteğe bağlı native MCP bridge'i de açıklayan [IDA rehberi](../ida-provider.md) ve [Windows için deneysel Ghidra desteği](../windows-ghidra-p0.md) belgelerine bakın. Ghidra ayrıca [16 bit DOS analizini](../ghidra-dos.md) destekler. Büyük ikili dosyalar için sağlayıcının başlatma zaman aşımını `REA_GHIDRA_STARTUP_TIMEOUT_MS` veya `REA_HOPPER_STARTUP_TIMEOUT_MS` ile artırın. Sağlayıcı seçimi için [CLI rehberine](../cli.md#choose-a-provider) bakın. En son npm sürümünden sonra eklenen özellikler için [sürümlerde kullanılabilirliği](../installation.md#released-package-and-main) kontrol edin.
+Yerel makine kodu içeren ikili dosya biçimleri ve ana bilgisayar desteği sağlayıcıya göre değişir. [Hopper ve Ghidra kurulumu](../installation.md#hopper), isteğe bağlı native MCP bridge'i de açıklayan [IDA rehberi](../ida-provider.md), [Kuna](../kuna-provider.md) ve [Windows için deneysel Ghidra desteği](../windows-ghidra-p0.md) belgelerine bakın. Ghidra ayrıca [16 bit DOS analizini](../ghidra-dos.md) destekler. Büyük ikili dosyalar için sağlayıcının başlatma zaman aşımını `REA_GHIDRA_STARTUP_TIMEOUT_MS` veya `REA_HOPPER_STARTUP_TIMEOUT_MS` ile artırın. Sağlayıcı seçimi için [CLI rehberine](../cli.md#choose-a-provider) bakın. En son npm sürümünden sonra eklenen özellikler için [sürümlerde kullanılabilirliği](../installation.md#released-package-and-main) kontrol edin.
 
 ## Örnek çalışmalar
 
@@ -277,7 +277,7 @@ Güvenlik açıklarını [SECURITY.md](../../SECURITY.md) üzerinden bildirin.
 
 ## Yıldız geçmişi
 
-🎉 **GitHub'da 200.000 yıldız — teşekkürler!**
+🎉 **GitHub'da 250.000 yıldız — teşekkürler!**
 
 REA'yı kullanan, hata bildiren, özellik talep eden, derlemeleri test eden ve düzeltmelere katkıda bulunan herkese teşekkürler.
 

docs/readme/README_uk.md

@@ -166,7 +166,7 @@ REA потребує Node.js 22.x (>=22.19), 24.x (>=24.11) або 26+, а та
 
 <a id="choosing-a-deep-analysis-provider"></a>
 
-Нативні формати й підтримка систем відрізняються залежно від провайдера. Дивіться [налаштування Hopper і Ghidra](../installation.md#hopper), [посібник IDA](../ida-provider.md) та [експериментальну підтримку Ghidra у Windows](../windows-ghidra-p0.md). Ghidra також підтримує [аналіз 16-бітного DOS](../ghidra-dos.md). Для вибору провайдера дивіться [посібник CLI](../cli.md#choose-a-provider). Для функцій, доданих після найновішої версії в npm, перевірте [доступність у випусках](../installation.md#released-package-and-main).
+Нативні формати й підтримка систем відрізняються залежно від провайдера. Дивіться [налаштування Hopper і Ghidra](../installation.md#hopper), [посібник IDA](../ida-provider.md), [Kuna](../kuna-provider.md) та [експериментальну підтримку Ghidra у Windows](../windows-ghidra-p0.md). Ghidra також підтримує [аналіз 16-бітного DOS](../ghidra-dos.md). Для вибору провайдера дивіться [посібник CLI](../cli.md#choose-a-provider). Для функцій, доданих після найновішої версії в npm, перевірте [доступність у випусках](../installation.md#released-package-and-main).
 
 ## Приклади
 
@@ -272,7 +272,7 @@ npx rea-agents@latest setup
 
 ## Історія зірок
 
-🎉 **200 000 зірок на GitHub — дякуємо!**
+🎉 **250 000 зірок на GitHub — дякуємо!**
 
 Дякуємо всім, хто використовує REA, повідомляє про помилки, пропонує нові функції, тестує збірки й допомагає з виправленнями.
 

docs/readme/README_vi.md

@@ -166,7 +166,7 @@ Kiểm tra tĩnh JavaScript và .NET đọc các tệp được cung cấp mà k
 
 <a id="choosing-a-deep-analysis-provider"></a>
 
-Định dạng mã máy và nền tảng chạy REA được hỗ trợ khác nhau tùy trình phân tích. Xem [thiết lập Hopper và Ghidra](../installation.md#hopper), [hướng dẫn IDA](../ida-provider.md) và [hỗ trợ Ghidra thử nghiệm trên Windows](../windows-ghidra-p0.md). Ghidra cũng hỗ trợ [phân tích DOS 16 bit](../ghidra-dos.md). Để chọn trình phân tích, xem [hướng dẫn CLI](../cli.md#choose-a-provider). Kiểm tra [khả năng sử dụng theo bản phát hành](../installation.md#released-package-and-main) đối với tính năng được bổ sung sau phiên bản npm mới nhất.
+Định dạng mã máy và nền tảng chạy REA được hỗ trợ khác nhau tùy trình phân tích. Xem [thiết lập Hopper và Ghidra](../installation.md#hopper), [hướng dẫn IDA](../ida-provider.md), [Kuna](../kuna-provider.md) và [hỗ trợ Ghidra thử nghiệm trên Windows](../windows-ghidra-p0.md). Ghidra cũng hỗ trợ [phân tích DOS 16 bit](../ghidra-dos.md). Để chọn trình phân tích, xem [hướng dẫn CLI](../cli.md#choose-a-provider). Kiểm tra [khả năng sử dụng theo bản phát hành](../installation.md#released-package-and-main) đối với tính năng được bổ sung sau phiên bản npm mới nhất.
 
 ## Các nghiên cứu điển hình
 
@@ -272,7 +272,7 @@ Báo cáo lỗ hổng theo [SECURITY.md](../../SECURITY.md).
 
 ## Lịch sử sao
 
-🎉 **200.000 sao trên GitHub — xin cảm ơn!**
+🎉 **250.000 sao trên GitHub — xin cảm ơn!**
 
 Cảm ơn mọi người đã sử dụng REA, báo lỗi, đề xuất tính năng, kiểm thử bản dựng và đóng góp bản sửa lỗi.
 

docs/readme/README_zh-TW.md

@@ -170,7 +170,7 @@ REA 需要 Node.js 22.x（>=22.19）、24.x（>=24.11）或 26+，以及 npm。
 
 <a id="choosing-a-deep-analysis-provider"></a>
 
-不同提供者支援的原生格式和主機系統有所不同。請參閱 [Hopper 與 Ghidra 設定](../installation.md#hopper)、[IDA 指南](../ida-provider.md)和[實驗性 Windows Ghidra 支援](../windows-ghidra-p0.md)。Ghidra 也支援 [16 位 DOS 分析](../ghidra-dos.md)。提供者選擇見 [CLI 指南](../cli.md#choose-a-provider)。對於最新 npm 版本發佈後新增的功能，請查看[發佈可用性](../installation.md#released-package-and-main)。
+不同提供者支援的原生格式和主機系統有所不同。請參閱 [Hopper 與 Ghidra 設定](../installation.md#hopper)、[IDA 指南](../ida-provider.md)、[Kuna](../kuna-provider.md)和[實驗性 Windows Ghidra 支援](../windows-ghidra-p0.md)。Ghidra 也支援 [16 位 DOS 分析](../ghidra-dos.md)。提供者選擇見 [CLI 指南](../cli.md#choose-a-provider)。對於最新 npm 版本發佈後新增的功能，請查看[發佈可用性](../installation.md#released-package-and-main)。
 
 ## 案例
 
@@ -278,7 +278,7 @@ npx rea-agents@latest setup
 
 ## Star 歷史
 
-🎉 **GitHub Star 達到 200,000 個，感謝大家！**
+🎉 **GitHub Star 達到 250,000 個，感謝大家！**
 
 感謝每一位使用 REA、報告錯誤、提出功能需求、測試建置和貢獻修復的朋友。
 

docs/readme/README_zh.md

@@ -170,7 +170,7 @@ REA 需要 Node.js 22.x（>=22.19）、24.x（>=24.11）或 26+，以及 npm。
 
 <a id="choosing-a-deep-analysis-provider"></a>
 
-不同提供方支持的原生格式和主机系统有所不同。请参阅 [Hopper 与 Ghidra 设置](../installation.md#hopper)、[IDA 指南](../ida-provider.md)和[实验性 Windows Ghidra 支持](../windows-ghidra-p0.md)。Ghidra 也支持 [16 位 DOS 分析](../ghidra-dos.md)。提供方选择见 [CLI 指南](../cli.md#choose-a-provider)。对于最新 npm 版本发布后新增的功能，请查看[发布可用性](../installation.md#released-package-and-main)。
+不同提供方支持的原生格式和主机系统有所不同。请参阅 [Hopper 与 Ghidra 设置](../installation.md#hopper)、[IDA 指南](../ida-provider.md)、[Kuna](../kuna-provider.md)和[实验性 Windows Ghidra 支持](../windows-ghidra-p0.md)。Ghidra 也支持 [16 位 DOS 分析](../ghidra-dos.md)。提供方选择见 [CLI 指南](../cli.md#choose-a-provider)。对于最新 npm 版本发布后新增的功能，请查看[发布可用性](../installation.md#released-package-and-main)。
 
 ## 案例
 
@@ -278,7 +278,7 @@ npx rea-agents@latest setup
 
 ## Star 历史
 
-🎉 **GitHub Star 达到 200,000 个，感谢大家！**
+🎉 **GitHub Star 达到 250,000 个，感谢大家！**
 
 感谢每一位使用 REA、报告错误、提出功能需求、测试构建和贡献修复的朋友。
 

docs/reconstruction-coverage.md

@@ -0,0 +1,43 @@
+# Reconstruction coverage
+
+`rea evaluate-reconstruction-coverage coverage.json --json` evaluates an inline
+coverage workspace and its selected `boundary_id`. MCP exposes the same input
+through `evaluate_reconstruction_coverage`. Results contain a status, reasons,
+recommended probes and Evidence references; the supplied bundle retains the
+complete records.
+
+Evidence IDs and bundle manifests must authenticate before evaluation. A valid
+record can still be insufficient proof: a passing verifier summary cannot
+promote inferred Evidence into a replay observation. That case returns
+`unknown`, while a missing or tampered record remains an input error.
+
+## Comparable verifier Evidence
+
+At least one record explicitly cited by the latest verifier observation must
+support the declared contract authority:
+
+| Contract authority  | Evidence source                           | Confidence              |
+| ------------------- | ----------------------------------------- | ----------------------- |
+| `shipped-artifact`  | `shipped-artifact`                        | `observed` or `derived` |
+| `controlled-replay` | `controlled-replay`                       | `observed`              |
+| `live-observation`  | `controlled-replay` or `external-service` | `observed`              |
+| `external`          | `external-service`                        | `observed`              |
+
+Contextual references can accompany qualifying proof. An unrelated observed
+record elsewhere in the bundle cannot repair a result that cites only
+inference. Older observed results cannot replace the newest weak result.
+Equivalent observations at the same instant can share qualifying proof from
+one of those observations; conflicting tied assessments remain unknown.
+
+## Package and integration proof
+
+Build, clean-install and runtime-routing claims require observed runtime
+Evidence (`controlled-replay` or `external-service`). Package-content and
+authority-independence audits can also use observed or derived
+`shipped-artifact` Evidence. Each proof must cite its qualifying records and
+cover the selected boundary's artifact digests. An inference alone establishes
+neither a passing nor a failed package workflow.
+
+The evaluator preserves Evidence provenance and returns an actionable unknown
+when qualification is unavailable. It does not change the cited records to
+make their authority match a caller's declaration.

docs/reconstruction-obligation-ledgers.md

@@ -323,3 +323,6 @@ disposition but does not silently count it as verified.
 The tool returns every obligation inline. Its `ledger_id` and `closure_digest`
 identify the complete result, including summaries, reports, ownership and
 dependency graphs, Evidence links, and limitations.
+
+For named boundaries with explicit verifier contracts and package-proof kinds,
+see [reconstruction coverage](reconstruction-coverage.md).

docs/reverse-engineering-tools.md

@@ -16,11 +16,12 @@ can run.
 
 `inspect_with_objdump` and `rea inspect-with-objdump` capture stdout and stderr as inline Evidence. Artifact inspection Evidence records SHA-256 digests before and after the provider call and whether those observed digests differ; this cannot prove the exact bytes consumed if the artifact changes during execution. Choose one artifact operation: `file_headers`, `section_headers`, `symbols`, `relocations`, `dwarf`, `disassemble`, or `disassemble_all`. `disassemble` maps to `-d`; `disassemble_all` maps to `-D` and has different symbol-boundary semantics. The pathless `inspect_with_objdump {"operation":"architectures"}` MCP call or `rea list-objdump-architectures` CLI command maps to `-i` and produces subjectless provider-capability Evidence describing BFD targets compiled into the installed binary, not a guarantee that a target is supported.
 
-For DWARF, REA does not follow local debug links by default. Set `follow_debug_links` to `true` to pass `-WK`; external debug-file identities are not bound to Evidence when following links is enabled. The CLI option is `--follow-debug-links true`. REA clears `DEBUGINFOD_URLS` in the objdump child environment to disable debuginfod network access on GNU binutils versions that do not support `-WE`. The complete raw output is authoritative; REA does not normalize it into a version-independent disassembly schema.
+For DWARF, REA does not follow local debug links by default. Set `follow_debug_links` to `true` to pass `-WK`; external debug-file identities are not bound to Evidence when following links is enabled. The CLI flag is `--follow-debug-links`; `--no-follow-debug-links` or `--follow-debug-links=false` keeps the default. A separate `true` or `false` word after the flag is not its value. REA clears `DEBUGINFOD_URLS` in the objdump child environment to disable debuginfod network access on GNU binutils versions that do not support `-WE`. The complete raw output is authoritative; REA does not normalize it into a version-independent disassembly schema.
 
 ```sh
 rea inspect-with-objdump ./program disassemble
-rea inspect-with-objdump ./program dwarf --follow-debug-links false
+rea inspect-with-objdump ./program dwarf
+rea inspect-with-objdump ./program dwarf --follow-debug-links
 ```
 
 Captured stdout and stderr are retained up to a 16 MiB combined limit; `output_truncated` marks output beyond that limit.

docs/sqlite-databases.md

@@ -105,6 +105,20 @@ overall native heap and total resident memory are not capped. Row limits do not
 override these byte and resource limits. Resource-limit failures do not return
 an incomplete schema as successful complete evidence.
 
+If the combined database and WAL exceed the snapshot budget, the refusal is a
+`resource_constraint` and reports both the configured byte limit and the exact
+observed combined size as a decimal string. The remediation directs callers to
+create a smaller consistent offline snapshot containing only relevant data
+through the application's supported export workflow. A generic backup or copy
+does not guarantee a smaller result; REA does not truncate or modify the
+selected source.
+
+On hosts that cannot address the derived WAL or rollback-journal path for a
+long selected database name, REA reports that the SQLite capability is
+unavailable. It does not assume an inaccessible sidecar is absent. Use the
+application's supported consistent offline export or copy workflow to place
+the complete known file set at a shorter path.
+
 MCP also applies its configured response budget to the complete Evidence
 envelope. If delivery exceeds that budget, it reports a transport constraint
 with guidance for exporting retained Evidence, requesting a matching larger

docs/testing.md

@@ -199,6 +199,44 @@ Measure slow cases before pruning capacity coverage. Retain inputs beyond the
 formerly failing size/depth and distinguish cold runs, warm builds, coverage,
 and toolchain conditions when comparing timings.
 
+## Lifecycle regressions
+
+For changes to resource acquisition, readiness, cancellation, or cleanup, select
+the affected owners and applicable scenarios below. Include temporary snapshots
+and one-shot commands. Use public CLI/MCP workflows where practical and a
+controlled production seam for interleavings that cannot be reproduced reliably.
+
+| Scenario                                                                                | Assertion and control                                                                                                                                                                                                          |
+| --------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ |
+| Target admitted; startup still pending                                                  | Status preserves the readiness distinction and observed startup activity; readiness requires the provider's documented proof.                                                                                                  |
+| First handshake is late, absent, invalid, or recovered during close                     | Initial and recovery paths retain the same authenticated identity facts. Foreign run/token controls cannot authorize cleanup; authenticated idle/native work survives the unauthenticated listener deadline.                   |
+| Launch returns an error with a valid process handle                                     | Cover already-exited and still-live owned processes. Verify observed exit separately from creation; failed termination or unavailable identity remains unresolved.                                                             |
+| Cancellation arrives after workload exit, during result reads or temporary-root removal | Return typed cancellation with available partial observations and actual cleanup status. The non-aborted control succeeds; existing primary/cleanup failures keep their precedence.                                            |
+| Children have mixed health or cleanup outcomes                                          | Missing peer health does not erase an observed failure or activity. Idle plus unknown stays unknown. Retry a failed child without closing a successful peer again, including after a close-option change.                      |
+| Cleanup fails after success or primary failure                                          | Public output retains the completed observation or primary diagnostic, remaining resources, and owner recovery action. Restore the denied resource and verify the owning cleanup retry.                                        |
+| Target exits without an acknowledgement, or transport disappears while it stays alive   | Positive owned exit proof resolves target lifetime; other resources still need cleanup. Lost transport alone does not prove exit. Verify unrelated/attached targets survive and same-connection recovery works when supported. |
+
+Synchronize cancellation with the finalization boundary being tested; a fixed
+delay after dispatch does not establish that phase. Use markers, barriers, or
+owned cleanup seams while preserving real cleanup where feasible. Check server
+settlement and retained resources even if the MCP SDK rejects the client promise
+before the handler finishes. A cancellation during active execution does not
+cover cancellation during final cleanup.
+
+For atomic writes, exercise cancellation on both sides of the documented commit
+point and preserve any publication receipt. A late interruption does not prove
+that no output was published.
+
+Representative regressions include
+[process cleanup-time SIGINT](https://github.com/morluto/rea/blob/e78455338fe2d06c4cd03d59cbf33473d754998b/tests/boundary/process/processCaptureCancellationAdapters.test.ts),
+[late authenticated cleanup](https://github.com/morluto/rea/blob/e78455338fe2d06c4cd03d59cbf33473d754998b/tests/boundary/providers/hopper/hopperClient.handshakeRecovery.test.ts),
+[LLDB launch and finalization](https://github.com/morluto/rea/blob/e78455338fe2d06c4cd03d59cbf33473d754998b/tests/boundary/providers/native/lldbCallTracerFailures.test.ts),
+[composed child ownership](https://github.com/morluto/rea/blob/e78455338fe2d06c4cd03d59cbf33473d754998b/src/application/binary/CompositeProvider.test.ts),
+and [public CLI cleanup denial](https://github.com/morluto/rea/blob/e78455338fe2d06c4cd03d59cbf33473d754998b/tests/boundary/cli/wasmCleanupCli.test.ts).
+These controlled probes do not replace the matching real-provider lane. Record
+the base failure, passing control, exact host/provider, verified cleanup, and
+unavailable coverage; a missing engine is not a pass.
+
 ## Behavioral depths
 
 | Depth        | Location                | What it proves                                                                                        |
@@ -817,6 +855,38 @@ Native IDA MCP verification uses `"protocol": "native"` with an already-running
 
 The initial Linux x64 IDA Free 9.4 run used native MCP 1.1.2 with a local navigation correction. Windows x64 verification uses **IDA Pro 9.3**, the two-function public fixture, and the shipped Windows `whoami.exe` application. The native plugin was built from upstream `c9d15d98e8a143fc74f6e14037c48f63889b6d6c`, including the merged navigation fix, with a local loopback-listener change. Production CLI/MCP parity, switching, identity and cleanup passed. Correctness and performance measurements are recorded in the [Windows native verification notes](https://github.com/morluto/rea/blob/main/experiments/analysis-efficiency/ida-native-windows.md). Stock 1.1.2 switching, native macOS, and Windows IDA Free remain unverified. See the [native bridge prerequisites](ida-provider.md#optional-native-ida-mcp-bridge).
 
+## Kuna lane
+
+`npm run verify:kuna` requires a caller-supplied `kuna` in `REA_KUNA_COMMAND`
+and a host C compiler (`REA_CC`, default `cc`) with `strip` and `nm`. It builds a
+source-owned fixture with non-inlined calls, with symbols and stripped, then verifies through MCP that
+callers, callees and exact incoming xrefs agree with those calls, graph/path
+composition reaches the callee, reference-only queries leave project/staging
+entries unchanged, repeated relationships reuse the graph, and
+the first cold pseudocode answer comes from a single-function run or an export
+that already finished, assembly waits for and serves from that export, and
+pseudocode and both dossier facets match their recorded file spans and digests,
+a fresh server reuses the export without rebuilding it, staging is empty after
+a completed export, and no owned process survives. Pass
+`--targets DIR --functions a,b` to add caller-built binaries laid out as
+`DIR/<flavor>/<name>` beside `<name>.stripped`; addresses come from the
+symbol-bearing build. `--compare-ghidra` with `GHIDRA_INSTALL_DIR` runs those
+caller-built functions through Ghidra (it refuses to run without `--targets` and
+`--functions`, an empty matrix, or names missing from every supplied target) and
+reports latency, sampled process-tree CPU and peak memory, UTF-8 response size,
+and goto and warning-marker counts. Those counts are not a correctness judgement,
+and model tokens are not measured. Reference timings and response sizes are
+reported separately; reference caches expire when the server closes. `--report PATH` writes the JSON report.
+
+The first matrix run used Kuna 1.833 and Ghidra 12.1.4 with JDK 21 on macOS
+arm64. It covered zlib 1.3.2 `minigzip`, coreutils 9.9 `sort` and the sqlite
+3.54.0 shell, built at `-O2` for ELF x86-64, ELF AArch64, Mach-O arm64 and
+Mach-O x86-64, each with symbols and stripped. A separate Linux x86-64 run verified
+the fixture and system zlib 1.2.13 / SQLite 3.40.1 libraries through MCP, including
+cached byte spans and a Ghidra comparison. See the [Linux report](https://github.com/morluto/rea/pull/3854#issuecomment-6185870391)
+for measured scope and limitations. Windows, PE and universal Mach-O remain
+unverified by this lane.
+
 ## DOS Ghidra analysis
 
 `npm run verify:ghidra:dos` requires the supported Ghidra and JDK installation
@@ -957,7 +1027,19 @@ the Apple or Windows curated lanes invoke explicitly, which also select that
 lane. Edits to verifier entrypoints run only by manual, licensed or self-hosted
 workflows run static checks and the Linux suite. Website changes run website
 checks. Provider workflow changes validate the workflow and run that provider's
-verifier. Manual-provider/Android workflow changes run workflow validation only.
+verifier. Changes under `bridge/ghidra/` also select the `ghidra_bridge` lane,
+which compiles the packaged bridge sources against the pinned Ghidra 12.1.4
+jars with `npm run verify:ghidra:bridge-sources`. It never launches Ghidra or
+imports a target, so a missing import fails the PR instead of every Ghidra
+session. The launcher, compiler and package checks share named required source
+roles. The manifest must match the complete Java cohort, so deleting a helper
+or adding an unregistered dependency cannot silently change the shipped set.
+The preflight reuses the provider's Java Properties parser with the pinned Node
+toolchain's native type stripping, without an npm install or runtime build.
+The shared release policy owns the verified version and JDK major. CI checks
+the archive pin before downloading and verifies the extracted SDK version;
+changes to these source owners also select the compile lane.
+Manual-provider/Android workflow changes run workflow validation only.
 The four-version SQLite compatibility matrix has its own `sqlite` scope for
 SQLite implementation, contracts, CLI/MCP registration, fixtures, and shared
 runtime boundaries. SQLite and server-lifecycle test edits and the SQLite npm

docs/tool-design.md

@@ -68,6 +68,54 @@ satisfy. A plan without execution does not establish runtime behavior.
   constraints. Report pagination, truncation, cancellation, and partial failure.
   An assumed agent budget is insufficient justification for a limit.
 
+## Own the lifecycle
+
+For tools that acquire resources, identify the owner and the observations that
+establish each transition. Include temporary files and one-shot commands, as
+well as persistent engine sessions. Use the existing provider contracts;
+providers need not expose identical state machines.
+
+- Keep target admission, provider readiness, active work, observed process exit,
+  and resource release separate. A valid process handle, successful launcher,
+  or selected target establishes only its own fact. Publish available identity
+  and exit observations even when launch fails; retain unknown ownership.
+- Authenticate and capture identity through the same boundary on initial and
+  recovery handshakes. Bound abandoned unauthenticated listeners where the
+  protocol permits it; that deadline must not expire authenticated idle sessions
+  or long-running native work. Listener expiry does not prove target closure.
+- Shared startup has one producer and independently cancellable caller waits.
+  Keep it running while any caller still needs it. The final departing caller
+  waits for owned cleanup and retains its failure. Provider closure is distinct
+  from caller cancellation. Retained cleanup capabilities block replacement;
+  derive readiness from the current startup's completion and live connection.
+- For cancellable observations, carry cancellation through result consumption
+  and awaited cleanup. Before returning success, check the signal after the
+  last awaited finalization step.
+  Retain collected observations under typed cancellation. Preserve an existing
+  primary failure and let incomplete cleanup retain its own classification.
+  Client cancellation alone does not prove native work stopped.
+  For atomic publication, document the cancellation cutoff and preserve any
+  committed output's publication receipt.
+- Release only owned resources. Attaching owns a connection or session unless
+  the provider contract explicitly establishes target ownership. A helper exit,
+  missing reply, or missing PID does not establish target exit.
+- Retain successful child cleanup and retry only unresolved children, including
+  when close options change. Preserve observed child failure or activity when
+  another child's health is unavailable; idle plus missing coverage stays
+  unknown.
+  Bind cleanup failures to their exact resource owner. Resolve current cleanup
+  uncertainty only after that owner confirms release, while preserving the
+  original failure and completed observations. Later failure composition must
+  report only resources whose cleanup remains unresolved.
+- Await one-shot cleanup without discarding the completed result or primary
+  diagnostic. Preserve partial observations, remaining resources, and the cleanup
+  owner's recovery instructions. CLI adapters should reuse
+  [`finishAnalysisCommand`](https://github.com/morluto/rea/blob/main/src/cli/finishAnalysisCommand.ts) where its contract
+  applies rather than duplicate close/finally handling.
+
+Exercise the applicable [lifecycle regression scenarios](testing.md#lifecycle-regressions)
+through the producing owner and affected CLI/MCP consumers.
+
 ## Implement and evaluate
 
 Put provider protocols/parsing in adapters and shared CLI/MCP workflows in the

docs/verification/javascript-analysis-effects-2026-10-15.json

@@ -0,0 +1,249 @@
+{
+  "base_revision": "e598b4656ad90d59ede2d10454a0ac34f3813558",
+  "scope": "Canonical effect metadata only; real static Obsidian reads, no target execution",
+  "baseline_source_sha256": {
+    "src/contracts/toolEffects.ts": "39505b5f1501768515229a37ff67e8a1a512faedfb6669aee7a690a2ac3ac633",
+    "src/javascript/analysis/JavaScriptAnalysisWorker.ts": "49ea4e6344060fae3529ffe6e14bfadc534132067cfa5c370263dc7476375be0",
+    "scripts/verify/javascript/module-references.mjs": "9d5f0b8b35ed2ebb4591f9dc3baea0fc116a0c3fd7d8e5ce30ec476c84671ed1"
+  },
+  "final_source_sha256": {
+    "src/contracts/toolEffects.ts": "feb66f1b71c4a6849d1730227c9c74e4efbb67b4537d75fd41ce9ff615c25753",
+    "src/javascript/analysis/JavaScriptAnalysisWorker.ts": "49ea4e6344060fae3529ffe6e14bfadc534132067cfa5c370263dc7476375be0",
+    "scripts/verify/javascript/module-references.mjs": "9d5f0b8b35ed2ebb4591f9dc3baea0fc116a0c3fd7d8e5ce30ec476c84671ed1",
+    "tests/boundary/mcp/electronMcp.test.ts": "e483bec8512cfeafa7f951c8faf25e1aecd6e6987ea74aead14230501912928d",
+    "tests/boundary/mcp/toolSchemaValidity.test.ts": "ab6d9ba37035a6ff9279b820aa8660c50a3ef9f31d25f838c0ba40929148e89c"
+  },
+  "cohorts": {
+    "baseline": {
+      "metadata": {
+        "tool": "analyze_javascript_application",
+        "advertised_annotations": {
+          "readOnlyHint": false,
+          "destructiveHint": false,
+          "idempotentHint": true,
+          "openWorldHint": false
+        },
+        "session_effects": ["mutatesSession", "idempotent"],
+        "session_annotations": ["idempotent"],
+        "catalog_tool_count": 207
+      },
+      "metadata_probe_resources": {
+        "elapsed_seconds": 10.711,
+        "cpu_seconds": 10.503415,
+        "peak_cgroup_bytes": 842891264,
+        "payload_cleanup_verified": true,
+        "events_at_payload_exit": {
+          "memory.events": {
+            "low": 0,
+            "high": 0,
+            "max": 0,
+            "oom": 0,
+            "oom_kill": 0,
+            "oom_group_kill": 0
+          },
+          "pids.events": {
+            "max": 0
+          }
+        }
+      },
+      "real_workflow": {
+        "artifact": {
+          "path": "/tmp/rea-codemode-experiment-20261015/artifacts/obsidian-1.14.4.asar",
+          "sha256": "146ef8470d8cbb4a6db209bdc19bf2bad13350444186c0fd93c31087b142b92f"
+        },
+        "parent_evidence_id": "ev_73ef06e3b9b944f54c5257f7960279127d8518d2a73cd8618bd045dbe44e6b5a",
+        "samples": [
+          {
+            "round": 0,
+            "mode": "references",
+            "elapsed_ms": 109.19401700000162,
+            "calls": 1,
+            "structured_reply_bytes": 22158,
+            "facts_sha256": "d3be5e91d2ec3fd9a95851cfbbdef9c168bb4c7153f78810b8b235e005401299",
+            "matching_modules": 23,
+            "coverage": {
+              "status": "page",
+              "examined": 23,
+              "total": 23,
+              "next_offset": null,
+              "exhausted": true
+            }
+          }
+        ],
+        "token_costs": {
+          "tokenizer": "cl100k_base",
+          "tiktoken_version": "0.12.0",
+          "costs": {
+            "references": {
+              "selected_catalog": 7977,
+              "full_catalog": 195522,
+              "common_request": 53,
+              "common_summary": 2675,
+              "requests": 102,
+              "structured_replies": 5741,
+              "selected_catalog_task_total": 16548,
+              "full_catalog_task_total": 204093
+            }
+          }
+        },
+        "cli_parity": {
+          "evidence_id": "ev_b3586dfdf7d4f01157f8f2978326aaa7103987110f2df4fa5bf3331ff2898749",
+          "facts_sha256": "d3be5e91d2ec3fd9a95851cfbbdef9c168bb4c7153f78810b8b235e005401299",
+          "exact_cli_mcp_view_parity": true
+        }
+      },
+      "real_workflow_resources": {
+        "elapsed_seconds": 30.47,
+        "cpu_seconds": 29.939807,
+        "peak_cgroup_bytes": 1551282176,
+        "payload_cleanup_verified": true,
+        "events_at_payload_exit": {
+          "memory.events": {
+            "low": 0,
+            "high": 0,
+            "max": 0,
+            "oom": 0,
+            "oom_kill": 0,
+            "oom_group_kill": 0
+          },
+          "pids.events": {
+            "max": 0
+          }
+        }
+      }
+    },
+    "final": {
+      "metadata": {
+        "tool": "analyze_javascript_application",
+        "advertised_annotations": {
+          "readOnlyHint": false,
+          "destructiveHint": false,
+          "idempotentHint": true,
+          "openWorldHint": true
+        },
+        "session_effects": [
+          "mutatesSession",
+          "writesFilesystem",
+          "launchesProcess",
+          "idempotent"
+        ],
+        "session_annotations": ["idempotent", "open_world"],
+        "catalog_tool_count": 207
+      },
+      "metadata_probe_resources": {
+        "elapsed_seconds": 10.909,
+        "cpu_seconds": 10.664997,
+        "peak_cgroup_bytes": 853958656,
+        "payload_cleanup_verified": true,
+        "events_at_payload_exit": {
+          "memory.events": {
+            "low": 0,
+            "high": 0,
+            "max": 0,
+            "oom": 0,
+            "oom_kill": 0,
+            "oom_group_kill": 0
+          },
+          "pids.events": {
+            "max": 0
+          }
+        }
+      },
+      "real_workflow": {
+        "artifact": {
+          "path": "/tmp/rea-codemode-experiment-20261015/artifacts/obsidian-1.14.4.asar",
+          "sha256": "146ef8470d8cbb4a6db209bdc19bf2bad13350444186c0fd93c31087b142b92f"
+        },
+        "parent_evidence_id": "ev_73ef06e3b9b944f54c5257f7960279127d8518d2a73cd8618bd045dbe44e6b5a",
+        "samples": [
+          {
+            "round": 0,
+            "mode": "references",
+            "elapsed_ms": 110.17946600000141,
+            "calls": 1,
+            "structured_reply_bytes": 22158,
+            "facts_sha256": "d3be5e91d2ec3fd9a95851cfbbdef9c168bb4c7153f78810b8b235e005401299",
+            "matching_modules": 23,
+            "coverage": {
+              "status": "page",
+              "examined": 23,
+              "total": 23,
+              "next_offset": null,
+              "exhausted": true
+            }
+          }
+        ],
+        "token_costs": {
+          "tokenizer": "cl100k_base",
+          "tiktoken_version": "0.12.0",
+          "costs": {
+            "references": {
+              "selected_catalog": 7977,
+              "full_catalog": 195522,
+              "common_request": 53,
+              "common_summary": 2675,
+              "requests": 102,
+              "structured_replies": 5741,
+              "selected_catalog_task_total": 16548,
+              "full_catalog_task_total": 204093
+            }
+          }
+        },
+        "cli_parity": {
+          "evidence_id": "ev_b3586dfdf7d4f01157f8f2978326aaa7103987110f2df4fa5bf3331ff2898749",
+          "facts_sha256": "d3be5e91d2ec3fd9a95851cfbbdef9c168bb4c7153f78810b8b235e005401299",
+          "exact_cli_mcp_view_parity": true
+        }
+      },
+      "real_workflow_resources": {
+        "elapsed_seconds": 31.059,
+        "cpu_seconds": 30.663163,
+        "peak_cgroup_bytes": 1584574464,
+        "payload_cleanup_verified": true,
+        "events_at_payload_exit": {
+          "memory.events": {
+            "low": 0,
+            "high": 0,
+            "max": 0,
+            "oom": 0,
+            "oom_kill": 0,
+            "oom_group_kill": 0
+          },
+          "pids.events": {
+            "max": 0
+          }
+        }
+      }
+    }
+  },
+  "conditions": "Separate fresh MCP sessions, uncontrolled warm filesystem/build and shared-host scheduling. One CPU, 8 GiB hard cap, 6 GiB early stop, no swap. Resource counters are at payload exit with the observer alive, not final whole-scope counters; memory includes charged file cache, not summed RSS. Whole real-verifier resources include analysis, export, CLI parity and tokenizer; metadata probe resources are separate. Static BPE uses name/description/inputSchema catalog projections plus requests/structured replies, excludes annotations, metadata-probe calls, rendering/replay/reasoning/final-answer/model usage. No performance improvement or equivalence bound established by one before/after pair.",
+  "regressions": {
+    "baseline": "20 executed: 2 expected metadata failures, 18 passed",
+    "final": "27 passed after the test ownership cleanup across canonical effects and MCP/catalog boundaries",
+    "interim_correction": "26 passed before metadata assertions were split into their own target-free case to satisfy function-size lint"
+  },
+  "selected_input": {
+    "input_path": "/tmp/rea-codemode-experiment-20261015/artifacts/obsidian-1.14.4.asar",
+    "detail": "summary",
+    "source_paths": ["package.json", "main.js"]
+  },
+  "selected_view": {
+    "kind": "javascript",
+    "collection": "modules",
+    "projection": "references",
+    "where": {
+      "path": "main.js",
+      "kind": "javascript-module"
+    }
+  },
+  "observer_sources_sha256": {
+    "resource-run.py": "7728862171415e7dee2c11af2ead2b9a54265fccaf82a7113956d36b29065e33",
+    "resource_payload.py": "6f089e645459eba80c5d73e39c21606c78173a44daf49fabc3b88c4631ce51c6"
+  },
+  "runtime": {
+    "platform": "Linux x64",
+    "node": "24.18.0",
+    "npm": "11.16.0",
+    "tiktoken": "0.12.0"
+  }
+}

docs/verification/javascript-analysis-effects-2026-10-15.md

@@ -0,0 +1,121 @@
+# JavaScript analysis effect metadata verification
+
+Tracks #3867 under #4027. The canonical effect record omitted existing internal
+worker launches and source staging. `CapabilityInventory` and SDK registration
+already consume that owner directly; no adapter override is required. The
+[worker](../../src/javascript/analysis/JavaScriptAnalysisWorker.ts) creates a
+private runtime root, stages `.source` files and launches an owned REA worker.
+These are source-derived facts; the analyzed application code is not executed.
+
+## Executed results
+
+Public stdio MCP before/after:
+
+| Surface                      | Before                         | After                                      |
+| ---------------------------- | ------------------------------ | ------------------------------------------ |
+| `tools/list` `openWorldHint` | `false`                        | `true`                                     |
+| `binary_session` effects     | `mutatesSession`, `idempotent` | also `writesFilesystem`, `launchesProcess` |
+| Session annotation flags     | `idempotent`                   | also `open_world`                          |
+| Advertised tools             | 207                            | 207                                        |
+
+Regression assertions failed on the original contract: **2 failed,
+18 passed**. After correction and separating the target-free metadata case, **27 passed** across canonical effects, public
+Electron/JavaScript MCP and SDK-converted catalog validity. The tests retain the
+non-destructive, idempotent hints and exact session effects; no foreign target
+execution is declared.
+
+The existing real verifier analyzed official Obsidian 1.14.4 with `package.json`
+and `main.js` selected, then read the typed module references and verified exact
+public CLI/MCP parity. Both revisions returned the same parent/view Evidence
+identities and 23-row fact digest. Query replies remained 22,158 bytes and the
+selected-schema task remained 16,548 `cl100k_base` BPE tokens (204,093 with the full
+name/description/inputSchema catalog projection). These are static counts, not
+model billing; annotations, the separate metadata probe, host rendering/replay,
+reasoning and final answers are outside this projection. No hosted agent was run.
+
+| Complete real verifier |        Before |         After |
+| ---------------------- | ------------: | ------------: |
+| Elapsed seconds        |        30.470 |        31.059 |
+| CPU seconds            |     29.939807 |     30.663163 |
+| Peak cgroup bytes      | 1,551,282,176 | 1,584,574,464 |
+
+The scope includes shared analysis, temporary export, CLI parity and tokenizer.
+The metadata-only probe separately used 10.711/10.909 s wall,
+10.503415/10.664997 s CPU and 842,891,264/853,958,656 B peak cgroup memory.
+All scopes used one CPU, an 8 GiB hard cap, 6 GiB early stop and no swap; cleanup
+was verified with no OOM. Memory includes charged file cache, not summed RSS.
+Counters were observed at payload exit while the observer remained alive.
+Fresh MCP sessions do not establish an OS-cold cache. Shared-host scheduling and
+filesystem/build warmness were uncontrolled; one pair establishes neither a
+performance improvement nor an equivalence bound. Linux x64 static ASAR analysis
+was executed; live Electron and other platforms were not exercised locally.
+[Compact records and source hashes](javascript-analysis-effects-2026-10-15.json)
+preserve these distinctions without archiving the downloadable application.
+
+## Reproduce
+
+Use pinned Node/npm, `npm ci`, then `npm run build:cached`. Run the following as
+an ES module from the checkout root (`node --input-type=module < probe.mjs`):
+
+```javascript
+import assert from "node:assert/strict";
+import { Client } from "@modelcontextprotocol/client";
+import { StdioClientTransport } from "@modelcontextprotocol/client/stdio";
+import { toolAvailabilityInventory } from "./dist/contracts/toolOutputSchemaPrimitives.js";
+import { closeVerifierResources } from "./scripts/lib/verifier-cleanup.mjs";
+
+const client = new Client({
+  name: "javascript-effect-verification",
+  version: "1",
+});
+const transport = new StdioClientTransport({
+  command: process.execPath,
+  args: ["scripts/rea.mjs", "mcp"],
+  stderr: "pipe",
+});
+let primaryError;
+try {
+  await client.connect(transport);
+  transport.stderr?.resume();
+  const { tools } = await client.listTools();
+  const tool = tools.find(
+    ({ name }) => name === "analyze_javascript_application",
+  );
+  assert.ok(tool);
+  const response = await client.callTool({
+    name: "binary_session",
+    arguments: {},
+  });
+  assert.notEqual(response.isError, true);
+  const inventory = toolAvailabilityInventory.parse(
+    response.structuredContent?.result?.tool_availability,
+  );
+  const group = inventory.groups.find(({ names }) => names.includes(tool.name));
+  assert.ok(group);
+  console.log(
+    JSON.stringify({
+      tool: tool.name,
+      advertised_annotations: tool.annotations,
+      session_effects: group.effects,
+      session_annotations: group.annotations,
+      catalog_tool_count: tools.length,
+    }),
+  );
+} catch (cause) {
+  primaryError = cause;
+  throw cause;
+} finally {
+  await closeVerifierResources(
+    [() => client.close(), () => transport.close()],
+    primaryError,
+  );
+}
+```
+
+For the real workflow, follow the artifact acquisition/digest instructions in
+[module-reference verification](javascript-module-references-2026-10-15.md).
+Run its existing command with `--mode references --rounds 1 --cli-parity` and the
+same selected source paths and tokenizer. That verifier is unchanged by this
+fix. The preserved 8 GiB cgroup observer and its source hashes are documented in
+the same guide. The machine-readable records separate metadata-probe resources
+from the complete real-workflow resources.

docs/verification/javascript-interruption-applicability-2026-10-15.md

@@ -0,0 +1,245 @@
+# Interrupted application processing: applicability and Evidence
+
+Tracks #3948 under #4027. Measured baseline:
+`c7f6b8bad1a34d5efb30ff2a017672dea1fcdabb`.
+
+The question is which processing phases remain unavailable after a source
+interruption, while keeping native inventory and already collected facts.
+Native identity and provider linkage must stay independent of application-wide
+partial coverage. This is static artifact analysis; no target application or
+native provider runs.
+
+## Observed failure and correction
+
+The interruption producer previously attached one generic failure to every
+remaining relevant file. Graph projection interpreted it as JavaScript AST work.
+Real Joplin reproduced an `analyze-javascript-source` / `ast-static-analysis`
+unavailable scope on keytar's native binary and on unvisited package, HTML and
+source-map files. Public CLI/MCP controls reproduced the same applicability
+problem. Deterministic cancellation before native metadata additionally produced
+a phantom JavaScript semantic module for the binary.
+
+The producer now records the applicable operation/authority and whether the
+phase was entered. `analysis_status: interrupted` includes worker startup and
+does not imply parsing completed; `not-started` identifies unvisited phases.
+Native add-ons need no post-inventory source-processing phase, so neither interrupted
+nor caller-excluded native files acquire AST scopes or semantic modules.
+Actual JavaScript failures, structured-file unknowns, caller selection and byte
+identity remain available. The graph consumes these produced facts.
+
+The AST-only unavailable Evidence helper is replaced by an explicitly typed
+static helper. Normal package/JSON/map gaps also retain their artifact-byte
+authority. Per-file progress uses the same applicability owner: JavaScript keeps
+its source phase, and metadata uses `process_javascript_artifact_metadata`.
+Remaining parsers do not run after interruption. Failure sharing is limited to
+five applicable phases during one interruption; no persistent cache is added.
+
+A second review found the alternate full-analysis semantic builder still
+created modules for every excluded file kind. An actual filesystem-producer
+regression on revision `323fae8cbf3e36677bcb7b22ba31fa1a7e3c5cd4` reproduced
+phantom native/JSON/HTML/package/source-map modules, while selected and excluded
+JavaScript were positive controls. Semantic applicability now belongs to the
+shared `recordUnavailableFile` projection entry point, before retaining limits,
+nodes, roots or unknowns. Three duplicate caller checks are removed. Both full
+and incremental construction use this boundary; actual excluded JavaScript
+continues to carry source-bound unknowns and partial semantic coverage.
+
+## Real input and reproduction
+
+Use the [official Joplin v3.7.21 acquisition and extraction instructions](https://github.com/morluto/rea/blob/c7f6b8bad1a34d5efb30ff2a017672dea1fcdabb/docs/verification/javascript-native-handoff-identity-2026-10-15.md).
+The AppImage SHA-256 is
+`5d202dcaca31e14206826aa3f04125e7c218dbc3dc3d98479d811efbd6b856c1`;
+ASAR SHA-256 is
+`767df409b18effe60421fbefcca873de237c7e813eac744ecdeebbc5e825a33a`.
+The 6,324-file inventory includes 46 unpacked entries. Both cohorts stage all
+three native companions; the other 43 unpacked entries remain unavailable.
+
+Build with pinned Node 24.18.0/npm 11.16.0 and locked dependencies. The extended
+existing verifier also runs against the baseline runtime; copy the verifier and
+`scripts/lib/javascript-file-scope-oracle.mjs` from the fixed revision into a
+built baseline checkout when reproducing the failure.
+
+```bash
+artifact_path="/absolute/path/to/cases/available/app.asar"
+common=(
+  --artifact "$artifact_path"
+  --addon-path node_modules/keytar/build/Release/keytar.node
+  --native-sha256 0700016bcd761cbe02d275f2046c7cc42f39f730a50d66e3d3307aa9d413373a
+  --source-path ElectronAppWrapper.js
+  --source-path ElectronAppWrapper.js.map
+  --source-path index.html
+  --source-path package.json
+  --source-path node_modules/keytar/build/Release/keytar.node
+  --structured-path ElectronAppWrapper.js.map
+  --structured-path index.html
+  --structured-path package.json
+  --structured-expectations '{"ElectronAppWrapper.js.map":{"operation":"parse-local-source-map","analysis_status":"not-started"},"index.html":{"operation":"parse-html-script-references","analysis_status":"not-started"},"package.json":{"operation":"parse-package-json","analysis_status":"not-started"}}'
+  --rounds 3 --cli-parity
+  --tokenizer-python /absolute/path/to/python-with-tiktoken-0.12.0
+)
+node scripts/verify/javascript/native-handoffs.mjs "${common[@]}" \
+  --analysis-timeout-ms 1 --expect-timeout --require-safe
+node scripts/verify/javascript/native-handoffs.mjs "${common[@]}" \
+  --validate-structured-from-export --require-safe
+```
+
+Omit `--require-safe` on the baseline timeout run to record its incorrect
+applicability. Its diagnostic exit 0 is not a passing analyst task. A supported
+1 ms deadline includes worker startup; the verifier requires a typed timeout
+and retained partial reference for that cohort. Each structured path supplies an
+independent expected operation and phase-entry status; the safety oracle matches
+all gaps exactly, including unavailable state and unknown confidence. Missing or
+conflicting expectations do not pass.
+
+The native path is explicitly selected. An earlier structured-only run excluded
+it and therefore did not reproduce the native interruption. Input selections
+and that preliminary result remain recorded. Both final cohorts preserve
+partial coverage for unselected/unavailable scopes.
+
+The supplied native Evidence is simulated solely to check exact subject linkage.
+The verifier checks physical companion bytes and the exported parent's root
+digest, validates public CLI/MCP native-trace parity, and reports only after
+owned transport/command/private-file cleanup. Applications and large Evidence
+are regenerable and are not archived.
+
+## Two distinct task projections
+
+The **timeout task** includes the expected typed-timeout preparation, one inline
+native trace and three inline structured-path traces. It checks that native has
+no AST failure and that each unvisited structured phase remains unavailable with
+the appropriate artifact-byte authority and `not-started` status.
+
+The **default-deadline control** models the summary plus one inline native trace.
+Its named structured paths are additionally validated from the already exported
+complete parent, explicitly labelled `exported-parent-validation`. Those checks
+are diagnostic overhead outside the modeled native-task tokens and inside total
+verifier resources. They are not presented as inline metadata answers.
+
+A control that also requested the complete outgoing `package.json` feature trace
+hit stdio admission twice: at least 10,587,373 bytes against a 10,484,736-byte
+result budget, with identical retained parent/trace identities. It is not a
+completed equivalent task or an efficiency baseline. [#4190](https://github.com/morluto/rea/issues/4190)
+records the separate need for exact retained metadata/native fact views. Complete
+export remains available. Do not compare costs across these different tasks.
+
+## Measurements
+
+| Cohort                         | Revision |  Wall s |      CPU s | Peak cgroup bytes | Applicability |
+| ------------------------------ | -------- | ------: | ---------: | ----------------: | ------------- |
+| Timeout, four inline traces    | baseline |  60.738 |  60.289464 |     3,092,045,824 | wrong scope   |
+| Timeout, four inline traces    | final    |  57.515 |  57.089883 |     2,763,276,288 | pass          |
+| Default control, native inline | baseline |  61.542 |  61.135522 |     2,808,336,384 | pass          |
+| Default control, native inline | final    | 186.085 | 142.371100 |     2,959,585,280 | pass          |
+
+| Defined task   | Selected catalog + parsed/structured replies BPE | Selected catalog + complete tool replies BPE |
+| -------------- | -----------------------------------------------: | -------------------------------------------: |
+| Timeout        |                                  21,136 → 20,161 |                              36,468 → 34,471 |
+| Native control |                                  11,776 → 11,720 |                              17,738 → 17,632 |
+
+The full-catalog alternatives and per-query samples are preserved in JSON.
+Native content identity, observed byte provenance, source selection and handoff
+provider Evidence remain equal. Corrected global provenance changes derived
+parent/trace identities; null partial-Evidence subjects remain null. Both final
+runs have exact CLI/MCP native-trace parity and match the earlier post-fix
+controls in every recorded non-timing field.
+
+The whole verifier includes input hashing, analysis, three native trace rounds,
+applicable structured queries, export, CLI parity and tokenization. Every heavy
+run uses the preserved [resource observer](https://github.com/morluto/rea/blob/b3bdabab694af1f9801eff7caf75001b92314a58/experiments/analysis-efficiency/resource-run.py):
+one CPU, 8 GiB hard cap, 6 GiB early stop and no swap. Cgroup peaks include charged
+file cache, not summed RSS. Counters are observed at payload exit while the
+observer remains alive, not final whole-scope counters. Completed runs verify
+owned cleanup and report no OOM.
+
+Each MCP process is fresh, but input hashing warms its file cache. Build/OS caches
+and shared-host scheduling are uncontrolled. First and repeated retained-query
+wall samples remain separate; isolated query CPU/memory are unmeasured. An
+earlier fixed timeout took 150.173 s wall / 119.468013 s CPU; its exact unchanged
+runtime repeat took 58.827 s / 58.482996 s, with identical non-timing facts,
+Evidence and BPE. Both receipts are retained. The final default control also took 186.085 s wall / 142.371100 s CPU; its
+unchanged-runtime repeat took 62.651 s / 62.240964 s with identical non-timing
+results. The slow final row remains reported. Attribution remains unresolved;
+these runs establish no speedup or performance-equivalence bound.
+
+The stricter post-review verifier reruns also passed without changing production
+code: timeout 162.645 s wall / 125.876114 s CPU / 2,809,643,008 B peak; default
+control 61.367 s / 60.970077 s / 2,526,949,376 B. Both preserve the original final
+analytical fields, identities, coverage and BPE exactly; only timings and the new
+explicit oracle-input field differ. These additional receipts do not replace the
+slower rows above.
+
+After centralizing semantic applicability, both real public controls passed again:
+timeout 60.443 s wall / 59.970344 s CPU / 3,078,295,552 B peak; default control
+63.574 s / 63.177884 s / 3,189,719,040 B. Analytical results, identities, coverage,
+BPE and exact CLI/MCP parity remain unchanged. JSON identifies the runtime and
+verifier source manifests for each cohort and keeps all earlier receipts.
+
+Static `cl100k_base`/tiktoken 0.12.0 projections include the selected or complete
+name/description/inputSchema catalog, requests and structured Evidence or parsed
+text errors. Complete-tool-reply alternatives include MCP text/structured
+duplication. The selected two-tool catalog is 5,570 BPE; the complete projection
+is 195,522. Output schemas, host rendering/replay, notifications, model reasoning,
+final answers and billing are unmeasured. Repeated rounds, export, CLI parity,
+tokenization and exported-parent diagnostics are validation overhead outside the
+defined task tokens but inside resource receipts. Removing unsupported claims
+from the incorrect timeout baseline earns no efficiency credit.
+
+## Verification and limitations
+
+The initial production fix had 49 distinct passing cases across eight files. The initial
+pre-progress-cleanup seven-file selection passed 44/44. After progress cleanup,
+the same selection passed 43 and hit the existing 30-second test deadline in one
+native identity case; an unchanged two-file rerun passed both native cases and
+five public CLI cancellation cases. The original deadline was not increased.
+The runtime verifier collects interruption-bearing scopes; separate regressions
+cover native caller exclusion and malformed structured data. Source selection,
+malformed structures, actual worker resource constraints,
+retained large partial observations, cancellation/publication/cleanup, native
+identity and both public adapters remain covered.
+
+The first review found that the real verifier previously accepted any non-JavaScript
+operation with artifact-byte authority and ignored phase-entry status. The
+scenario now supplies exact operation/status expectations independently of the
+production phase map. Six additional oracle cases use the actual retained
+Joplin snapshots and reject wrong metadata operations, entered or missing
+statuses, and absent or conflicting gaps. They passed, bringing the distinct
+coverage to 55 cases across nine files. The production implementation did not
+change during this review correction.
+
+After the shared semantic-boundary correction from the second review, 78
+distinct cases passed across ten files: the nine-file selection passed 65/65,
+and the original root reconstruction file passed 13/13. This includes the
+alternate-builder owning test file's 23 cases. The new producer-boundary
+regression checks both construction paths, all five excluded metadata/native
+kinds, and selected/excluded JavaScript controls. Real Joplin controls exercise
+the public incremental workflow; the alternate full-builder path has filesystem
+boundary verification rather than a separately measured real-app workflow.
+
+Recorded attempts also include the verifier's initial non-null-subject assumption
+and an invalid top-level semantic-module-path absence check. Both were corrected
+before their assertions were used as proof. Corrected cancellation reproduction
+shows the actual native AST scope and phantom module on the baseline. Final
+source and execution-driver versions, source pins, input digests, prior attempts
+and resource receipts are in the [compact JSON record](javascript-interruption-applicability-2026-10-15.json).
+
+Before review, `npm run check && npm run docs:check` passed, including five consistency
+tests: 165.553 s wall / 150.247257 s CPU /
+4,232,491,008 B peak cgroup memory, clean owned exit and no OOM.
+The first full check stopped on a 172-line test callback exceeding the 150-line
+rule. A named assertion helper now shares scope lookup across all four public
+results; the strengthened six-case file passed again before the final gates.
+After the review correction, the full static/docs gate passed again, including
+the five consistency tests: 111.395 s wall / 110.979710 s CPU /
+4,347,125,760 B peak, with clean owned exit and no OOM.
+After centralizing semantic applicability, the final full static/docs gate also
+passed, including five consistency tests: 117.636 s wall / 117.194118 s CPU /
+4,263,153,664 B peak, clean owned exit and no OOM.
+
+At the initial PR revision, 36 CI checks passed; macOS `hdiutil create` failed
+while creating its DMG fixture before REA verification, also failing the aggregate
+check. No Apple fixture/workflow code changed. That attempt provides no passing
+Apple runtime evidence.
+
+Live Electron/application execution, real native-provider behavior, other-platform
+runtime coverage and hosted-agent costs are **NOT_ENOUGH_INFO**. The verification
+does not claim them.

docs/verification/javascript-literal-byte-budget-2026-10-15.json

@@ -0,0 +1,547 @@
+{
+  "base_revision": "e9a71efa31d6c3675b67938b8a79fa4ddf01604c",
+  "scope": "Semantic per-value string-byte normalization; real static shipped-application control, no hosted-model performance claim",
+  "runtime": {
+    "platform": "Linux x64",
+    "node": "24.18.0",
+    "npm": "11.16.0",
+    "tiktoken": "0.12.0"
+  },
+  "baseline_source_sha256": {
+    "src/domain/javascript/javascriptSemanticValues.ts": "df3fc30b5ba91901a85fe56f200355930da32d054bb026a747e3414af233b6b5",
+    "src/domain/javascript/javascriptSemanticPrimitives.ts": "c12b29f0faf4b6511777cba7ed8ac007c6d6170d657f01e43e194feacb3e0601",
+    "src/domain/javascript/javascriptSemanticResourceLimits.ts": "0203d207e0d6bb8aaa17746090f008876a11653afd660fce3dc44f3fda646a6c",
+    "scripts/verify/javascript/module-references.mjs": "9d5f0b8b35ed2ebb4591f9dc3baea0fc116a0c3fd7d8e5ce30ec476c84671ed1"
+  },
+  "final_source_sha256": {
+    "scripts/verify/javascript/module-references.mjs": "9d5f0b8b35ed2ebb4591f9dc3baea0fc116a0c3fd7d8e5ce30ec476c84671ed1",
+    "src/application/javascript/JavaScriptSemanticGraphDataProjection.ts": "844d4639401a9f050ce42c80192572a6d0d74fe4f81b37e93a2c20aaca9fb9a6",
+    "src/domain/javascript/javascriptSemanticAnalysis.ts": "a3f802ca3446c2b5376cc58b19d818eff318e6775bd47752793c3f649994f8a2",
+    "src/domain/javascript/javascriptSemanticAnalysis.unions.test.ts": "2e78717d30e637b5fc5bab6d0eff063f7a62e0199d33697c1cdfaf8317baa698",
+    "src/domain/javascript/javascriptSemanticDataEffectHelpers.ts": "8f9cf1a36f05b03dd816eaf9a6cdba0160d34ad7d17a961206e1bc32444d2ccc",
+    "src/domain/javascript/javascriptSemanticDataEffects.ts": "e69408f2f5910a0107fd5d0cae0ff60ed706448857bc8e535c37cac1a0cd4d37",
+    "src/domain/javascript/javascriptSemanticIr.ts": "68dad360d0a9e6bd72108b2717cae756b84ca1efb48455248ded2ff699d2c113",
+    "src/domain/javascript/javascriptSemanticPrimitives.ts": "899eeb6ccdf666f504f8b722327599997c924c8e4fb13a1cbd13edaf08c96640",
+    "src/domain/javascript/javascriptSemanticResourceLimits.ts": "0993a397f2a2a48c2b6c35494e940b3fce5c2c9032cfe33c7cf0401128c79dde",
+    "src/domain/javascript/javascriptSemanticState.ts": "bfcface5dcda22594f47ed390efa15f9140fe41cb2d5995896e52dd0ee31b978",
+    "src/domain/javascript/javascriptSemanticValues.ts": "a4ba51cc60fff38ef11fd5af35a2e1ad753cdb2dea9d44895f91d89320c5ddeb",
+    "tests/acceptance/applications/javascriptEndpointPayload.test.ts": "b3cfcde1c5b5e0698f84afd51f4b53e42e56101267707c20c0d038320f5cfbd2",
+    "tests/boundary/mcp/javascriptLiteralBudget.test.ts": "070106115b482269794799e448bd20b8e02df1f511f4ceb2b8fa75612c7f78e4",
+    "tests/boundary/mcp/javascriptLiteralPayload.test.ts": "03618272c605f8877ed171ab39e037e4d80eac72b7478421640076f739adfb50"
+  },
+  "artifact": {
+    "path": "/tmp/rea-codemode-experiment-20261015/artifacts/obsidian-1.14.4.asar",
+    "sha256": "146ef8470d8cbb4a6db209bdc19bf2bad13350444186c0fd93c31087b142b92f"
+  },
+  "selected_source_paths": ["package.json", "main.js"],
+  "where": {
+    "path": "main.js",
+    "kind": "javascript-module"
+  },
+  "identical_parent_evidence_id": "ev_73ef06e3b9b944f54c5257f7960279127d8518d2a73cd8618bd045dbe44e6b5a",
+  "cli_parity": {
+    "evidence_id": "ev_b3586dfdf7d4f01157f8f2978326aaa7103987110f2df4fa5bf3331ff2898749",
+    "facts_sha256": "d3be5e91d2ec3fd9a95851cfbbdef9c168bb4c7153f78810b8b235e005401299",
+    "exact_cli_mcp_view_parity": true
+  },
+  "unchanged_query_shapes": {
+    "legacy": {
+      "calls": 24,
+      "structured_reply_bytes": 325201,
+      "facts_sha256": "d3be5e91d2ec3fd9a95851cfbbdef9c168bb4c7153f78810b8b235e005401299",
+      "matching_modules": 23,
+      "coverage": {
+        "status": "page",
+        "examined": 55,
+        "total": 55,
+        "next_offset": null,
+        "exhausted": true
+      }
+    },
+    "references": {
+      "calls": 1,
+      "structured_reply_bytes": 22158,
+      "facts_sha256": "d3be5e91d2ec3fd9a95851cfbbdef9c168bb4c7153f78810b8b235e005401299",
+      "matching_modules": 23,
+      "coverage": {
+        "status": "page",
+        "examined": 23,
+        "total": 23,
+        "next_offset": null,
+        "exhausted": true
+      }
+    }
+  },
+  "unchanged_token_costs": {
+    "tokenizer": "cl100k_base",
+    "tiktoken_version": "0.12.0",
+    "costs": {
+      "legacy": {
+        "selected_catalog": 7977,
+        "full_catalog": 195522,
+        "common_request": 53,
+        "common_summary": 2675,
+        "requests": 2836,
+        "structured_replies": 71761,
+        "selected_catalog_task_total": 85302,
+        "full_catalog_task_total": 272847
+      },
+      "references": {
+        "selected_catalog": 7977,
+        "full_catalog": 195522,
+        "common_request": 53,
+        "common_summary": 2675,
+        "requests": 102,
+        "structured_replies": 5741,
+        "selected_catalog_task_total": 16548,
+        "full_catalog_task_total": 204093
+      }
+    }
+  },
+  "cohorts": {
+    "baseline": {
+      "preparation_ms": 19089.119622,
+      "query_samples": [
+        {
+          "round": 0,
+          "mode": "legacy",
+          "elapsed_ms": 383.8260569999984
+        },
+        {
+          "round": 0,
+          "mode": "references",
+          "elapsed_ms": 15.778886999996757
+        },
+        {
+          "round": 1,
+          "mode": "references",
+          "elapsed_ms": 13.15173700000014
+        },
+        {
+          "round": 1,
+          "mode": "legacy",
+          "elapsed_ms": 240.44363500000327
+        },
+        {
+          "round": 2,
+          "mode": "legacy",
+          "elapsed_ms": 346.04796499999793
+        },
+        {
+          "round": 2,
+          "mode": "references",
+          "elapsed_ms": 13.727306999997381
+        }
+      ],
+      "resources": {
+        "elapsed_seconds": 33.394,
+        "cpu_seconds": 33.016802,
+        "peak_cgroup_bytes": 1979863040,
+        "payload_cleanup_verified": true,
+        "events_at_payload_exit": {
+          "memory.events": {
+            "low": 0,
+            "high": 0,
+            "max": 0,
+            "oom": 0,
+            "oom_kill": 0,
+            "oom_group_kill": 0
+          },
+          "pids.events": {
+            "max": 0
+          }
+        }
+      }
+    },
+    "final": {
+      "preparation_ms": 19061.292544,
+      "query_samples": [
+        {
+          "round": 0,
+          "mode": "legacy",
+          "elapsed_ms": 344.6316079999997
+        },
+        {
+          "round": 0,
+          "mode": "references",
+          "elapsed_ms": 14.438351000000694
+        },
+        {
+          "round": 1,
+          "mode": "references",
+          "elapsed_ms": 11.38763100000142
+        },
+        {
+          "round": 1,
+          "mode": "legacy",
+          "elapsed_ms": 298.41579600000114
+        },
+        {
+          "round": 2,
+          "mode": "legacy",
+          "elapsed_ms": 211.45682099999976
+        },
+        {
+          "round": 2,
+          "mode": "references",
+          "elapsed_ms": 10.275370999999723
+        }
+      ],
+      "resources": {
+        "elapsed_seconds": 32.949,
+        "cpu_seconds": 32.559187,
+        "peak_cgroup_bytes": 1880096768,
+        "payload_cleanup_verified": true,
+        "events_at_payload_exit": {
+          "memory.events": {
+            "low": 0,
+            "high": 0,
+            "max": 0,
+            "oom": 0,
+            "oom_kill": 0,
+            "oom_group_kill": 0
+          },
+          "pids.events": {
+            "max": 0
+          }
+        }
+      }
+    }
+  },
+  "conditions": "First query is after shared artifact analysis; subsequent queries share one MCP session. Legacy item reads are serial; a concurrent legacy baseline is unmeasured. Filesystem/build warmness uncontrolled. Token costs count one post-change catalog, preparation request/summary and each request/structured reply once; client rendering, replay, reasoning, final answer and model latency are unmeasured. Preparation and parity/tokenization overhead are outside per-query elapsed times; external family-resource receipts include the complete verifier and children. Before/after used separate fresh MCP sessions. CPU/memory includes analysis, query arms, export, CLI parity and tokenization; cgroup memory includes file cache, not summed RSS. Counters are at payload exit while observer lives, not final scope counters. One before/after control establishes no speedup or equivalence bound. Catalog BPE projection excludes annotations/output schemas.",
+  "resource_policy": {
+    "memory_max_bytes": 8589934592,
+    "early_stop_bytes": 6442450944,
+    "cpu_quota_percent": 100,
+    "swap": false
+  },
+  "observer_sources_sha256": {
+    "resource-run.py": "7728862171415e7dee2c11af2ead2b9a54265fccaf82a7113956d36b29065e33",
+    "resource_payload.py": "6f089e645459eba80c5d73e39c21606c78173a44daf49fabc3b88c4631ce51c6"
+  },
+  "regressions": {
+    "original_main": "Quoted/template oversized literals failed through public CLI/MCP; addition/small controls passed. Corrected domain boundary reproduced 2 failures/16 passes.",
+    "review_head": "8 failed/25 passed before fixes for projection tags, encountered-but-discarded limits and bare quoted/template defaults; includes exact null default control.",
+    "interim_review_fix": "65/66 focused passed; old doubling-chain expectation omitted a genuine expression-depth limit now observed. Adjusted that assertion to preserve the byte-limit requirement while allowing other observed classes.",
+    "final_focused": "66 passed across five files on the restored 38d263a runtime. Final corrected public selection passed 7/7 in two files: the six literal/default/member cases plus the augmented original large-endpoint acceptance. 67 distinct focused cases across invocations; not a single 67-case invocation.",
+    "final_semantic_suite": "932 passed in 36 files on 38d263a. Final production src tree (including these domain tests) verified byte-identical to that revision; no repeated semantic run claimed.",
+    "arithmetic": "Computed 174762*6+2=1048574 and 174763*6+2=1048580. Corrected the erroneous published 1048578 total.",
+    "final_grouped_domain": "27 passed after grouping observations/defaults/allocation within function-size rules; no scenarios removed.",
+    "endpoint_contract_correction": "Restored exact endpoint lookup identities rather than shrinking the existing large-endpoint fixture. Added the identical oversized string as an exported primitive in that acceptance workflow: literal normalization must report its bound while full endpoint lookup/feature trace/semantic trace/version comparison continue to work. Production src tree restored byte-identical to 38d263a; 932 semantic tests from that state remain applicable. Final corrected public tests/gates recorded after execution.",
+    "final_corrected_gates": "Full static gate and docs gate (5 consistency tests) passed after endpoint identity restoration and strengthened unchanged-large-input acceptance. Production src remains byte-identical to 38d263a; final public 7/7 and fresh real control passed."
+  },
+  "prior_attempts": [
+    {
+      "stage": "Initial literal normalization, incomplete for member/default paths identified in review",
+      "revision": "40b9d3c5b9cbeae7661d8e351e1b5ef6e5338931",
+      "source_sha256": {
+        "src/domain/javascript/javascriptSemanticValues.ts": "2debbb4f32d7d189819f2413f0f7d05b9bdcf497451a1c578475e3f2ec6c8554",
+        "src/domain/javascript/javascriptSemanticPrimitives.ts": "899eeb6ccdf666f504f8b722327599997c924c8e4fb13a1cbd13edaf08c96640",
+        "src/domain/javascript/javascriptSemanticResourceLimits.ts": "0203d207e0d6bb8aaa17746090f008876a11653afd660fce3dc44f3fda646a6c",
+        "src/domain/javascript/javascriptSemanticAnalysis.unions.test.ts": "747d893d6ddd91cf24d3b62b7eb50e0af9b1715ab355bbfc2e10ed0dc171ec17",
+        "tests/boundary/mcp/javascriptLiteralBudget.test.ts": "46ceb5a169de3eec3a335b05ef4aa7b1b341a5ac06ad325bd191e862c5b6be70",
+        "tests/boundary/mcp/javascriptLiteralPayload.test.ts": "03618272c605f8877ed171ab39e037e4d80eac72b7478421640076f739adfb50",
+        "scripts/verify/javascript/module-references.mjs": "9d5f0b8b35ed2ebb4591f9dc3baea0fc116a0c3fd7d8e5ce30ec476c84671ed1"
+      },
+      "real_control_cohort": {
+        "preparation_ms": 18876.764036,
+        "query_samples": [
+          {
+            "round": 0,
+            "mode": "legacy",
+            "elapsed_ms": 346.7870870000006
+          },
+          {
+            "round": 0,
+            "mode": "references",
+            "elapsed_ms": 18.34765800000241
+          },
+          {
+            "round": 1,
+            "mode": "references",
+            "elapsed_ms": 11.56626999999935
+          },
+          {
+            "round": 1,
+            "mode": "legacy",
+            "elapsed_ms": 227.17703500000061
+          },
+          {
+            "round": 2,
+            "mode": "legacy",
+            "elapsed_ms": 251.5148330000011
+          },
+          {
+            "round": 2,
+            "mode": "references",
+            "elapsed_ms": 10.888227999999799
+          }
+        ],
+        "resources": {
+          "elapsed_seconds": 31.347,
+          "cpu_seconds": 31.05517,
+          "peak_cgroup_bytes": 1589915648,
+          "payload_cleanup_verified": true,
+          "events_at_payload_exit": {
+            "memory.events": {
+              "low": 0,
+              "high": 0,
+              "max": 0,
+              "oom": 0,
+              "oom_kill": 0,
+              "oom_group_kill": 0
+            },
+            "pids.events": {
+              "max": 0
+            }
+          }
+        }
+      },
+      "validation": {
+        "baseline_public": "Quoted and cooked-template cases retained 200000-unit normalized literals and failed; addition and small exact controls passed. Aggregate-capacity test passed.",
+        "baseline_domain_corrected": "2 over-bound failures, 16 passed. Initial combined-run domain assertions had incorrectly required an empty optional resourceLimits array; corrected before using the domain reproduction as evidence.",
+        "fixed_focused": "55 passed across five files.",
+        "semantic_suite": "923 passed across 36 files.",
+        "capacity_after_fixture_cleanup": "1 passed; all 16 bounded values, aggregate size >= original, typed identities and retained tracing preserved.",
+        "interim_static_failure": "Typecheck rejected resume() on the new test SDK stderr Stream declaration. Switched to its declared data event, matching existing typed tests; analyzer source unchanged.",
+        "public_boundary_after_stream_type_fix": "4 passed across actual CLI and stdio MCP; analyzer source unchanged."
+      }
+    },
+    {
+      "stage": "Projection/default correction; final review subsequently found unnormalized request endpoints",
+      "revision": "38d263a369e7853ccfae1c8e0a3bd387f920c7cf",
+      "source_sha256": {
+        "scripts/verify/javascript/module-references.mjs": "9d5f0b8b35ed2ebb4591f9dc3baea0fc116a0c3fd7d8e5ce30ec476c84671ed1",
+        "src/application/javascript/JavaScriptSemanticGraphDataProjection.ts": "844d4639401a9f050ce42c80192572a6d0d74fe4f81b37e93a2c20aaca9fb9a6",
+        "src/domain/javascript/javascriptSemanticAnalysis.ts": "a3f802ca3446c2b5376cc58b19d818eff318e6775bd47752793c3f649994f8a2",
+        "src/domain/javascript/javascriptSemanticAnalysis.unions.test.ts": "2e78717d30e637b5fc5bab6d0eff063f7a62e0199d33697c1cdfaf8317baa698",
+        "src/domain/javascript/javascriptSemanticDataEffectHelpers.ts": "8f9cf1a36f05b03dd816eaf9a6cdba0160d34ad7d17a961206e1bc32444d2ccc",
+        "src/domain/javascript/javascriptSemanticDataEffects.ts": "e69408f2f5910a0107fd5d0cae0ff60ed706448857bc8e535c37cac1a0cd4d37",
+        "src/domain/javascript/javascriptSemanticIr.ts": "68dad360d0a9e6bd72108b2717cae756b84ca1efb48455248ded2ff699d2c113",
+        "src/domain/javascript/javascriptSemanticPrimitives.ts": "899eeb6ccdf666f504f8b722327599997c924c8e4fb13a1cbd13edaf08c96640",
+        "src/domain/javascript/javascriptSemanticResourceLimits.ts": "0993a397f2a2a48c2b6c35494e940b3fce5c2c9032cfe33c7cf0401128c79dde",
+        "src/domain/javascript/javascriptSemanticState.ts": "bfcface5dcda22594f47ed390efa15f9140fe41cb2d5995896e52dd0ee31b978",
+        "src/domain/javascript/javascriptSemanticValues.ts": "a4ba51cc60fff38ef11fd5af35a2e1ad753cdb2dea9d44895f91d89320c5ddeb",
+        "tests/boundary/mcp/javascriptLiteralBudget.test.ts": "070106115b482269794799e448bd20b8e02df1f511f4ceb2b8fa75612c7f78e4",
+        "tests/boundary/mcp/javascriptLiteralPayload.test.ts": "03618272c605f8877ed171ab39e037e4d80eac72b7478421640076f739adfb50"
+      },
+      "real_control_cohort": {
+        "preparation_ms": 19599.643274,
+        "query_samples": [
+          {
+            "round": 0,
+            "mode": "legacy",
+            "elapsed_ms": 396.9265450000021
+          },
+          {
+            "round": 0,
+            "mode": "references",
+            "elapsed_ms": 15.658120000000054
+          },
+          {
+            "round": 1,
+            "mode": "references",
+            "elapsed_ms": 12.24905100000251
+          },
+          {
+            "round": 1,
+            "mode": "legacy",
+            "elapsed_ms": 237.03392900000108
+          },
+          {
+            "round": 2,
+            "mode": "legacy",
+            "elapsed_ms": 357.78818299999693
+          },
+          {
+            "round": 2,
+            "mode": "references",
+            "elapsed_ms": 14.37622100000226
+          }
+        ],
+        "resources": {
+          "elapsed_seconds": 33.969,
+          "cpu_seconds": 33.499223,
+          "peak_cgroup_bytes": 1893015552,
+          "payload_cleanup_verified": true,
+          "events_at_payload_exit": {
+            "memory.events": {
+              "low": 0,
+              "high": 0,
+              "max": 0,
+              "oom": 0,
+              "oom_kill": 0,
+              "oom_group_kill": 0
+            },
+            "pids.events": {
+              "max": 0
+            }
+          }
+        }
+      },
+      "validation": {
+        "original_main": "Quoted/template oversized literals failed through public CLI/MCP; addition/small controls passed. Corrected domain boundary reproduced 2 failures/16 passes.",
+        "review_head": "8 failed/25 passed before fixes for projection tags, encountered-but-discarded limits and bare quoted/template defaults; includes exact null default control.",
+        "interim_review_fix": "65/66 focused passed; old doubling-chain expectation omitted a genuine expression-depth limit now observed. Adjusted that assertion to preserve the byte-limit requirement while allowing other observed classes.",
+        "final_focused": "66 passed in five files.",
+        "final_semantic_suite": "932 passed in 36 files.",
+        "arithmetic": "Computed 174762*6+2=1048574 and 174763*6+2=1048580. Corrected the erroneous published 1048578 total.",
+        "final_grouped_domain": "27 passed after grouping observations/defaults/allocation within function-size rules; no scenarios removed.",
+        "endpoint_review_baseline": "Production at 38d263a: 7 failed/14 passed across 13 new domain and 8 CLI/MCP cases before endpoint normalization. The later public linked-response case was not in this baseline run. Regression tests were uncommitted; final formatted test hashes are preserved, not claimed as the baseline test bytes.",
+        "endpoint_selection_attempt": "First six-file selection exited before execution because a signed-zero test filename was mistyped; corrected the exact path. No test result claimed for that attempt."
+      }
+    },
+    {
+      "stage": "Rejected endpoint normalization: violated existing full-endpoint lookup/trace contract; reverted after CI acceptance failure",
+      "revision": "580887209c4fceda432eac4083fac82a0dd78f15",
+      "source_sha256": {
+        "scripts/verify/javascript/module-references.mjs": "9d5f0b8b35ed2ebb4591f9dc3baea0fc116a0c3fd7d8e5ce30ec476c84671ed1",
+        "src/application/javascript/JavaScriptSemanticGraphDataProjection.ts": "a648f97e631985edcb5259ff4f1fdf0332dca18b8c10b57e050e36a785c1c524",
+        "src/domain/javascript/javascriptSemanticAnalysis.requestBudget.test.ts": "8e84ffa70d834f5589d04118f0aaf33b0383b15ea97545ff69c66a74473aca73",
+        "src/domain/javascript/javascriptSemanticAnalysis.ts": "a3f802ca3446c2b5376cc58b19d818eff318e6775bd47752793c3f649994f8a2",
+        "src/domain/javascript/javascriptSemanticAnalysis.unions.test.ts": "2e78717d30e637b5fc5bab6d0eff063f7a62e0199d33697c1cdfaf8317baa698",
+        "src/domain/javascript/javascriptSemanticDataEffectHelpers.ts": "8f9cf1a36f05b03dd816eaf9a6cdba0160d34ad7d17a961206e1bc32444d2ccc",
+        "src/domain/javascript/javascriptSemanticDataEffects.ts": "eed961e0b1d8a320e76bbba933408e950b4a4a446110401e8cf470f1e74d8315",
+        "src/domain/javascript/javascriptSemanticIr.ts": "3b2ede73dbcfc74c4014e4d1affcbf00ecccc29a422ed495a7a3f2f517056c9b",
+        "src/domain/javascript/javascriptSemanticPrimitives.ts": "899eeb6ccdf666f504f8b722327599997c924c8e4fb13a1cbd13edaf08c96640",
+        "src/domain/javascript/javascriptSemanticResourceLimits.ts": "0993a397f2a2a48c2b6c35494e940b3fce5c2c9032cfe33c7cf0401128c79dde",
+        "src/domain/javascript/javascriptSemanticState.ts": "bfcface5dcda22594f47ed390efa15f9140fe41cb2d5995896e52dd0ee31b978",
+        "src/domain/javascript/javascriptSemanticValues.ts": "a4ba51cc60fff38ef11fd5af35a2e1ad753cdb2dea9d44895f91d89320c5ddeb",
+        "tests/boundary/mcp/javascriptLiteralBudget.test.ts": "61319ea954de7f256841f43ca9355a1706ce8e93d30f2a8ba851d5e717bd19e7",
+        "tests/boundary/mcp/javascriptLiteralPayload.test.ts": "03618272c605f8877ed171ab39e037e4d80eac72b7478421640076f739adfb50"
+      },
+      "real_control_cohort": {
+        "preparation_ms": 19939.160272,
+        "query_samples": [
+          {
+            "round": 0,
+            "mode": "legacy",
+            "elapsed_ms": 348.3974309999976
+          },
+          {
+            "round": 0,
+            "mode": "references",
+            "elapsed_ms": 13.97052899999835
+          },
+          {
+            "round": 1,
+            "mode": "references",
+            "elapsed_ms": 12.060379000002285
+          },
+          {
+            "round": 1,
+            "mode": "legacy",
+            "elapsed_ms": 212.01217800000086
+          },
+          {
+            "round": 2,
+            "mode": "legacy",
+            "elapsed_ms": 223.73954800000138
+          },
+          {
+            "round": 2,
+            "mode": "references",
+            "elapsed_ms": 10.450989999997546
+          }
+        ],
+        "resources": {
+          "elapsed_seconds": 33.062,
+          "cpu_seconds": 32.71624,
+          "peak_cgroup_bytes": 1897406464,
+          "payload_cleanup_verified": true,
+          "events_at_payload_exit": {
+            "memory.events": {
+              "low": 0,
+              "high": 0,
+              "max": 0,
+              "oom": 0,
+              "oom_kill": 0,
+              "oom_group_kill": 0
+            },
+            "pids.events": {
+              "max": 0
+            }
+          }
+        }
+      },
+      "validation": {
+        "original_main": "Quoted/template oversized literals failed through public CLI/MCP; addition/small controls passed. Corrected domain boundary reproduced 2 failures/16 passes.",
+        "review_head": "8 failed/25 passed before fixes for projection tags, encountered-but-discarded limits and bare quoted/template defaults; includes exact null default control.",
+        "interim_review_fix": "65/66 focused passed; old doubling-chain expectation omitted a genuine expression-depth limit now observed. Adjusted that assertion to preserve the byte-limit requirement while allowing other observed classes.",
+        "final_focused": "81 passed in six files; the same CLI/MCP file then passed 9/9 after adding a linked-response case. That follow-up replaces its earlier 8-case coverage; 82 distinct cases across the final six-file selection, not one 82-case invocation.",
+        "final_semantic_suite": "945 passed in 37 files, overlapping focused domain coverage.",
+        "arithmetic": "Computed 174762*6+2=1048574 and 174763*6+2=1048580. Corrected the erroneous published 1048578 total.",
+        "final_grouped_domain": "27 passed after grouping observations/defaults/allocation within function-size rules; no scenarios removed.",
+        "endpoint_review_baseline": "Production at 38d263a: 7 failed/14 passed across 13 new domain and 8 CLI/MCP cases before endpoint normalization. The later public linked-response case was not in this baseline run. Regression tests were uncommitted; final formatted test hashes are preserved, not claimed as the baseline test bytes.",
+        "endpoint_selection_attempt": "First six-file selection exited before execution because a signed-zero test filename was mistyped; corrected the exact path. No test result claimed for that attempt.",
+        "endpoint_typecheck_attempt": "Full static gate found the inferred optional dataField in the public test computed key. Replaced the in-property check with an explicit bare === true discriminant; scenario inputs and runtime implementation unchanged. Corrected static/public checks recorded separately.",
+        "final_gates": "Full static gate passed after the explicit test discriminant; final public file passed 9/9 again, then docs gate passed including 5 consistency tests. Runtime implementation unchanged after the 945-test semantic suite and real control."
+      },
+      "ci_failure": {
+        "run_id": 38758262643,
+        "job_id": 116503076738,
+        "test": "tests/acceptance/applications/javascriptEndpointPayload.test.ts",
+        "result": "1 failed/1573 passed/19 skipped; request endpoint no longer matched its complete source value. Other local tests and real Obsidian control passed but did not establish this compatibility contract.",
+        "aggregate_result": "Merged source/built report: 1 failed/15625 passed/138 skipped. Portable tests and CI required failed as aggregates of the same endpoint assertion; no separate additional failed test."
+      }
+    }
+  ],
+  "review_baseline_revision": "40b9d3c5b9cbeae7661d8e351e1b5ef6e5338931",
+  "review_baseline_source_sha256": {
+    "scripts/verify/javascript/module-references.mjs": "9d5f0b8b35ed2ebb4591f9dc3baea0fc116a0c3fd7d8e5ce30ec476c84671ed1",
+    "src/application/javascript/JavaScriptSemanticGraphDataProjection.ts": "ee541e634264363822bf4ce05c42c72272029c7e8da72bbeed643ad1341d0756",
+    "src/domain/javascript/javascriptSemanticAnalysis.ts": "a1cf791a61ee4742c351c006b69b64eba6bef3068f7a8450ed8f9366bfe4ff88",
+    "src/domain/javascript/javascriptSemanticAnalysis.unions.test.ts": "747d893d6ddd91cf24d3b62b7eb50e0af9b1715ab355bbfc2e10ed0dc171ec17",
+    "src/domain/javascript/javascriptSemanticDataEffectHelpers.ts": "5ee7af18dd847726d8fe1828ea25f386ff9ef317d996a3f7969253bf73973cfa",
+    "src/domain/javascript/javascriptSemanticDataEffects.ts": "4c62462016246bc7198294a74a60e87598fc009c53f1601abc2d4662d41b5bd1",
+    "src/domain/javascript/javascriptSemanticIr.ts": "14955cb4b7912ed6a4f26f804a2480f841dcb6084dbddd15856375fa62cdcfa1",
+    "src/domain/javascript/javascriptSemanticPrimitives.ts": "899eeb6ccdf666f504f8b722327599997c924c8e4fb13a1cbd13edaf08c96640",
+    "src/domain/javascript/javascriptSemanticResourceLimits.ts": "0203d207e0d6bb8aaa17746090f008876a11653afd660fce3dc44f3fda646a6c",
+    "src/domain/javascript/javascriptSemanticState.ts": "d94c5a88bb105284fab192406a8c88860d59f80c93394ed3ee7264969df254a6",
+    "src/domain/javascript/javascriptSemanticValues.ts": "2debbb4f32d7d189819f2413f0f7d05b9bdcf497451a1c578475e3f2ec6c8554",
+    "tests/boundary/mcp/javascriptLiteralBudget.test.ts": "46ceb5a169de3eec3a335b05ef4aa7b1b341a5ac06ad325bd191e862c5b6be70",
+    "tests/boundary/mcp/javascriptLiteralPayload.test.ts": "03618272c605f8877ed171ab39e037e4d80eac72b7478421640076f739adfb50"
+  },
+  "implementation": "Per-analysis state records encountered resource classifications at expression/default normalization; final coverage consumes those facts. Slot projection preserves tagged unknown values. Removed retained-value recursive coverage scan and duplicate raw default reader. Defaults retain optional valueResourceLimit and graph value_resource_limit plus an anchored resource-limit unknown; known null/false/0/empty remain exact. Endpoint identities intentionally remain exact lookup values, outside the normalized primitive lattice; rejected endpoint normalization and its CI failure are preserved separately.",
+  "restored_production_revision": "38d263a369e7853ccfae1c8e0a3bd387f920c7cf",
+  "final_corrected_resource_receipts": {
+    "public_7": {
+      "elapsed_seconds": 43.621,
+      "cpu_seconds": 43.097453,
+      "peak_cgroup_bytes": 1520324608,
+      "payload_cleanup_verified": true,
+      "events_at_payload_exit": {
+        "memory.events": {
+          "low": 0,
+          "high": 0,
+          "max": 0,
+          "oom": 0,
+          "oom_kill": 0,
+          "oom_group_kill": 0
+        },
+        "pids.events": {
+          "max": 0
+        }
+      }
+    },
+    "static_and_docs": {
+      "elapsed_seconds": 111.403,
+      "cpu_seconds": 110.830839,
+      "peak_cgroup_bytes": 4268244992,
+      "payload_cleanup_verified": true,
+      "events_at_payload_exit": {
+        "memory.events": {
+          "low": 0,
+          "high": 0,
+          "max": 0,
+          "oom": 0,
+          "oom_kill": 0,
+          "oom_group_kill": 0
+        },
+        "pids.events": {
+          "max": 0
+        }
+      }
+    }
+  }
+}

docs/verification/javascript-literal-byte-budget-2026-10-15.md

@@ -0,0 +1,86 @@
+# Semantic literal byte-budget verification
+
+Tracks #3933 under #4027. All normalized semantic literals and configuration
+defaults use the primitive owner. Its singleton path validates strings/nonfinite
+numbers/negative zero without a Set, sorting or canonical-key serialization.
+Each source analysis owns a bounded set of encountered limit classifications;
+coverage consumes these observations after evaluation. This replaces a recursive
+scan that only saw retained binding/return values. No persistent cache is added.
+
+## Correctness and coverage
+
+Public CLI and actual stdio MCP cases cover quoted strings, cooked templates,
+equivalent concatenation, member receivers, a configuration-only expression
+statement and a small exact control. Direct/default oversized values become
+`resource-limit` unknowns with partial application coverage and
+`javascript_semantic_primitive_bytes = 1048576`. Defaults carry
+`value_resource_limit` inline and an unknown anchored to their config node.
+Known null/false/zero/empty defaults remain exact. Original source/digests and
+CLI/MCP identity parity are preserved.
+
+Member and destructuring projections use the shared slot owner's tagged value.
+An encountered limit stays visible even when a subsequent projection selects a
+healthy sibling: the selected value remains exactly `7`, while file coverage is
+partial. This avoids inferring completeness from the final retained values alone.
+The duplicate raw default reader and superseded retained-value scan are removed.
+
+Boundary controls admit 174,762 UTF-16 units (1,048,574 estimated JSON bytes)
+and reject 174,763 (1,048,580 bytes). Actual arithmetic corrected the earlier
+published 1,048,578 total; the unit-count boundary assertions were already correct.
+The estimate counts six bytes per unit plus two quotes. It does not establish
+actual ASCII JSON size above 1 MiB or an observed OOM.
+
+The initial main baseline reproduced both public literal failures, with addition
+and small controls passing. A mistaken optional-limit-array control was corrected
+before the separate domain reproduction (2 failed/16 passed). Review head
+`40b9d3c5` then reproduced the additional projection/default gaps (8 failed,
+25 passed). The first review correction passed 65/66 focused tests; a doubling
+chain now truthfully retains an additional encountered expression-depth limit,
+so its byte-growth assertion was updated to allow other recorded limits.
+Prior validation on the restored runtime: **66 focused tests** in five files and **932 semantic tests**
+in 36 files, with overlapping domain coverage. The large-wire fixture retains
+16 individually admitted values and at least the former aggregate UTF-16 payload,
+plus complete CLI/MCP delivery, typed identities, empty strings and retained
+tracing. Inputs regenerate from the tests.
+
+## Endpoint identity compatibility
+
+An attempted endpoint normalization at `5808872` passed its new local cases and real Obsidian control, but failed the existing CI endpoint-payload acceptance workflow (1 failed/1,573 passed/19 skipped in that shard). It violated the established full endpoint lookup/trace contract. This extension was reverted; endpoint values are lookup identities, outside the normalized primitive lattice. The original large endpoint is preserved. Corrected public validation passed **7/7** across two files; this adds one acceptance workflow to the previous 66-case focused selection (**67 distinct cases**, not one 67-case invocation). Its acceptance workflow now also normalizes that identical string as an exported primitive, requiring the typed value bound while endpoint lookup, both trace workflows and version comparison remain usable. Prior measurements and failed attempts remain in the JSON record. Production source is restored byte-identical to `38d263a`; the 932-test semantic run belongs to that state. Final full static and documentation gates (five consistency tests) passed after the compatibility correction.
+
+## Real application control and costs
+
+Official Obsidian 1.14.4 ASAR, selected `package.json` and `main.js`. Follow the
+acquisition/digest and existing verifier command in
+[module-reference verification](javascript-module-references-2026-10-15.md).
+Use three alternating-order query pairs with CLI parity on each revision.
+Application code is not executed.
+
+The final correction retains the exact baseline full parent/view Evidence IDs,
+23-row fact digest, source coverage and unknowns. Static `cl100k_base` task costs
+remain 16,548 with selected name/description/inputSchema contracts, or 204,093
+with the full catalog projection, for the reference workflow. Serial legacy
+reads also keep their original costs. The query projection improvement belongs
+to #4097; these are controls for semantic normalization and coverage.
+
+| Complete real verifier |        Before | Final correction |
+| ---------------------- | ------------: | ---------------: |
+| Elapsed seconds        |        33.394 |           32.949 |
+| CPU seconds            |     33.016802 |        32.559187 |
+| Peak cgroup bytes      | 1,979,863,040 |    1,880,096,768 |
+
+Includes initial analysis, both query arms, export, CLI parity and tokenizer.
+One CPU, 8 GiB hard cap, 6 GiB early stop, no swap; cleanup verified and no OOM.
+Memory includes charged file cache, not summed RSS. Counters are at payload exit
+while the observer remains alive. Fresh MCP sessions do not establish OS-cold
+cache conditions. Shared-host scheduling and filesystem/build warmness were
+uncontrolled. This one before/after control establishes neither a speedup nor
+an equivalence bound. Static BPE excludes annotations/output schemas, client
+rendering/replay, reasoning and final answers; no hosted agent was run. Linux
+static ASAR analysis was exercised, not live Electron or other platforms.
+
+[Compact records, samples, source hashes and prior attempts](javascript-literal-byte-budget-2026-10-15.json)
+preserve the initial incomplete correction separately from the final measurements.
+No downloadable application or complete bundle is archived. Reproduce boundary
+outcomes through the listed public tests and reuse the existing real verifier
+and preserved 8 GiB observer linked above. Observer hashes and result-affecting
+touched source hashes are pinned in the records.

docs/verification/javascript-module-references-2026-10-15.json

@@ -0,0 +1,691 @@
+{
+  "base_revision": "c9f9b83c6f5878f1d5b102ab206a2ce214749258",
+  "source_sha256": {
+    "scripts/verify/javascript/module-references.mjs": "9d5f0b8b35ed2ebb4591f9dc3baea0fc116a0c3fd7d8e5ce30ec476c84671ed1",
+    "src/contracts/analysisViewToolContracts.ts": "f7c64b3472e7ec1090f9d47ede864b8c9630d506d896454a800747f2c90a8960",
+    "src/domain/analysisView/analysisView.ts": "854b03988791c19642b01d17813b379b51d8edc10d2190a3c236c9a1b2feb365",
+    "src/domain/analysisView/javascriptApplicationView.ts": "f2c41ad15fdc9f891412c9161a832b60baf3dae55e3c4d60159098d07f7d1ef1",
+    "src/domain/analysisView/javascriptModuleReferences.ts": "34c7a7ff466e9c49d43f1b11b72f878041de61a1bc09086ec190a5833b454806",
+    "scripts/fixtures/tokenizer-family-wrapper.py": "2e9b7092ee539ae1c6a477492f05bb6007d4612dadfff9b90051d621de43049b"
+  },
+  "verification": {
+    "scope": "Completed module-fact reads from identical retained Evidence; no hosted-model performance claim",
+    "artifact": {
+      "path": "/tmp/rea-codemode-experiment-20261015/artifacts/obsidian-1.14.4.asar",
+      "sha256": "146ef8470d8cbb4a6db209bdc19bf2bad13350444186c0fd93c31087b142b92f"
+    },
+    "parent_evidence_id": "ev_73ef06e3b9b944f54c5257f7960279127d8518d2a73cd8618bd045dbe44e6b5a",
+    "source_selection": {
+      "requested_paths": ["package.json", "main.js"],
+      "unselected_files": 36
+    },
+    "source_coverage": {
+      "application": {
+        "status": "partial",
+        "truncated": false,
+        "omitted_count": null,
+        "limits": []
+      },
+      "semantic": {
+        "status": "partial",
+        "truncated": false,
+        "omitted_nodes": null,
+        "omitted_relations": null,
+        "limits": [],
+        "families": [
+          {
+            "family": "boundary",
+            "status": "partial",
+            "retained_relations": 14,
+            "omitted_relations": null,
+            "unknown_count": 31
+          },
+          {
+            "family": "call-flow",
+            "status": "partial",
+            "retained_relations": 845,
+            "omitted_relations": null,
+            "unknown_count": 1278
+          },
+          {
+            "family": "child-process",
+            "status": "partial",
+            "retained_relations": 0,
+            "omitted_relations": null,
+            "unknown_count": 31
+          },
+          {
+            "family": "closure",
+            "status": "partial",
+            "retained_relations": 1187,
+            "omitted_relations": null,
+            "unknown_count": 31
+          },
+          {
+            "family": "configuration",
+            "status": "partial",
+            "retained_relations": 0,
+            "omitted_relations": null,
+            "unknown_count": 31
+          },
+          {
+            "family": "data-flow",
+            "status": "partial",
+            "retained_relations": 9388,
+            "omitted_relations": null,
+            "unknown_count": 1135
+          },
+          {
+            "family": "event",
+            "status": "partial",
+            "retained_relations": 127,
+            "omitted_relations": null,
+            "unknown_count": 89
+          },
+          {
+            "family": "object-flow",
+            "status": "partial",
+            "retained_relations": 1462,
+            "omitted_relations": null,
+            "unknown_count": 2292
+          },
+          {
+            "family": "promise-ownership",
+            "status": "partial",
+            "retained_relations": 80,
+            "omitted_relations": null,
+            "unknown_count": 98
+          },
+          {
+            "family": "request",
+            "status": "partial",
+            "retained_relations": 0,
+            "omitted_relations": null,
+            "unknown_count": 31
+          },
+          {
+            "family": "resource-lifecycle",
+            "status": "partial",
+            "retained_relations": 4,
+            "omitted_relations": null,
+            "unknown_count": 31
+          },
+          {
+            "family": "timer",
+            "status": "partial",
+            "retained_relations": 8,
+            "omitted_relations": null,
+            "unknown_count": 31
+          }
+        ]
+      },
+      "application_unknowns": 36,
+      "semantic_unknowns": 5109
+    },
+    "where": {
+      "path": "main.js",
+      "kind": "javascript-module"
+    },
+    "preparation_ms": 20978.249922,
+    "samples": [
+      {
+        "round": 0,
+        "mode": "legacy",
+        "elapsed_ms": 363.02500200000213,
+        "calls": 24,
+        "structured_reply_bytes": 325201,
+        "facts_sha256": "d3be5e91d2ec3fd9a95851cfbbdef9c168bb4c7153f78810b8b235e005401299",
+        "matching_modules": 23,
+        "coverage": {
+          "status": "page",
+          "examined": 55,
+          "total": 55,
+          "next_offset": null,
+          "exhausted": true
+        }
+      },
+      {
+        "round": 0,
+        "mode": "references",
+        "elapsed_ms": 14.895039999999426,
+        "calls": 1,
+        "structured_reply_bytes": 22158,
+        "facts_sha256": "d3be5e91d2ec3fd9a95851cfbbdef9c168bb4c7153f78810b8b235e005401299",
+        "matching_modules": 23,
+        "coverage": {
+          "status": "page",
+          "examined": 23,
+          "total": 23,
+          "next_offset": null,
+          "exhausted": true
+        }
+      },
+      {
+        "round": 1,
+        "mode": "references",
+        "elapsed_ms": 12.951290000000881,
+        "calls": 1,
+        "structured_reply_bytes": 22158,
+        "facts_sha256": "d3be5e91d2ec3fd9a95851cfbbdef9c168bb4c7153f78810b8b235e005401299",
+        "matching_modules": 23,
+        "coverage": {
+          "status": "page",
+          "examined": 23,
+          "total": 23,
+          "next_offset": null,
+          "exhausted": true
+        }
+      },
+      {
+        "round": 1,
+        "mode": "legacy",
+        "elapsed_ms": 232.7533779999976,
+        "calls": 24,
+        "structured_reply_bytes": 325201,
+        "facts_sha256": "d3be5e91d2ec3fd9a95851cfbbdef9c168bb4c7153f78810b8b235e005401299",
+        "matching_modules": 23,
+        "coverage": {
+          "status": "page",
+          "examined": 55,
+          "total": 55,
+          "next_offset": null,
+          "exhausted": true
+        }
+      },
+      {
+        "round": 2,
+        "mode": "legacy",
+        "elapsed_ms": 235.16672899999685,
+        "calls": 24,
+        "structured_reply_bytes": 325201,
+        "facts_sha256": "d3be5e91d2ec3fd9a95851cfbbdef9c168bb4c7153f78810b8b235e005401299",
+        "matching_modules": 23,
+        "coverage": {
+          "status": "page",
+          "examined": 55,
+          "total": 55,
+          "next_offset": null,
+          "exhausted": true
+        }
+      },
+      {
+        "round": 2,
+        "mode": "references",
+        "elapsed_ms": 10.566158999998152,
+        "calls": 1,
+        "structured_reply_bytes": 22158,
+        "facts_sha256": "d3be5e91d2ec3fd9a95851cfbbdef9c168bb4c7153f78810b8b235e005401299",
+        "matching_modules": 23,
+        "coverage": {
+          "status": "page",
+          "examined": 23,
+          "total": 23,
+          "next_offset": null,
+          "exhausted": true
+        }
+      }
+    ],
+    "token_costs": {
+      "tokenizer": "cl100k_base",
+      "tiktoken_version": "0.12.0",
+      "costs": {
+        "legacy": {
+          "selected_catalog": 7977,
+          "full_catalog": 195446,
+          "common_request": 53,
+          "common_summary": 2675,
+          "requests": 2836,
+          "structured_replies": 71761,
+          "selected_catalog_task_total": 85302,
+          "full_catalog_task_total": 272771
+        },
+        "references": {
+          "selected_catalog": 7977,
+          "full_catalog": 195446,
+          "common_request": 53,
+          "common_summary": 2675,
+          "requests": 102,
+          "structured_replies": 5741,
+          "selected_catalog_task_total": 16548,
+          "full_catalog_task_total": 204017
+        }
+      }
+    },
+    "cli_parity": {
+      "evidence_id": "ev_b3586dfdf7d4f01157f8f2978326aaa7103987110f2df4fa5bf3331ff2898749",
+      "facts_sha256": "d3be5e91d2ec3fd9a95851cfbbdef9c168bb4c7153f78810b8b235e005401299",
+      "exact_cli_mcp_view_parity": true
+    },
+    "conditions": "First query is after shared artifact analysis; subsequent queries share one MCP session. Legacy item reads are serial; a concurrent legacy baseline is unmeasured. Filesystem/build warmness uncontrolled. Token costs count one post-change catalog, preparation request/summary and each request/structured reply once; client rendering, replay, reasoning, final answer and model latency are unmeasured. Preparation and parity/tokenization overhead are outside per-query elapsed times; external family-resource receipts include the complete verifier and children."
+  },
+  "resources": {
+    "scope": "Complete verifier, MCP/analysis descendants, temporary export, CLI parity and tokenizer; cgroup memory includes charged file cache and is not summed RSS",
+    "elapsed_seconds": 35.797,
+    "cpu_seconds": 35.197074,
+    "peak_cgroup_bytes": 1938313216,
+    "memory_max_bytes": 8589934592,
+    "memory_early_stop_bytes": 6442450944,
+    "cpu_quota_percent": 100,
+    "payload_exit_code": 0,
+    "payload_cleanup_verified": true,
+    "events_at_payload_exit": {
+      "memory.events": {
+        "low": 0,
+        "high": 0,
+        "max": 0,
+        "oom": 0,
+        "oom_kill": 0,
+        "oom_group_kill": 0
+      },
+      "pids.events": {
+        "max": 0
+      }
+    },
+    "observer_sources_sha256": {
+      "resource-run.py": "7728862171415e7dee2c11af2ead2b9a54265fccaf82a7113956d36b29065e33",
+      "resource_payload.py": "6f089e645459eba80c5d73e39c21606c78173a44daf49fabc3b88c4631ce51c6"
+    }
+  },
+  "prior_attempts": [
+    {
+      "scope": "Initial verifier incorrectly requested full MCP delivery; completed analysis exceeded existing response admission",
+      "elapsed_seconds": 21.971,
+      "cpu_seconds": 21.78546,
+      "peak_cgroup_bytes": 1080860672,
+      "outcome": "resource_constraint transport; verifier corrected to summary-first; no performance credit"
+    },
+    {
+      "scope": "Interim successful parity run before eliminating verifier assertion serialization and counting the bootstrap request",
+      "elapsed_seconds": 31.335,
+      "cpu_seconds": 31.07044,
+      "peak_cgroup_bytes": 1652293632,
+      "outcome": "parity passed; superseded measurement, not pooled with final samples"
+    },
+    {
+      "scope": "Successful three-pair probe before adding the shared analyze_javascript_application schema to selected-catalog accounting",
+      "preparation_ms": 18317.110148,
+      "samples": [
+        {
+          "round": 0,
+          "mode": "legacy",
+          "elapsed_ms": 313.59318599999824,
+          "calls": 24,
+          "structured_reply_bytes": 325201,
+          "facts_sha256": "d3be5e91d2ec3fd9a95851cfbbdef9c168bb4c7153f78810b8b235e005401299",
+          "matching_modules": 23,
+          "coverage": {
+            "status": "page",
+            "examined": 55,
+            "total": 55,
+            "next_offset": null,
+            "exhausted": true
+          }
+        },
+        {
+          "round": 0,
+          "mode": "references",
+          "elapsed_ms": 13.595478999999614,
+          "calls": 1,
+          "structured_reply_bytes": 22158,
+          "facts_sha256": "d3be5e91d2ec3fd9a95851cfbbdef9c168bb4c7153f78810b8b235e005401299",
+          "matching_modules": 23,
+          "coverage": {
+            "status": "page",
+            "examined": 23,
+            "total": 23,
+            "next_offset": null,
+            "exhausted": true
+          }
+        },
+        {
+          "round": 1,
+          "mode": "references",
+          "elapsed_ms": 11.230690000000322,
+          "calls": 1,
+          "structured_reply_bytes": 22158,
+          "facts_sha256": "d3be5e91d2ec3fd9a95851cfbbdef9c168bb4c7153f78810b8b235e005401299",
+          "matching_modules": 23,
+          "coverage": {
+            "status": "page",
+            "examined": 23,
+            "total": 23,
+            "next_offset": null,
+            "exhausted": true
+          }
+        },
+        {
+          "round": 1,
+          "mode": "legacy",
+          "elapsed_ms": 206.19418099999893,
+          "calls": 24,
+          "structured_reply_bytes": 325201,
+          "facts_sha256": "d3be5e91d2ec3fd9a95851cfbbdef9c168bb4c7153f78810b8b235e005401299",
+          "matching_modules": 23,
+          "coverage": {
+            "status": "page",
+            "examined": 55,
+            "total": 55,
+            "next_offset": null,
+            "exhausted": true
+          }
+        },
+        {
+          "round": 2,
+          "mode": "legacy",
+          "elapsed_ms": 242.2037199999977,
+          "calls": 24,
+          "structured_reply_bytes": 325201,
+          "facts_sha256": "d3be5e91d2ec3fd9a95851cfbbdef9c168bb4c7153f78810b8b235e005401299",
+          "matching_modules": 23,
+          "coverage": {
+            "status": "page",
+            "examined": 55,
+            "total": 55,
+            "next_offset": null,
+            "exhausted": true
+          }
+        },
+        {
+          "round": 2,
+          "mode": "references",
+          "elapsed_ms": 13.734700000000885,
+          "calls": 1,
+          "structured_reply_bytes": 22158,
+          "facts_sha256": "d3be5e91d2ec3fd9a95851cfbbdef9c168bb4c7153f78810b8b235e005401299",
+          "matching_modules": 23,
+          "coverage": {
+            "status": "page",
+            "examined": 23,
+            "total": 23,
+            "next_offset": null,
+            "exhausted": true
+          }
+        }
+      ],
+      "elapsed_seconds": 31.037,
+      "cpu_seconds": 30.786998,
+      "peak_cgroup_bytes": 1886543872,
+      "outcome": "Facts and CLI/MCP parity passed. Selected-catalog totals omitted the common analysis-tool schema; superseded, not pooled. Shared-host timing differed substantially from the final run."
+    },
+    {
+      "scope": "Completed three-pair probe with full token accounting before the review-added tokenizer deadline",
+      "preparation_ms": 66696.731143,
+      "samples": [
+        {
+          "round": 0,
+          "mode": "legacy",
+          "elapsed_ms": 1114.080965000001,
+          "calls": 24,
+          "structured_reply_bytes": 325201,
+          "facts_sha256": "d3be5e91d2ec3fd9a95851cfbbdef9c168bb4c7153f78810b8b235e005401299",
+          "matching_modules": 23,
+          "coverage": {
+            "status": "page",
+            "examined": 55,
+            "total": 55,
+            "next_offset": null,
+            "exhausted": true
+          }
+        },
+        {
+          "round": 0,
+          "mode": "references",
+          "elapsed_ms": 46.572758999987855,
+          "calls": 1,
+          "structured_reply_bytes": 22158,
+          "facts_sha256": "d3be5e91d2ec3fd9a95851cfbbdef9c168bb4c7153f78810b8b235e005401299",
+          "matching_modules": 23,
+          "coverage": {
+            "status": "page",
+            "examined": 23,
+            "total": 23,
+            "next_offset": null,
+            "exhausted": true
+          }
+        },
+        {
+          "round": 1,
+          "mode": "references",
+          "elapsed_ms": 37.09324900001229,
+          "calls": 1,
+          "structured_reply_bytes": 22158,
+          "facts_sha256": "d3be5e91d2ec3fd9a95851cfbbdef9c168bb4c7153f78810b8b235e005401299",
+          "matching_modules": 23,
+          "coverage": {
+            "status": "page",
+            "examined": 23,
+            "total": 23,
+            "next_offset": null,
+            "exhausted": true
+          }
+        },
+        {
+          "round": 1,
+          "mode": "legacy",
+          "elapsed_ms": 675.7124039999908,
+          "calls": 24,
+          "structured_reply_bytes": 325201,
+          "facts_sha256": "d3be5e91d2ec3fd9a95851cfbbdef9c168bb4c7153f78810b8b235e005401299",
+          "matching_modules": 23,
+          "coverage": {
+            "status": "page",
+            "examined": 55,
+            "total": 55,
+            "next_offset": null,
+            "exhausted": true
+          }
+        },
+        {
+          "round": 2,
+          "mode": "legacy",
+          "elapsed_ms": 777.2660329999926,
+          "calls": 24,
+          "structured_reply_bytes": 325201,
+          "facts_sha256": "d3be5e91d2ec3fd9a95851cfbbdef9c168bb4c7153f78810b8b235e005401299",
+          "matching_modules": 23,
+          "coverage": {
+            "status": "page",
+            "examined": 55,
+            "total": 55,
+            "next_offset": null,
+            "exhausted": true
+          }
+        },
+        {
+          "round": 2,
+          "mode": "references",
+          "elapsed_ms": 36.31178899999941,
+          "calls": 1,
+          "structured_reply_bytes": 22158,
+          "facts_sha256": "d3be5e91d2ec3fd9a95851cfbbdef9c168bb4c7153f78810b8b235e005401299",
+          "matching_modules": 23,
+          "coverage": {
+            "status": "page",
+            "examined": 23,
+            "total": 23,
+            "next_offset": null,
+            "exhausted": true
+          }
+        }
+      ],
+      "elapsed_seconds": 108.372,
+      "cpu_seconds": 81.551676,
+      "peak_cgroup_bytes": 1884962816,
+      "outcome": "All facts/costs and CLI/MCP parity passed. Superseded execution source after adding a tokenizer deadline; retained separately, not pooled."
+    },
+    {
+      "scope": "Successful data comparison with direct-child tokenizer timeout before adopting shared owned-process supervision",
+      "preparation_ms": 19051.368892000002,
+      "samples": [
+        {
+          "round": 0,
+          "mode": "legacy",
+          "elapsed_ms": 330.7313439999998,
+          "calls": 24,
+          "structured_reply_bytes": 325201,
+          "facts_sha256": "d3be5e91d2ec3fd9a95851cfbbdef9c168bb4c7153f78810b8b235e005401299",
+          "matching_modules": 23,
+          "coverage": {
+            "status": "page",
+            "examined": 55,
+            "total": 55,
+            "next_offset": null,
+            "exhausted": true
+          }
+        },
+        {
+          "round": 0,
+          "mode": "references",
+          "elapsed_ms": 13.62184999999954,
+          "calls": 1,
+          "structured_reply_bytes": 22158,
+          "facts_sha256": "d3be5e91d2ec3fd9a95851cfbbdef9c168bb4c7153f78810b8b235e005401299",
+          "matching_modules": 23,
+          "coverage": {
+            "status": "page",
+            "examined": 23,
+            "total": 23,
+            "next_offset": null,
+            "exhausted": true
+          }
+        },
+        {
+          "round": 1,
+          "mode": "references",
+          "elapsed_ms": 12.475999999998749,
+          "calls": 1,
+          "structured_reply_bytes": 22158,
+          "facts_sha256": "d3be5e91d2ec3fd9a95851cfbbdef9c168bb4c7153f78810b8b235e005401299",
+          "matching_modules": 23,
+          "coverage": {
+            "status": "page",
+            "examined": 23,
+            "total": 23,
+            "next_offset": null,
+            "exhausted": true
+          }
+        },
+        {
+          "round": 1,
+          "mode": "legacy",
+          "elapsed_ms": 187.18667300000016,
+          "calls": 24,
+          "structured_reply_bytes": 325201,
+          "facts_sha256": "d3be5e91d2ec3fd9a95851cfbbdef9c168bb4c7153f78810b8b235e005401299",
+          "matching_modules": 23,
+          "coverage": {
+            "status": "page",
+            "examined": 55,
+            "total": 55,
+            "next_offset": null,
+            "exhausted": true
+          }
+        },
+        {
+          "round": 2,
+          "mode": "legacy",
+          "elapsed_ms": 293.4882039999975,
+          "calls": 24,
+          "structured_reply_bytes": 325201,
+          "facts_sha256": "d3be5e91d2ec3fd9a95851cfbbdef9c168bb4c7153f78810b8b235e005401299",
+          "matching_modules": 23,
+          "coverage": {
+            "status": "page",
+            "examined": 55,
+            "total": 55,
+            "next_offset": null,
+            "exhausted": true
+          }
+        },
+        {
+          "round": 2,
+          "mode": "references",
+          "elapsed_ms": 8.146389999998064,
+          "calls": 1,
+          "structured_reply_bytes": 22158,
+          "facts_sha256": "d3be5e91d2ec3fd9a95851cfbbdef9c168bb4c7153f78810b8b235e005401299",
+          "matching_modules": 23,
+          "coverage": {
+            "status": "page",
+            "examined": 23,
+            "total": 23,
+            "next_offset": null,
+            "exhausted": true
+          }
+        }
+      ],
+      "elapsed_seconds": 30.742,
+      "cpu_seconds": 30.425218,
+      "peak_cgroup_bytes": 1592938496,
+      "outcome": "Fact and cost parity passed. That harness only killed its immediate tokenizer child; superseded by process-family supervision and detached-helper verification."
+    }
+  ],
+  "prior_tokenizer_timeout_verifications": [
+    {
+      "artifact_sha256": "146ef8470d8cbb4a6db209bdc19bf2bad13350444186c0fd93c31087b142b92f",
+      "scenario": "Real Obsidian references plus CLI parity, then a deliberately stalled tokenizer",
+      "injected_tokenizer_source": "#!/bin/sh\nexec /bin/sleep 30\n",
+      "timeout_ms": 20,
+      "observed_error_code": "ETIMEDOUT",
+      "expected_payload_exit_code": 1,
+      "payload_cleanup_verified": true,
+      "remaining_owned_temporary_files": 0,
+      "elapsed_seconds": 31.258,
+      "cpu_seconds": 31.00025,
+      "peak_cgroup_bytes": 1644077056,
+      "events_at_payload_exit": {
+        "memory.events": {
+          "low": 0,
+          "high": 0,
+          "max": 0,
+          "oom": 0,
+          "oom_kill": 0,
+          "oom_group_kill": 0
+        },
+        "pids.events": {
+          "max": 0
+        }
+      },
+      "scope": "Robustness check; no efficiency credit",
+      "limitation": "Immediate child only; did not test detached descendants and is superseded by the owned-family check."
+    }
+  ],
+  "tokenizer_timeout_verification": {
+    "artifact_sha256": "146ef8470d8cbb4a6db209bdc19bf2bad13350444186c0fd93c31087b142b92f",
+    "scenario": "Real Obsidian references and CLI parity, then an owned tokenizer wrapper with a detached SIGTERM-ignoring helper",
+    "fixture": "scripts/fixtures/tokenizer-family-wrapper.py",
+    "fixture_sha256": "2e9b7092ee539ae1c6a477492f05bb6007d4612dadfff9b90051d621de43049b",
+    "timeout_ms": 2000,
+    "observed_failure_type": "OwnedCommandFailure",
+    "observed_reason": "timeout",
+    "cleanup_failure": null,
+    "expected_payload_exit_code": 1,
+    "payload_cleanup_verified": true,
+    "remaining_owned_temporary_files": 0,
+    "helper_observation": {
+      "pid": 2871108,
+      "group": 2871108,
+      "parent": 2871107,
+      "run_id": "rea-module-token-f139b74a-dd83-498b-b663-e765104159ec",
+      "detached": true,
+      "ignores_sigterm": true
+    },
+    "elapsed_seconds": 55.08,
+    "cpu_seconds": 45.245603,
+    "peak_cgroup_bytes": 1563475968,
+    "events_at_payload_exit": {
+      "memory.events": {
+        "low": 0,
+        "high": 0,
+        "max": 0,
+        "oom": 0,
+        "oom_kill": 0,
+        "oom_group_kill": 0
+      },
+      "pids.events": {
+        "max": 0
+      }
+    },
+    "scope": "Robustness check; no efficiency credit. PIDs are observations; signaling authority comes from the shared process owner."
+  }
+}

docs/verification/javascript-module-references-2026-10-15.md

@@ -0,0 +1,123 @@
+# Retained JavaScript module-reference verification
+
+Tracks #4092 under #4027 and the decision in #4091. This verifies an existing
+tool's data projection, without a code-mode host or SDK. [Machine-readable
+results](javascript-module-references-2026-10-15.json) pin the base revision,
+executed source hashes, samples, tokenizer and resource observations.
+
+## Real application and question
+
+Official Obsidian 1.14.4 ASAR, 27,477,966 bytes, SHA-256
+`146ef8470d8cbb4a6db209bdc19bf2bad13350444186c0fd93c31087b142b92f`.
+The [official gzip](https://github.com/obsidianmd/obsidian-releases/releases/download/v1.14.4/obsidian-1.14.4.asar.gz)
+has SHA-256 `d1ed428c363968774f0f3906e67d8a53a058ece0b6cdd924b07865977b2ada21`.
+Select `package.json` and `main.js`; the other 36 inventoried files remain
+explicitly unselected. No target code is executed.
+
+Question: retrieve every retained `javascript-module` identity at exact path
+`main.js`, its retained source-range observation count, and distinct support
+states. Compare the existing module page plus full item reads with one filtered
+typed reference view. All six completed query samples use one identical retained
+parent. All 23 selected rows, identities, counts and support states match. Public
+CLI inline input and MCP retained input produce the exact same view and Evidence
+ID. This establishes producer-fact consistency, not independent extractor
+correctness or proof of runtime loading.
+
+## Completed read costs
+
+| Measurement                                                   | Existing page + items | Typed reference view |
+| ------------------------------------------------------------- | --------------------: | -------------------: |
+| Calls per query task                                          |                    24 |                    1 |
+| Structured reply bytes                                        |               325,201 |               22,158 |
+| First query in its arm, after preparation                     |             363.03 ms |             14.90 ms |
+| Subsequent same-session query times                           |    232.75 / 235.17 ms |     12.95 / 10.57 ms |
+| Request + structured-reply BPE tokens                         |                74,597 |                5,843 |
+| With selected tool catalog + shared bootstrap request/summary |                85,302 |               16,548 |
+| With complete tool catalog + shared bootstrap request/summary |               272,771 |              204,017 |
+
+Order alternates: existing/reference, reference/existing, existing/reference.
+Times include verifier-side fact normalization and equality checks; byte/token
+accounting follows the timed reads. The initial artifact analysis and summary
+took 20.978 seconds and is shared preparation, not a query-speed improvement.
+The retained source remains partial; unknown counts and source selection are
+preserved independently of exhausted pages. No query cache was introduced.
+
+Tokens are `cl100k_base` counts with tiktoken 0.12.0 over serialized values, not
+hosted-model usage. Catalog projections contain name/description/inputSchema;
+both workflows use the same post-change catalog. Each request/reply is counted
+once. Output schemas, host-specific rendering/replay, reasoning, final answers,
+agent success rates and model latency are unmeasured. This is a completed
+data-retrieval comparison; it does not establish a whole-agent speedup or a
+before/after catalog-size improvement. Legacy item reads are serial; a concurrent
+legacy baseline was not measured. Selected discovery includes both the
+analysis and view tool schemas. Builds/filesystem are warm or uncontrolled,
+so the first query is not an OS-cold analysis measurement. Shared-host scheduling
+was uncontrolled: an earlier successful probe had substantially higher wall/CPU
+times. Its samples are retained separately; neither run establishes a portable
+latency guarantee.
+
+The complete final verifier scope used 35.797 seconds elapsed, 35.197 seconds
+CPU and 1,938,313,216 bytes peak cgroup memory. This includes shared analysis,
+both query arms, temporary export, CLI parity and tokenization; CPU/memory were
+not isolated by arm. Cgroup memory includes charged file cache and is not summed
+process RSS. The whole scope had one CPU, an 8 GiB hard memory cap, 6 GiB early
+stop and no swap. Payload cleanup was verified; no OOM/capacity event occurred.
+An earlier full-response transport failure and interim probes are recorded
+separately in the JSON and earn no performance credit.
+
+A review follow-up exercised a stalled tokenizer after real Obsidian reference
+reads and CLI parity. The committed POSIX fault fixture forks a helper into a
+separate process group, ignores SIGTERM and stalls. With a caller-selected
+2,000 ms deadline, the shared owned-command supervisor reported
+`OwnedCommandFailure` with reason `timeout` and no cleanup failure. The verifier
+failed as expected, its complete owned scope was cleaned, and its selected
+temporary root was empty. The separate robustness run used 55.080 s wall /
+45.246 s CPU / 1,563,475,968 B peak cgroup memory, with cleanup verified and no
+OOM. It earns no efficiency credit. An earlier 20 ms immediate-child timeout
+probe did not establish descendant cleanup; it remains a separate prior attempt.
+The detached-helper scenario was executed on Linux, not Windows or macOS.
+
+A deterministic CI regression also asserts two retained
+observations at the same source range, excludes an unavailable location, and
+checks agreement with the full item view; the count deliberately measures
+observations rather than unique source sites.
+
+## Reproduce
+
+Use the pinned Node/npm toolchain and `npm ci`, then `npm run build:cached`.
+Download/decompress the versioned application outside the checkout and check its
+digest. Run with an operator-selected absolute artifact path:
+
+```bash
+node scripts/verify/javascript/module-references.mjs \
+  --artifact /absolute/path/obsidian-1.14.4.asar \
+  --source-path package.json --source-path main.js \
+  --path main.js --kind javascript-module --rounds 3 --cli-parity \
+  --tokenizer-python /absolute/path/to/tiktoken-venv/bin/python
+```
+
+The optional Python interpreter needs tiktoken 0.12.0; omit that flag to skip
+token accounting. It has a 60-second wall-time deadline, configurable with the
+positive integer `--tokenizer-timeout-ms`. The tokenizer uses the existing
+owned-command supervisor, with native identity checks and inherited run-token
+sweeping for descendants. Inputs and temporary Evidence use the existing private
+runtime root. Cleanup attempts preserve the primary error; incomplete command
+cleanup retains its owner and private files for recovery. `--mode legacy` or
+`--mode references` runs one arm. Fresh runs
+reanalyze the artifact and can produce different parent identities/resource
+coverage; the verifier requires equal facts within its paired run. The measured
+run additionally used the preserved [8 GiB cgroup observer](https://github.com/morluto/rea/blob/b3bdabab694af1f9801eff7caf75001b92314a58/experiments/analysis-efficiency/resource-run.py)
+and its adjacent `resource_payload.py`; observer hashes are in the JSON.
+No application, complete bundle, raw observations or ambient environment is
+archived. Temporary exported/inline Evidence files are removed after successful
+owned-command cleanup.
+
+For the Linux descendant-timeout check, create an operator-selected private
+directory and set `REA_TOKENIZER_PROBE_RECORD` to a file within it. Run the same
+real-artifact command with `--mode references --rounds 1`, replace the tokenizer
+with the absolute path to `scripts/fixtures/tokenizer-family-wrapper.py`, and add
+`--tokenizer-timeout-ms 2000`. Expect a timeout failure. The explicit marker
+confirms the detached helper started; the resource observer must separately
+verify the complete owned scope is empty, and the selected temporary root must
+contain no verifier files. The recorded PID is an observation, not authority to
+signal that process.

docs/verification/javascript-native-handoff-identity-2026-10-15.md

@@ -0,0 +1,179 @@
+# Native handoff byte-identity verification
+
+Tracks #3926 under #4027. Baseline: `9f8659e0536e4e50a438f5b9d5661f3c9624f6ac`.
+Final integration base: `62bf09809ef7bb224028916424f8f07652a20f4f`.
+
+The analyst already knows an ASAR-relative native path and supplies prior native
+Evidence. The question is which Evidence can be linked to that component's exact
+bytes. Discovery and native-provider execution are outside this task.
+
+## Observed behavior and correction
+
+The native-node producer distinguishes a content digest from a canonical path
+whose bytes are unavailable. Its unavailable observation legitimately cites the
+containing archive. The handoff consumer previously treated that supporting
+artifact digest as the native component's digest.
+
+Public CLI/stdio MCP regression: the missing-companion case failed and the
+available-byte control passed. Real Joplin reproduced the same result: missing
+keytar bytes produced an `evidence-linked` handoff with the ASAR digest and the
+simulated container observation. CLI/MCP agreed on that incorrect answer.
+
+The consumer now uses only the node owner's content-digest identity. The fallback
+observation scan is removed. Unavailable native nodes retain their canonical
+path, availability, container-backed provenance and partial coverage in the
+inline graph; they yield no false actionable handoff or native Evidence link.
+An empty handoff list describes eligibility, not absence of native components.
+The per-query node index is allocated only after an eligible target exists; no
+persistent cache, schema, tool or target authority is added.
+
+Final real checks passed both cases with exact CLI/MCP trace parity. The available
+control preserved the complete parent/trace Evidence IDs, native facts, links,
+coverage and all token projections. The missing control preserved its parent,
+native node, source provenance and coverage, while removing the wrong handoff
+and container Evidence link.
+
+## Real input and scope
+
+[Official Joplin v3.7.21 Linux AppImage](https://github.com/laurent22/joplin/releases/download/v3.7.21/Joplin-3.7.21.AppImage),
+published 2026-09-25, 201,462,481 bytes, SHA-256
+`5d202dcaca31e14206826aa3f04125e7c218dbc3dc3d98479d811efbd6b856c1`.
+
+Read-only 7z 26.02 inspection found 6,324 ASAR files, 46 unpacked entries and
+three `.node` components. Extract only the ASAR and those three binaries:
+
+```sh
+input_root="$(mktemp -d -t rea-native-handoff-XXXXXX)"
+curl --fail --location --output "$input_root/Joplin-3.7.21.AppImage" \
+  https://github.com/laurent22/joplin/releases/download/v3.7.21/Joplin-3.7.21.AppImage
+sha256sum "$input_root/Joplin-3.7.21.AppImage"
+7z x -mmt=1 "-o$input_root/extracted" "$input_root/Joplin-3.7.21.AppImage" \
+  resources/app.asar \
+  resources/app.asar.unpacked/node_modules/keytar/build/Release/keytar.node \
+  resources/app.asar.unpacked/node_modules/sqlite3/lib/binding/napi-v6-linux-glibc-x64/node_sqlite3.node \
+  resources/app.asar.unpacked/node_modules/onnxruntime-node/bin/napi-v6/linux/x64/onnxruntime_binding.node
+python3 - "$input_root" <<'PY'
+import os, sys
+from pathlib import Path
+root = Path(sys.argv[1])
+source = root / 'extracted/resources'
+for mode in ('available', 'missing'):
+    case = root / 'cases' / mode
+    case.mkdir(parents=True)
+    os.link(source / 'app.asar', case / 'app.asar')
+    if mode == 'available':
+        for native in (source / 'app.asar.unpacked').rglob('*.node'):
+            target = case / native.relative_to(source)
+            target.parent.mkdir(parents=True, exist_ok=True)
+            os.link(native, target)
+PY
+```
+
+Verify the asset digest above before extraction. The ASAR is 233,575,014 bytes,
+SHA-256 `767df409b18effe60421fbefcca873de237c7e813eac744ecdeebbc5e825a33a`.
+Keytar is 76,960 bytes, SHA-256
+`0700016bcd761cbe02d275f2046c7cc42f39f730a50d66e3d3307aa9d413373a`.
+All extracted component hashes are in the JSON record. 7z reported one
+trailing-data warning for the AppImage's embedded SquashFS and exited 0.
+
+Both cases select only `package.json` for static source analysis while retaining
+the inventory. The available case stages all three native companions; the other
+43 unpacked entries are not staged. The missing case stages none. Whole-application
+coverage is therefore partial in both cases, independently of the selected
+native component's known or unknown byte identity. Neither the application nor
+a native analysis provider is executed. Native Evidence is explicitly simulated
+to test exact subject linkage. Live Electron, real Ghidra/Hopper/IDA behavior and
+Windows/macOS verification are **NOT_ENOUGH_INFO**.
+
+## Reproduce the public workflow
+
+Use pinned Node 24.18.0/npm 11.16.0, `npm ci`, then `npm run build:cached`.
+The verifier returns compact records and deletes its private exported parent and
+CLI input after owned cleanup. No downloadable application or large Evidence is
+archived.
+
+```sh
+for mode in available missing; do
+  node scripts/verify/javascript/native-handoffs.mjs \
+    --artifact "$input_root/cases/$mode/app.asar" \
+    --addon-path node_modules/keytar/build/Release/keytar.node \
+    --native-sha256 0700016bcd761cbe02d275f2046c7cc42f39f730a50d66e3d3307aa9d413373a \
+    --source-path package.json --rounds 3 --cli-parity --require-safe
+done
+```
+
+Add `--tokenizer-python /absolute/path/to/python-with-tiktoken-0.12.0` for the
+reported BPE projections. The npm alias `verify:javascript:native-handoffs`
+builds first; recorded costs use the direct Node invocation above after building.
+For a pre-fix observation, run the saved verifier on the baseline build without
+`--require-safe`; `safe_identity: false` is an incorrect answer, not a passing
+analyst task. Paths affect serialized requests and some Evidence identities.
+
+The exact resource runs additionally used the preserved
+[8 GiB observer](https://github.com/morluto/rea/blob/b3bdabab694af1f9801eff7caf75001b92314a58/experiments/analysis-efficiency/resource-run.py)
+and adjacent `resource_payload.py`; both hashes are pinned in the JSON.
+
+## Completed workflow costs and limits
+
+| Case                   | Revision |  Wall s |      CPU s | Peak cgroup bytes | Identity oracle            |
+| ---------------------- | -------- | ------: | ---------: | ----------------: | -------------------------- |
+| Native bytes available | Before   |  60.369 |  59.806670 |     2,662,629,376 | pass                       |
+| Native bytes available | After    | 258.885 | 106.221003 |     2,678,456,320 | pass, exact baseline facts |
+| Native bytes missing   | Before   |  57.930 |  57.479109 |     2,719,457,280 | wrong identity/linkage     |
+| Native bytes missing   | After    |  59.738 |  59.254776 |     2,571,235,328 | pass, unknown retained     |
+
+Whole-verifier resources include input hashing, analysis, three retained trace
+queries, export, CLI parity and tokenization. One CPU, 8 GiB hard cap, 6 GiB early
+stop, no swap; cleanup verified and no OOM. Memory includes charged file cache,
+not summed RSS. Counters are at payload exit while the observer remains alive.
+The available control's final wall and CPU times are substantially higher than
+baseline. Earlier post-fix controls on the original base took 61.657/74.956 s wall
+and 61.135962/60.352428 s CPU for available/missing cases; their complete receipts
+remain separately recorded. The final missing case took 59.738 s wall. This
+variability is unresolved and cannot be attributed to the native consumer change.
+These runs establish neither speedup nor performance equivalence. Input hashing warms the ASAR
+cache, build/OS caches and shared-host scheduling are uncontrolled. Fresh MCP
+sessions do not establish OS-cold conditions. First post-analysis and repeated
+warm-query samples are preserved separately; query CPU/memory are unmeasured.
+
+One defined task is one analysis summary plus one retained trace, with the path
+and prior Evidence already supplied. Static `cl100k_base`/tiktoken 0.12.0 counts
+include both requests, replies and name/description/inputSchema contracts:
+
+| Available-byte task, unchanged before/after     | BPE tokens |
+| ----------------------------------------------- | ---------: |
+| Selected catalog + structured replies           |     12,340 |
+| Selected catalog + complete tool replies        |     18,906 |
+| Full catalog projection + structured replies    |    202,292 |
+| Full catalog projection + complete tool replies |    208,858 |
+
+These are alternative reply/catalog projections. The selected catalog alone is
+5,570 BPE; the full projection is 195,522. Output schemas, annotations, host
+rendering/replay, reasoning, final answers and model billing are unmeasured.
+Repeated queries/export/CLI/tokenization are validation overhead outside this
+one-task token count, but inside resource receipts. Missing-case structured-task
+counts change 11,372 → 11,103 by removing unsupported claims; the incorrect
+baseline earns no efficiency credit.
+
+## Regression and verification record
+
+The initial four-file focused selection passed 24 tests. The two public cases
+were rerun after adding named summary-result parsing and correcting an optional
+architecture factory input. Omitting that field instead of passing null produced
+byte-identical normalized simulated Evidence in all four real-case inputs.
+The initial CLI-selector harness error and first typecheck failure are retained
+as harness/validation attempts, not product reproductions.
+
+After syncing the reviewed upstream changes, the updated application workflow
+CLI acceptance file passed 34 tests, bringing the focused selection to 58 distinct
+cases across five files. Both Joplin controls were then repeated on the final
+integration base. Canonical byte identity is also used by the related version
+matching/content comparison consumers; no equivalent observation-digest fallback
+was found there.
+
+Final `npm run check && npm run docs:check` passed on the integration base,
+including five documentation consistency tests: 128.732 s wall, 115.518920 s CPU,
+4,301,684,736 B peak cgroup memory, clean owned exit and no OOM.
+
+Final public and static/docs outcomes are recorded in the
+[compact measurements and source pins](javascript-native-handoff-identity-2026-10-15.json).

docs/windows-ghidra-p0.md

@@ -17,7 +17,7 @@ additional permission flag or degraded mode is required.
 - Node.js 22.x (>=22.19), 24.x (>=24.11), or 26+;
 - an operator-installed Ghidra 12.1.x distribution (verified with 12.1.4);
 - a 64-bit full JDK inside that installation's declared Java range (JDK 21 or newer, with no maximum, for current 12.1 releases);
-- an explicit native, non-managed x86 or x86-64 PE application/DLL, or a
+- an explicit native, non-managed x86, x86-64, ARMNT (Thumb-2) or ARM64 PE application/DLL, or a
   [raw NMOS 6502 image](raw-binary.md#nmos-6502) with a caller-supplied entry; and
 - the 26 read-only Ghidra inventory, memory, and function-analysis operations.
 
@@ -135,9 +135,55 @@ and artifact SHA-256 before loading the package-owned addon. Native failure
 reasons remain distinct from a valid installation and an unsupported target.
 
 PE header classification and SHA-256 come from the same open file. The Windows
-provider admits native x86 and x86-64 PE applications and DLLs and rejects unsupported roles,
+provider admits native x86, x86-64, ARMNT (Thumb-2) and ARM64 PE applications and DLLs and rejects unsupported roles,
 architectures, managed images, and malformed headers. The original selected
-source coordinate is preserved for native admission.
+source coordinate is preserved for native admission. The original COFF machine is
+retained separately from the normalized CPU family. ARM admission requires the
+observed ARMNT machine 0x01C4; ARM64 requires 0xAA64. Legacy ARM (0x01C0), Thumb
+(0x01C2), and unavailable machine facts do not select this Windows lane.
+Without a caller-selected processor override, Ghidra's PE loader
+selects the language from the PE machine field: ARM (`IMAGE_FILE_MACHINE_ARMNT`)
+imports as `ARM:LE:32:v8` (Thumb), ARM64 (`IMAGE_FILE_MACHINE_ARM64`) as
+`AARCH64:LE:64:v8A`. Explicit language/compiler selections remain caller choices
+in the analysis profile and are recorded as explicit imports. The source
+fixture verifier compares the complete observed language ID, not just its CPU
+family prefix. ARM/ARM64 admission is a bounded extension of the x86/x64
+lane. `npm run verify:ghidra:windows -- --arm` or `-- --arm64`, optionally
+combined with `--dll`, selects encoded Thumb or ARM64 PE fixtures on a Windows
+x64 host. The packaged verifier accepts the same flags. These fixtures are not
+compiler-produced programs. ARM64EC (machine 0xA641) and ARM64X are not admitted;
+this target support does not establish Windows-on-Arm host support.
+
+Real Windows x64 verification with Ghidra 12.1.4 and JDK 21 completed the ARM64
+application and ARMNT/Thumb DLL fixtures' 26 read-only operations, constant-return
+decompilation, byte searches, CLI/MCP file-offset parity, source identity, and
+owned cleanup. The DLL fixture also verifies the exported caller/callee
+relationship. The x86/x64 compact-reference value-flow battery does not cover
+ARM or ARM64 instructions. ARM64 EXE file-offset verification includes ambiguous
+mappings; the single-section ARM32 DLL cannot exercise that case.
+
+The runs used two logical CPUs, 512 MiB Node/JVM heaps, a 900-second startup
+budget, and an explicit `REA_GHIDRA_DISABLED_ANALYZERS=WindowsResourceReference`
+test configuration. Default-analyzer runs exceeded the selected startup budget;
+resource-reference analysis and default-configuration completion are not
+established by these results. A Microsoft-signed Coreinfo ARM64 application also
+completed focused MCP decompilation, repeated readback, source-mapped memory
+reads, unchanged-input and cleanup checks. On this constrained host the first
+function query, including cold analysis, took about 434 seconds; its identical
+same-session read took about 23 ms and returned about 4.6 KiB of Evidence. These
+are individual observations, not general latency guarantees. Public ARM32
+DLL verification also completed against Microsoft Visual C++ 2013 ARM
+`msvcr120.dll`, including decompilation of its exported `abs`, repeated readback,
+source-mapped bytes, unchanged-input and cleanup checks. Its first function
+query took about 601 seconds; the same-session repeat took about 11 ms and
+returned about 4.8 KiB of Evidence. After preserving original COFF machines in the
+admission/profile, a further real Windows check rejected legacy ARM/Thumb and
+retained ARMNT/ARM64 admission. A rebuilt-candidate MSVCR120 query completed
+in about 407 seconds with a 15 ms repeat (4,830 bytes of Evidence), unchanged
+input and complete cleanup. That retry selected an 1,800-second startup budget
+after an additional 900-second attempt timed out. The resource bounds were
+unchanged; these observations do not establish a speedup. Configured processor
+selection is covered by profile tests, not these real-provider runs.
 
 Admission opens and checks every path component without following a reparse
 point. It retains handles, verifies local NTFS semantics, and observes volume
@@ -200,6 +246,8 @@ npm run verify:ghidra:windows
 npm run verify:ghidra:windows -- --x86
 npm run verify:ghidra:windows -- --dll
 npm run verify:ghidra:windows -- --x86 --dll
+npm run verify:ghidra:windows -- --arm64
+npm run verify:ghidra:windows -- --arm64 --dll
 npm run verify:ghidra:windows:package -- --x86
 npm run verify:ghidra:windows:package -- --dll
 npm run verify:ghidra:windows:package -- --x86 --dll

experiments/analysis-efficiency/agent-trial.py

@@ -4,6 +4,7 @@ import argparse
 import hashlib
 import json
 import os
+import re
 from pathlib import Path
 import shutil
 import signal
@@ -22,7 +23,31 @@ def file_digest(path):
     return digest.hexdigest()
 
 
-def runtime_fingerprint(root):
+def pinned_experimental_engine(settings):
+    expected = settings.get('REA_ATOMIC_NATIVE_IR_ENGINE_SHA256')
+    if expected is None:
+        return None
+    if not re.fullmatch(r'[0-9a-f]{64}', expected):
+        raise ValueError('Experimental native engine pin must be a SHA-256')
+    path = Path(settings.get('REA_KUNA_COMMAND', ''))
+    if not path.is_absolute() or not path.is_file() or not os.access(path, os.X_OK):
+        raise ValueError('Pinned experimental engine requires an absolute executable file')
+    actual = file_digest(path)
+    if actual != expected:
+        raise ValueError('Experimental engine digest differs from its pin; use a preserved build artifact')
+    return {'path': str(path), 'sha256': actual}
+
+
+def experimental_engine_unchanged(receipt):
+    if receipt is None:
+        return None
+    try:
+        return file_digest(Path(receipt['path'])) == receipt['sha256']
+    except OSError:
+        return False
+
+
+def runtime_fingerprint(root, entrypoint=None):
     files = []
     for name in ['dist', 'bridge']:
         for path in sorted((root / name).rglob('*')):
@@ -30,6 +55,18 @@ def runtime_fingerprint(root):
                 raise ValueError('Runtime fingerprint does not follow symlinks')
             if path.is_file():
                 files.append({'path': path.relative_to(root).as_posix(), 'sha256': file_digest(path)})
+    if entrypoint is not None:
+        # Include the experimental modules and their installed QuickJS JS/WASM,
+        # plus the shared fact owner; the entrypoint alone does not pin them.
+        for directory in [entrypoint.parent, root / 'experiments/atomic-re-tools']:
+            for path in sorted(directory.rglob('*')):
+                if path.suffix not in {'.mjs', '.cjs', '.js', '.wasm'} and path.name not in {'package.json', 'package-lock.json'}:
+                    continue
+                if path.is_symlink():
+                    raise ValueError('Experimental runtime fingerprint does not follow symlinks')
+                if path.is_file():
+                    label = path.relative_to(root).as_posix() if path.is_relative_to(root) else path.as_posix()
+                    files.append({'path': label, 'sha256': file_digest(path)})
     files.sort(key=lambda row: row['path'])
     text = json.dumps(files, separators=(',', ':'), ensure_ascii=False)
     return files, hashlib.sha256(text.encode()).hexdigest()
@@ -39,6 +76,12 @@ def main():
     parser = argparse.ArgumentParser()
     parser.add_argument('--mode', choices=['direct', 'client-code'], default='direct')
     parser.add_argument('--runtime-root', type=Path, required=True)
+    parser.add_argument('--mcp-entrypoint', type=Path,
+                        help='Caller-selected experimental stdio MCP entrypoint; leaves production dispatcher unchanged')
+    parser.add_argument('--server-name', default='rea')
+    parser.add_argument('--protocol-profile', choices=['canonical', 'atomic', 'atomic-direct'], default='canonical')
+    parser.add_argument('--profile-parent', type=Path,
+                        help='Absolute parent outside the repository for the isolated model workspace')
     parser.add_argument('--revision', required=True)
     parser.add_argument('--prompt', type=Path, required=True)
     parser.add_argument('--artifact-path', help='Replace {{artifact_path}} in a question template')
@@ -54,6 +97,10 @@ def main():
                         help='Retain owned tool/usage items; excludes account/ambient state')
     parser.add_argument('--guide-enabled-tools', action='store_true',
                         help='Opt-in exact-name discovery guidance for selected REA tools')
+    parser.add_argument('--normalize-mcp-guidance', action='store_true',
+                        help='Opt-in common envelope/stopping guidance, identical across protocol lanes')
+    parser.add_argument('--guide-long-waits', action='store_true',
+                        help='Opt-in common client wait guidance; does not change analysis contracts or deadlines')
     args = parser.parse_args()
     if args.guide_enabled_tools and args.enabled_tools is None:
         parser.error('--guide-enabled-tools requires --enabled-tools')
@@ -94,13 +141,19 @@ def main():
     if undeclared:
         raise ValueError('Declare inherited provider settings explicitly in --providers: '
                          + ', '.join(undeclared) + '. Values are not logged.')
+    # Fail before copying credentials or spending model tokens. Cargo test/build
+    # can replace a mutable target/release path with a different feature build.
+    experimental_engine = pinned_experimental_engine(provider_env)
     source_auth_home = (Path(os.environ['CODEX_HOME']) if 'CODEX_HOME' in os.environ
                         else Path.home() / '.codex')
     source_auth = source_auth_home / 'auth.json'
     args.output.mkdir(parents=True, exist_ok=False, mode=0o700)
-    runtime_files, runtime_digest = runtime_fingerprint(repository)
+    runtime_files, runtime_digest = runtime_fingerprint(repository, args.mcp_entrypoint)
     (args.output / 'runtime-files.json').write_text(json.dumps(runtime_files))
-    profile = Path(tempfile.mkdtemp(prefix='profile-', dir=args.output))
+    if args.profile_parent and (not args.profile_parent.is_absolute()
+            or args.profile_parent.resolve().is_relative_to(repository.resolve())):
+        raise ValueError('profile-parent must be an absolute directory outside runtime-root')
+    profile = Path(tempfile.mkdtemp(prefix='rea-model-profile-', dir=args.profile_parent or args.output))
     trial_codex_home = profile / 'codex'
     account = profile / 'account'
     workspace = profile / 'workspace'
@@ -119,7 +172,12 @@ def main():
         'NODE_OPTIONS': '--max-old-space-size=512',
     })
     target_digest = file_digest(Path(args.artifact_path)) if args.artifact_path else None
-    command_args = [str(repository / 'scripts/rea.mjs'), 'mcp']
+    if not args.server_name.replace('-', '').replace('_', '').isalnum():
+        raise ValueError('server-name must contain only letters, digits, hyphens or underscores')
+    if args.mcp_entrypoint and not args.mcp_entrypoint.is_absolute():
+        raise ValueError('mcp-entrypoint must be absolute')
+    command_args = ([str(args.mcp_entrypoint)] if args.mcp_entrypoint
+                    else [str(repository / 'scripts/rea.mjs'), 'mcp'])
     instruction = (
         'Use the registered REA MCP server for this investigation. Static analysis '
         'only: do not execute target code, fetch URLs, launch a GUI, or modify targets. '
@@ -132,6 +190,23 @@ def main():
         'Prefer summary/focused views where appropriate. Do not invent successful facts '
         'after a tool error. No other agents. Stop once the requested facts are supported.'
     )
+    if args.protocol_profile in {'atomic', 'atomic-direct'}:
+        instruction = (
+            'Use the registered ' + args.server_name + ' MCP server for this static binary investigation. '
+            'Call reference to inspect API contracts, then compose the supported analysis calls. '
+            'Do not execute target code, fetch URLs, launch a GUI, modify targets, use shell/file tools, '
+            'or use other agents. API methods return {ok,context,data} or {ok:false,error}; '
+            'execute_javascript returns the script value, with script-derived authority. '
+            'Preserve real evidence_id values, artifact digest, uncertainty and traversal boundaries. '
+            'Check failures before reading data. Do not infer absence from partial coverage. '
+            'Do not return complete inventories or source reports when a focused answer suffices. '
+            'Close the binary when finished. No precomputed answer is available to you.'
+        )
+        if args.protocol_profile == 'atomic-direct':
+            instruction = instruction.replace(
+                'execute_javascript returns the script value, with script-derived authority. ',
+                'Call the registered facts_* tools directly; no JavaScript execution tool is registered. '
+            )
     if args.authority == 'browser':
         instruction = (
             'Use the registered REA tools for the explicitly selected public browser scenario. '
@@ -147,6 +222,13 @@ def main():
             'possible and select the requested normalized_result fields before returning '
             'intermediate output to model context. Keep real Evidence IDs and limitations.'
         )
+    if args.normalize_mcp_guidance:
+        instruction += (
+            ' MCP calls return transport envelopes: prefer structuredContent when present; '
+            'otherwise parse the text content JSON. Do not print both duplicate encodings. '
+            'Apply the same rule to every tool. Stop once the requested facts are supported, '
+            'state missing coverage explicitly, and close the binary when finished.'
+        )
     if args.guide_enabled_tools:
         instruction += (
             ' Configured REA tool function names for this trial: '
@@ -155,6 +237,15 @@ def main():
             'return only their metadata. Preserve the original analyst question, '
             'dependency order and source Evidence.'
         )
+    if args.guide_long_waits:
+        instruction += (
+            ' When the client executor reports a running cell during a long analysis, '
+            'resume that same cell with wait and yield_time_ms:30000, within the '
+            'advertised tool schema, rather than repeated one-second polls. '
+            'The wait may return early when the result is ready. Do not sleep or '
+            'start duplicate analysis calls just to check completion. Keep the '
+            'original analysis dependencies, evidence and stopping requirements.'
+        )
     quote = json.dumps
     config = '\n'.join([
         'approval_policy = "never"', 'sandbox_mode = "read-only"',
@@ -166,16 +257,16 @@ def main():
         'code_mode = ' + ('true' if args.mode == 'client-code' else 'false'),
         'code_mode_only = ' + ('true' if args.mode == 'client-code' else 'false'),
         '[history]', 'persistence = "none"',
-        '[mcp_servers.rea]', 'command = ' + quote(node),
+        '[mcp_servers.' + args.server_name + ']', 'command = ' + quote(node),
         'args = ' + quote(command_args), 'startup_timeout_sec = 60',
         'tool_timeout_sec = 600',
         *([] if enabled_tools is None else ['enabled_tools = ' + quote(enabled_tools)]),
-        '[mcp_servers.rea.env]',
+        '[mcp_servers.' + args.server_name + '.env]',
         *[key + ' = ' + quote(value) for key, value in provider_env.items()],
     ]) + '\n'
     # MCP read-only hints cover normal static tools. Explicitly authorize only
     # the session-opening primitive needed by the requested native tasks.
-    config += '[mcp_servers.rea.tools.open_binary]\napproval_mode = "approve"\n'
+    config += '[mcp_servers.' + args.server_name + '.tools.open_binary]\napproval_mode = "approve"\n'
     for tool in ['analyze_javascript_application', 'inspect_analysis_view',
                  'inspect_android_package', 'search_android_classes',
                  'inspect_android_class', 'inspect_android_method',
@@ -189,7 +280,7 @@ def main():
                  'inspect_evm_interface', 'get_evidence_bundle', 'export_evidence_bundle',
                  'capture_browser_scenario',
                  'resolve_containing_procedure', 'analyze_function', 'close_binary']:
-        config += '[mcp_servers.rea.tools.' + tool + ']\napproval_mode = "approve"\n'
+        config += '[mcp_servers.' + args.server_name + '.tools.' + tool + ']\napproval_mode = "approve"\n'
     try:
         if source_auth.exists():
             shutil.copyfile(source_auth, auth)
@@ -209,10 +300,17 @@ def main():
             'mode': args.mode, 'model': args.model, 'reasoning': args.reasoning,
             'capture_tool_context': args.capture_tool_context,
             'guide_enabled_tools': args.guide_enabled_tools,
+            'normalize_mcp_guidance': args.normalize_mcp_guidance,
+            'guide_long_waits': args.guide_long_waits,
             'driver_commit': subprocess.check_output(
                 ['git', 'rev-parse', 'HEAD'], cwd=driver_repository, text=True).strip(),
             'runtime_revision': args.revision,
             'runtime_root': str(repository),
+            'mcp_server_name': args.server_name,
+            'protocol_profile': args.protocol_profile,
+            'isolated_profile_parent': str(args.profile_parent) if args.profile_parent else str(args.output),
+            'mcp_entrypoint': str(args.mcp_entrypoint) if args.mcp_entrypoint else None,
+            'mcp_entrypoint_sha256': file_digest(args.mcp_entrypoint) if args.mcp_entrypoint else None,
             'selected_artifact_path': args.artifact_path,
             'selected_artifact_sha256_before': target_digest,
             'enabled_tools': enabled_tools,
@@ -230,6 +328,7 @@ def main():
             'question_template_sha256': template_digest,
             'provider_settings_sha256': hashlib.sha256(
                 json.dumps(provider_env, sort_keys=True).encode()).hexdigest(),
+            'experimental_engine': experimental_engine,
             'provider_search_path_sha256': hashlib.sha256(
                 provider_env.get('PATH', os.environ.get('PATH', '')).encode()).hexdigest(),
             'provider_environment_policy_sha256': hashlib.sha256(
@@ -282,6 +381,9 @@ def main():
         summary = {
             'mode': args.mode, 'requested_model': args.model, 'reasoning': args.reasoning,
             'runtime_revision': args.revision,
+            'runtime_inputs_unchanged': runtime_fingerprint(repository, args.mcp_entrypoint)[1] == runtime_digest,
+            'experimental_engine': experimental_engine,
+            'experimental_engine_unchanged': experimental_engine_unchanged(experimental_engine),
             'capture_tool_context_requested': args.capture_tool_context,
             'selected_artifact_sha256_before': target_digest,
             'selected_artifact_sha256_after': final_target_digest,
@@ -296,6 +398,10 @@ def main():
         }
         (args.output / 'summary.json').write_text(json.dumps(summary, indent=2) + '\n')
         print(json.dumps(summary))
+        if not summary['runtime_inputs_unchanged']:
+            raise RuntimeError('Runtime inputs changed during the trial; retain costs but reject the comparison')
+        if summary['experimental_engine_unchanged'] is False:
+            raise RuntimeError('Experimental engine changed during the trial; retain costs but reject the comparison')
     finally:
         try:
             if args.capture_tool_context:

experiments/analysis-efficiency/test_engine_pin.py

@@ -0,0 +1,46 @@
+import hashlib
+import importlib.util
+from pathlib import Path
+import tempfile
+import unittest
+
+
+spec = importlib.util.spec_from_file_location('agent_trial', Path(__file__).with_name('agent-trial.py'))
+trial = importlib.util.module_from_spec(spec)
+spec.loader.exec_module(trial)
+
+
+class ExperimentalEnginePinTests(unittest.TestCase):
+    def test_missing_pin_preserves_other_provider_trials(self):
+        self.assertIsNone(trial.pinned_experimental_engine({}))
+        self.assertIsNone(trial.experimental_engine_unchanged(None))
+
+    def test_feature_build_replacement_and_removal_invalidate_pinned_engine(self):
+        with tempfile.TemporaryDirectory() as directory:
+            engine = Path(directory) / 'kuna'
+            engine.write_bytes(b'embedded-spec-build')
+            engine.chmod(0o755)
+            settings = {'REA_KUNA_COMMAND': str(engine),
+                        'REA_ATOMIC_NATIVE_IR_ENGINE_SHA256': hashlib.sha256(engine.read_bytes()).hexdigest()}
+            receipt = trial.pinned_experimental_engine(settings)
+            self.assertTrue(trial.experimental_engine_unchanged(receipt))
+            engine.write_bytes(b'non-embedded-test-build')
+            with self.assertRaisesRegex(ValueError, 'differs from its pin'):
+                trial.pinned_experimental_engine(settings)
+            self.assertFalse(trial.experimental_engine_unchanged(receipt))
+            engine.unlink()
+            self.assertFalse(trial.experimental_engine_unchanged(receipt))
+            with self.assertRaisesRegex(ValueError, 'absolute executable'):
+                trial.pinned_experimental_engine(settings)
+
+    def test_malformed_pin_and_relative_command_are_rejected(self):
+        for settings in [
+            {'REA_ATOMIC_NATIVE_IR_ENGINE_SHA256': 'bad'},
+            {'REA_ATOMIC_NATIVE_IR_ENGINE_SHA256': '0' * 64, 'REA_KUNA_COMMAND': './kuna'},
+        ]:
+            with self.assertRaises(ValueError):
+                trial.pinned_experimental_engine(settings)
+
+
+if __name__ == '__main__':
+    unittest.main()

experiments/atomic-codemode/README.md

@@ -0,0 +1,301 @@
+# Atomic RE through Code Mode
+
+This is an **in-progress experiment**, not an integrated provider release. Branch
+`experiment/atomic-codemode-20261016` starts at main
+`42fc6c53ae188978deb3bedfe88287b6b316ba22` and carries the previous atomic-facts
+prototype. Historical measurements in `../atomic-re-tools/` retain their original
+baselines. No PR or merge is intended.
+
+The goal is a real model completing reverse-engineering questions through
+composable facts, with Code Mode as the calling adapter. A configured feature flag,
+a scripted verifier or a small response does not prove that goal.
+
+The reference is [Hex-Rays IDA MCP](https://github.com/HexRaysSA/ida-mcp/blob/df6de7985b534a964c125d5a86e41b26b17b14f6/ida_mcp/mcp.py):
+an on-demand `reference` lookup and an execution entrypoint over a richer analysis
+API. This experiment uses JavaScript and a QuickJS WASM environment rather than
+Python in the host process. It exposes only declared static analysis capabilities.
+It does not add target execution, patching or GUI control.
+
+Three boundaries own the implementation:
+
+- `provider.mjs`, `native-ir.mjs`: existing REA native target/process owners and parsed Kuna
+  string/reference/SSA facts, retained sources, freshness and cleanup.
+- `api.mjs`: named, validated fact/relationship contracts and session admission.
+  SDK callers, CLI scripts and both MCP adapters use these same methods.
+- `executor.mjs`, `server.mjs`, `cli.mjs`: bounded JavaScript composition and
+  presentation. No analysis is reimplemented in the execution environment.
+
+The complete Code Mode MCP catalog is `open_binary`, `reference`,
+`execute_javascript`, `close_binary`. Setting `REA_ATOMIC_ADAPTER=direct` exposes
+the same fact API as `facts_*` MCP operations for comparison. This is a distinct
+experimental server; the production REA catalog is not filtered or replaced.
+
+Use the pinned Node, an existing REA build and a caller-supplied Kuna 1.833:
+
+```sh
+npm ci --prefix experiments/atomic-codemode --ignore-scripts
+node --test experiments/atomic-codemode/*.test.mjs
+REA_KUNA_COMMAND=/absolute/path/to/kuna node experiments/atomic-codemode/server.mjs
+```
+
+The local MCP registration for real model trials is named `rea-codemode`.
+`agent-trial.py --mcp-entrypoint /absolute/path/to/server.mjs
+--server-name rea-codemode --protocol-profile atomic --profile-parent /tmp` creates an isolated
+configuration, home and empty working directory for the model. Shell/file tools,
+web search and delegation are disabled. No repository or oracle is placed in that
+workspace. The copied authentication file is removed with the isolated profile.
+Model output, actual MCP calls, original errors and usage remain in the trial
+directory. `REA_ATOMIC_TRACE` and `REA_ATOMIC_EVIDENCE_DIR` retain actual API calls
+and complete sources outside the model workspace when explicitly selected by the
+evaluator. The ambient environment and credentials are not logged.
+
+Current bounds: 64 MiB guest heap, 32 KiB code and script value/error payloads, 64 host calls per execution,
+120 seconds wall deadline and 2 seconds of guest synchronous execution; native requests have existing supervision, an 8 MiB
+capture budget and 1 GiB function-memory limit. Retained source storage is 64 MiB
+per target, with two resident indexes (each derived from at most 2 MiB diagnostic
+or 8 MiB native wire source). Compressed native graphs admit at most 32 MiB decoded
+JSON plus independent row/text/relationship bounds. One decoded Evidence text view
+up to 32 MiB is also retained per lease; these source bounds are not parsed-heap
+size guarantees. Sources are
+content-digested and retained until close; explicit evaluator copies survive
+close. Capacity failures are explicit, not clipped facts. Leases and snapshot
+handles prevent cross-session joins.
+
+Invocation context and observed source IDs are additional bounded receipt
+metadata; the script payload bound is not a 32 KiB whole-MCP-wire limit. A real
+MCP probe found that a 40,000-byte guest exception bypassed that payload policy
+and produced a 40,183-byte response. Oversized errors now return explicit
+`output_budget` with omission metadata. Escaping expansion is checked too;
+ordinary smaller errors remain intact. `verify-executor.mjs` confirms real MCP
+delivery and continued access to source Evidence collected before the exception.
+It measures response bytes, not model tokens or an analysis speedup. Targeted JS
+tests at this stage totaled 42, alongside the separate real-provider and native-writer checks.
+
+The diagnostic SSA format still has incomplete CFG and phi predecessor coverage.
+The opt-in [native recovered IR experiment](native-engine.md) adds full recovered
+CFG edges, exact value/operand identities, phi incoming-edge bindings, selected
+`values`/`blocks` reads and derived control dependence. It requires an explicitly
+pinned patched engine; stock Kuna does not gain those capabilities from a wrapper.
+Runtime successors, memory aliases and machine-semantic equivalence remain
+unknown. Other providers are not yet wired to this experimental service.
+
+The initial [actual model comparison](trials.md) favors the focused traditional
+workflow. The seeded value-origin task has lower cumulative input and wall time
+through Code Mode, but initially exposes an IR-to-machine-path error. The revised
+trial correctly explains that machine path with lower observed task costs;
+broader claims still require more than this single seeded task.
+
+The large SQLite VM diagnostic task also completes when its native graph exceeds
+capacity. Atomic direct beats Code Mode on cumulative input and wall time in
+that sample, and a traditional reused-project run is fastest. The actual CLI
+client already exposes JavaScript dispatch in every lane, regardless of requested
+feature flags. MCP request counts therefore do not establish fewer model rounds;
+the ledger now records actual model responses and client dispatch/wait calls.
+
+`instructions` reads selected machine byte/decode windows through the same
+owner. Recovered operations use `origin_address` and blocks `origin_start`;
+these locations cannot be substituted for machine branch destinations. The
+value-origin trial demonstrates why this boundary matters. The experiment
+now has fourteen fact methods, still behind the same four-tool Code Mode
+catalog.
+
+`calls` connects recovered call operations to native target specifications and
+separates argument SSA values from passing storage. `values(is_input:true)`
+selects retained callee inputs for exact storage comparisons. These candidate
+links preserve snapshot identities, missing specifications, unknown storage and
+runtime-binding limits. The [cross-function task and oracle](call-oracle.md)
+are fixed before its real-model trials; the graph wire format is version 3, with independent version-1 call facts.
+
+`references` now selects incoming exact-address references or outgoing
+instruction/function references, with native kind filtering, field selection and
+paging. Source Evidence plus `source_row` identifies an occurrence. Function
+applicability, excluded/unattributed rows, indirect targets and discovery stay
+explicit. `fields:[]` requests only the always-retained identity columns across
+projection contracts. The [reference question and independent oracle](reference-oracle.md)
+cover real SQLite VM aggregates and a machine call omitted from optimized IR.
+
+`bytes` reads explicit virtual-address ranges as native loaded-image hex without
+decoding them as instructions. Batched reads preserve short reads and individual
+failures, with 64 KiB aggregate selected-byte capacity to bound hex/bridge growth.
+Complete hexdumps remain in Evidence. The original-file-offset mapping and live
+runtime memory remain unknown. This closes a direct byte-reading contract gap
+for data tables; it does not establish arbitrary memory-alias analysis.
+
+`functions` provides selected name/alias/entry reads. Native snapshots report
+code and recovered-IR facets separately, preserving readable code when graph
+extraction is unavailable. Real capacity and cold/warm MCP checks are documented
+in [native-engine.md](native-engine.md). Whole-model costs and unequal-depth
+comparisons remain separate in the trial report.
+
+The corrected SQLite comparison reaches the formatter implementation in all
+three interfaces. Code Mode uses fewer calls than atomic direct in that sample;
+a traditional run reusing its persistent whole-project cache is faster than
+Code Mode's fresh lease. Cache state, completed-answer coverage and actual
+uncached tokens must therefore accompany any cost comparison.
+
+SDK views share immutable parsed inventories with their owner. A real SDK probe
+initially changed a returned string and made the next query disappear under the
+same Evidence ID. Freezing at the cache boundary fixes that producing rule,
+without copying or parsing the entire inventory for each read. `verify-sdk.mjs`
+checks the real provider's string/metadata/function views and complete retained
+source digest. The fixed probe passes under the bounded runner (1.328 seconds,
+1.144 family CPU seconds, 110.6 MiB kernel family peak, zero OOM kills and verified
+payload cleanup); it is a correctness check, not a model-cost measurement.
+
+Encoded size is another real coverage boundary: two SQLite functions initially
+lose even their C because pretty JSON exceeds transport capacity. The native
+patch now uses bounded compact serialization and a graph-only failure when
+needed. Parser's full graph fits; Pragma retains identical C with an explicit
+missing graph. The 8 MiB bound remains. See the exact sizes, rebuilt engine digest
+and verification scope in [native-engine.md](native-engine.md).
+
+Completion requires actual model Code Mode use, independently checked task
+answers, shared atomic/direct/Code Mode semantics, needed relation coverage,
+failure/cancellation/identity tests, and equivalent completed-task measurements
+including discovery/retries, actual model usage, cold/warm analysis, CPU and
+process-family peak memory. Negative cases such as the small SQLite function
+must remain in the comparison.
+
+The outgoing-reference real-model pair now completes with identical required
+facts and explicit coverage limits. In its single sample, Code Mode uses 264,680
+cumulative input / 93.837 seconds versus direct's 285,086 / 106.183. The shared
+CLI/API also verifies the VM aggregates on unpatched stock Kuna 1.833. These are
+scoped positive results alongside the earlier mixed adapter comparisons, not
+proof of general superiority. Details and retained failed/partial attempts are
+in [trials.md](trials.md).
+
+Recovered calls and their boundary values now survive full-graph capacity
+failure. The real VM capture preserves 416 call operations and 1,294 boundary
+values while its graph remains unavailable. Values clearly state their partial
+scope; definition IDs do not imply readable defining operations. See the
+[boundary question and oracle](boundary-oracle.md) and its completed real-model
+pair in [trials.md](trials.md). Both answers pass, using one VM capture each.
+Code Mode's single-sample latency is lower, with mixed context costs; broader
+efficiency remains unproven. The focused JS suite at that stage passes 49 tests.
+
+Available snapshot call facets now include `by_opcode` known/unknown-target
+counts and their own coverage. The owner derives them once from the same
+validated immutable calls used by selected reads, including dead operations;
+target knowledge is independent of opcode. Empty available facets report zeros,
+while unavailable facets omit counts. This small aggregate addresses an observed
+cost in the real boundary pair: paging the entire call set just to count it.
+Both subsequent real-model answers satisfy the unchanged question/oracle without
+that paging. In the new sample direct is faster and lower on cumulative/uncached
+input than Code Mode, reversing the preceding latency ordering. Whole-task
+effects remain mixed; see [the complete results](trials.md). This aggregate is a
+shared fact-layer improvement, available to either adapter.
+
+The shared owner also checks the native cleanup `Result` before releasing its
+lease or clearing retained state. A tagged cleanup failure now stays retryable;
+MCP errors retain the lease context and reported resources. The real SDK
+failure-injection verifier first reproduces the old false success, then verifies
+both close and target replacement retain the staged file/owner until a successful
+retry. Both public MCP adapters preserve that error and retry successfully.
+These use an injected tagged failure, not an observed engine cleanup incident;
+native reads and subsequent cleanup use the real provider. See the recorded
+scope and resource receipts in [native-engine.md](native-engine.md).
+
+A [subsequent diagnostic probe](projection-experiment.md) measures the VM's full
+17 MB recovered graph and much smaller selected definition neighborhoods. It
+supports trying a bounded operation/value facet, while preserving unresolved
+definitions and unavailable CFG. The following implementation now makes that
+facet callable; the original diagnostic remains separate from public results.
+
+`snapshot.facets.operations` distinguishes a complete recovered graph from a
+bounded call-rooted definition projection. `operations`, `values`, selected
+definition/use `relations` and `slice` share the same parsed operation owner in
+either case. Projection coverage preserves unexpanded definition counts and
+selected slice frontier reasons; CFG/controller reads remain unavailable.
+Native operation reads retain `is_dead` and literal operand/value identities.
+Coverage returns compact counts and bounded issue text, with complete native
+details retained in Evidence. [The new question/oracle](projection-oracle.md)
+are fixed before real-model trials of the PHI format alternatives.
+
+Both first PHI-format model answers pass the full oracle using one VM analysis
+and a six-operation selected trace, with the global graph/use limits intact.
+The trace reveals two format sources at each shared call, verified against byte
+and machine evidence. [Costs and source review](trials.md) also record a direct
+documentation-lookup miss; named discovery needs fixing before rerunning the
+same task. No general adapter speedup follows from that first pair.
+
+Named reference lookup now accepts canonical IDs or exact displayed names,
+including space/comma-separated lists. One presentation map supplies direct
+registration and documentation names; unmatched terms remain explicit. Selected
+schemas come from named contracts, use input semantics (defaulted fields stay
+optional), and are cached immutably by Zod schema identity for the process.
+Validation stays in the API registry; documentation has its own module. Both
+actual MCP catalogs and all fourteen reference schemas validate. The unchanged
+PHI question/oracle were rerun: both answers pass again, but direct has a native
+timeout followed by a successful retry. Complete costs and the unknown timeout
+cause remain in [trials.md](trials.md); this does not isolate an adapter speedup.
+
+The next observed atom boundary is a wide PHI. One real VM operation has 156
+inputs; its old default single-node slice is about 44 KB and exceeds the guest's
+32 KiB result budget. Native operation/slice defaults now return opcode, output
+and exact input/definition/constant counts. Explicit detailed fields still work.
+`relations({kind:"operands",seeds:[id],input_indices:[0,105,155]})` reads selected
+slots; omitting the selector pages all slots. Each slot retains value/definition
+and phi predecessor/edge identity, including null values and definition-free
+constants/inputs. Counts concern recovered facts, not source function arity.
+
+Full and projected native operations use the same owner. Relationship paging
+counts existing groups before materializing only the requested page; identity
+reads deduplicate requested IDs in caller order. Diagnostic SSA explicitly lacks
+native operand relations. The discriminated relation schema advertises the same
+seed types and selector applicability enforced at runtime. The fourteen-method
+API and both adapters remain shared; no new engine or output-budget increase is
+needed. The [seeded wide-operand task/oracle](operands-oracle.md) is separate from
+the earlier PHI-format question and is pinned before model trials.
+
+Both actual wide-operand model answers now pass that fixed oracle, including
+distinct edges sharing a value and absent-slot versus unretained-seed semantics.
+Each performs one native analysis and reads only selected slots, distinct values
+and their immediate definitions. Code Mode takes 95.699 seconds versus direct's
+91.817, with nearly equal cumulative input and lower uncached input. Its four
+discovery lookups prevent fewer MCP requests from becoming fewer model responses.
+These [mixed costs](trials.md#selected-wide-phi-input-relationships) validate the
+fact boundary while leaving broad efficiency unproven. The current shared JS
+suite passes 57 tests; the Python cost-ledger suite passes three.
+
+Discovery now also offers `reference({view:"index"})`: the complete fourteen
+method names and descriptions, without schemas or return contracts. Follow with
+`view:"contracts"` and exact names for only the needed inputs. The default
+contract view and named/topic lookup stay compatible. Both views use the same
+registry and presentation map; index access never constructs unrequested schemas.
+The index describes declared methods, while `snapshot` remains the live coverage
+authority. This addresses observed blind discovery without inventing aliases or
+changing analysis facts; its whole-task effect must be measured on the unchanged
+wide-operand question.
+
+The identical question now completes correctly through both indexed adapters.
+Code Mode uses the intended index plus one named lookup; discovery drops from
+four calls to two, without changing the seven fact reads. The new pair takes
+about 98.8 seconds each; cumulative input favors direct, uncached input favors
+Code Mode. All four wide-operand answers pass, with no retry or OOM. Costs remain
+mixed. A selected-slot read still incurs whole-function native decompilation on
+a fresh lease; phase profiling is the next computation question, separate from
+the now finer fact/query boundary.
+
+The [native phase study](native-compute-study.md) now measures the VM drive at
+about 34 seconds, C rendering at 0.22 seconds and projection at 0.03 seconds.
+Repeated new native processes do not gain the API owner's warm result reuse.
+Its subsequent [bounded codec implementation](native-graph-transport-study.md)
+now returns the complete 21,349-operation VM graph in about 3.79 MB, inside the
+unchanged 8 MiB wire budget. Selected definitions/CFG reads and decoded Evidence
+work through both real MCP adapters; all graph fields and ordinary C/calls match
+the prior complete golden. Same-lease reuse is about 67 ms after a 39.7-second
+fresh capture, with resource limits and no OOM. The [next question/oracle](savepoint-control-oracle.md)
+start from names/error messages and exercise the new condition/CFG coverage.
+Two [real-model pairs](savepoint-control-study.md), including a discovery-guided
+repeat, answer both local guards correctly without a whole-report fetch. Code
+Mode uses fewer MCP requests, but whole-task costs are mixed and a large capture
+time variation remains unexplained. Both clients already provide JavaScript
+orchestration. Source fidelity is not machine equivalence, provider parity or a
+general speedup. The [native switch experiment](switch-study.md) now exports and independently checks
+182 labels over those 149 targets, exposes selected case-to-edge relations and
+keeps them readable when a 2 MiB native output budget omits the full graph.
+Both real models now answer the selected-case/default question correctly using
+one retained capture and focused case/byte reads. Code Mode uses fewer tokens
+in this pair, but unequal native capture time and extra client polling confound
+the comparison. Whole-task costs and limits are in the same study; a general
+efficiency advantage remains unproved.

experiments/atomic-codemode/api.mjs

@@ -0,0 +1,420 @@
+// One named contract registry shared by SDK, CLI, direct MCP and Code Mode.
+import { z } from "zod";
+import { KunaFactOwner } from "./provider.mjs";
+import { FactFailure } from "./fact-error.mjs";
+import { MAX_BYTE_BATCH_BYTES } from "./limits.mjs";
+import { createReference } from "./reference.mjs";
+
+const hex = z.string().regex(/^0x(?:0|[1-9a-f][0-9a-f]{0,15})$/u);
+const snapshot = z.string().startsWith("snap_");
+const byteWindow = z.strictObject({
+  address: hex,
+  length: z.number().int().min(1).max(MAX_BYTE_BATCH_BYTES).default(256),
+});
+const page = {
+  offset: z.number().int().nonnegative().default(0),
+  limit: z.number().int().min(1).max(128).default(20),
+};
+const kinds = z.enum([
+  "value",
+  "load",
+  "store",
+  "phi",
+  "call",
+  "branch",
+  "switch",
+  "return",
+]);
+const fields = z
+  .array(
+    z.enum([
+      "kind",
+      "statement",
+      "definitions",
+      "constants",
+      "opcode",
+      "inputs",
+      "output",
+      "is_dead",
+      "input_count",
+      "definition_count",
+      "constant_count",
+    ]),
+  )
+  .optional();
+const definition = (description, input, method, example) => ({
+  description,
+  input: z.strictObject(input),
+  method,
+  example,
+});
+const operationSeeds = z.array(z.string()).min(1).max(128);
+const relationInput = z.discriminatedUnion("kind", [
+  z.strictObject({
+    snapshot,
+    kind: z.literal("cases"),
+    seeds: operationSeeds,
+    case_values: z.array(hex).max(128).optional(),
+    case_indices: z.array(z.number().int().nonnegative()).max(128).optional(),
+    ...page,
+  }),
+  z.strictObject({
+    snapshot,
+    kind: z.literal("operands"),
+    seeds: operationSeeds,
+    input_indices: z.array(z.number().int().nonnegative()).max(128).optional(),
+    ...page,
+  }),
+  z.strictObject({
+    snapshot,
+    kind: z.enum(["definitions", "uses"]),
+    seeds: operationSeeds,
+    ...page,
+  }),
+  z.strictObject({
+    snapshot,
+    kind: z.enum(["incoming", "outgoing", "controllers"]),
+    seeds: z.array(z.number().int().nonnegative()).min(1).max(128),
+    ...page,
+  }),
+]);
+export const methods = {
+  switches: definition(
+    "Inspect recovered indirect branches and their independent native jump-table metadata, without returning all case rows. Selector is the current recovered BRANCHIND input, which may be normalized rather than the machine jump address. Native consumed mask and signed rendering hint are not complete runtime/signedness proofs. Null tables/defaults stay unknown. The native model default out-edge is separate from any machine out-of-range guard destination. Use relations kind:cases for selected raw table rows, labels and their native mapping to recovered edges; row/edge ordinals are not case values. Available even when the full graph is unavailable if this independent facet was captured.",
+    {
+      snapshot,
+      ids: z.array(z.string()).max(128).optional(),
+      origin_addresses: z.array(hex).max(128).optional(),
+      fields: z.array(z.enum(["selector", "table"])).optional(),
+      ...page,
+    },
+    "switches",
+    'await rea.switches({snapshot:s,origin_addresses:["0x1000"]})',
+  ),
+  functions: definition(
+    "Search retained native callable entries by literal name/alias, exact name, or entry addresses. Names can match several entries, including PLT/import stubs and implementations; preserve all matches and select by address. A name does not establish runtime symbol binding.",
+    {
+      query: z.string().default(""),
+      exact: z.boolean().default(false),
+      case_sensitive: z.boolean().default(true),
+      addresses: z.array(hex).max(128).optional(),
+      fields: z.array(z.enum(["name", "aliases", "size"])).optional(),
+      ...page,
+    },
+    "functions",
+    'await rea.functions({query:"format_error",exact:true})',
+  ),
+  strings: definition(
+    "Search retained string inventory by literal text. Paging does not establish artifact completeness.",
+    { query: z.string(), ...page },
+    "strings",
+    'await rea.strings({query:"incorrect header check"})',
+  ),
+  references: definition(
+    "Batch incoming or outgoing lifted-code references without decompilation. Address scope selects exact target (incoming) or source instruction (outgoing); function scope selects outgoing native owning-function references and requires an entry address. Unknown function applicability stays unknown. Kind filtering is pushed to the native query; source rows, endpoints and Evidence identity remain. Recovered IR calls are a separate representation and may omit machine calls. Interior string references require a separate extent query.",
+    {
+      addresses: z.array(hex).min(1).max(16),
+      direction: z.enum(["incoming", "outgoing"]).default("incoming"),
+      scope: z.enum(["address", "function"]).default("address"),
+      kinds: z
+        .array(z.enum(["call", "jump", "data", "read", "write"]))
+        .min(1)
+        .optional(),
+      fields: z
+        .array(z.enum(["fromFunction", "toFunction", "instruction"]))
+        .optional(),
+      ...page,
+    },
+    "references",
+    'await rea.references({addresses:["0x2000"]})',
+  ),
+  snapshot: definition(
+    "Analyze an explicit function entry once; return a retained capture handle, facets and coverage. Reuses the source within this target lease. Code, recovered calls and values remain independently available when full recovered IR exceeds capacity. A bounded call-rooted operation/definition projection may also be available; inspect facets.operations and coverage before reads. CFG queries still require the complete recovered graph. Available calls include compact by_opcode counts of known/unknown targets over the whole retained call facet; no paging is needed for these counts. Selected calls and complete Evidence remain readable.",
+    { entry: hex },
+    "snapshot",
+    'await rea.snapshot({entry:"0x1000"})',
+  ),
+  bytes: definition(
+    "Read selected native loaded-image bytes without instruction decoding. Batch up to 16 exact virtual-address windows and 64 KiB total selected bytes; hex uses twice that space before provenance. Short reads remain partial. These are not live process memory or an original-file-offset mapping. Global code discovery is not performed.",
+    {
+      windows: z.array(byteWindow).min(1).max(16),
+    },
+    "bytes",
+    'await rea.bytes({windows:[{address:"0x2000",length:32}]})',
+  ),
+  instructions: definition(
+    "Batch selected machine instruction windows. Read exact bytes and native disassembly to verify actual control transfers; optimized recovered CFG edges can bypass machine blocks. Address+size with the artifact context identifies the byte range. A caller-selected decode start does not prove execution/alignment.",
+    {
+      windows: z
+        .array(
+          z.strictObject({
+            address: hex,
+            count: z.number().int().min(1).max(4096).default(16),
+          }),
+        )
+        .min(1)
+        .max(16),
+      fields: z
+        .array(z.enum(["bytes", "text", "mnemonic", "operands"]))
+        .optional(),
+    },
+    "instructions",
+    'await rea.instructions({windows:[{address:"0x1000",count:8}],fields:["bytes","text"]})',
+  ),
+  operations: definition(
+    "Select SSA operations by IDs, source instruction origins, block IDs, constant literal or kind. Native default rows are compact summaries: opcode/kind/output and input/definition/constant counts; request detailed fields explicitly or page input slots via relations kind:operands. Filters intersect; identities and native is_dead are retained. Availability may be a complete recovered graph or a bounded call-rooted projection; inspect coverage before inferring absence. A literal match is not object identity. origin_addresses selects optimized recovered IR origins; use instructions for actual machine control transfers.",
+    {
+      snapshot,
+      ids: z.array(z.string()).max(128).optional(),
+      origin_addresses: z.array(hex).max(128).optional(),
+      block_ids: z.array(z.number().int().nonnegative()).max(128).optional(),
+      constant: hex.optional(),
+      kind: kinds.optional(),
+      fields,
+      ...page,
+    },
+    "operations",
+    'await rea.operations({snapshot:s,constant:"0x2000",fields:["statement"]})',
+  ),
+  relations: {
+    description:
+      "Read native operands, definitions/uses or case rows for operation IDs; incoming/outgoing CFG edges or derived controllers for block IDs. cases pages one relation per case/out-edge, with a stable case_id; an unmapped case has one explicit unknown-edge row. total counts relations, not table rows. cases filters optional case_values (native labels tagged value) and case_indices (original row ordinals); both filters intersect and paging spans deduplicated seeds in caller order. Rows preserve missing/no-label status, original recovered target, and separate recovered edges. Partial/unavailable tables and unmapped rows are explicit; no runtime feasibility claim. operands pages every selected input slot, including constants, function inputs and null slots; input_indices optionally selects slots in caller order. Identity is snapshot + operation + input_index; value_id/definition and exact phi predecessor/edge bindings remain separate. Out-of-range slots of retained operations are absent; unretained_seeds remain unknown. A bounded call-rooted projection supports operands/definitions/uses but no CFG; uses cover retained selected operations only. Diagnostic SSA lacks native operands and has partial CFG/printed occurrences. controllers requires a complete recovered graph and concerns finite paths to recovered exits, not runtime feasibility.",
+    input: relationInput,
+    method: "relations",
+    example:
+      'await rea.relations({snapshot:s,kind:"operands",seeds:[op.id],input_indices:[0,2]})',
+  },
+  values: definition(
+    "Read identified native SSA values in a batch; scope may be a full graph, bounded call-rooted definition projection, or call-boundary values only; is_input filters retained operand/output values, not an original source signature. Offset is an address-space offset, not a computed arbitrary-width value. Width, constants, function inputs, annotation status and defining operation are separate facts. Requires the pinned native exporter.",
+    {
+      snapshot,
+      ids: z.array(z.string()).max(128).optional(),
+      is_input: z.boolean().optional(),
+      fields: z
+        .array(
+          z.enum([
+            "space",
+            "offset",
+            "size_bytes",
+            "is_constant",
+            "is_input",
+            "is_annotation",
+            "definition",
+            "type_name",
+          ]),
+        )
+        .optional(),
+      ...page,
+    },
+    "values",
+    "await rea.values({snapshot:s,ids:op.inputs.map(i=>i.value_id).filter(Boolean)})",
+  ),
+  calls: definition(
+    "Select structured recovered call sites and argument bindings. Target comes from native call specifications, never the fspec annotation offset or display text. Each argument separates its SSA value from its passing storage. Calls and their boundary values survive full graph capacity failure. target_value_id retains input zero independently of target resolution; it may be an fspec annotation or an unresolved indirect target value. A null output describes this recovered operation, not proof of a void source signature. Input slots are not original source parameter ordinals; missing storage and indirect/runtime targets remain unknown. To investigate across functions, snapshot the selected callee and select values with is_input:true, comparing exact space/offset/width. Requires the pinned native exporter.",
+    {
+      snapshot,
+      ids: z.array(z.string()).max(128).optional(),
+      origin_addresses: z.array(hex).max(128).optional(),
+      fields: z
+        .array(
+          z.enum([
+            "target",
+            "target_name",
+            "arguments",
+            "output",
+            "prototype",
+            "stack_offset",
+            "target_value_id",
+            "is_dead",
+          ]),
+        )
+        .optional(),
+      ...page,
+    },
+    "calls",
+    'await rea.calls({snapshot:s,origin_addresses:["0x1100"]})',
+  ),
+  blocks: definition(
+    "Read native recovered blocks by stable snapshot-local ID, with ordered predecessors, successor edges and member operation IDs. origin_start describes optimized IR provenance, not a machine block extent. Requires an available native recovered IR graph. Runtime successors remain unknown.",
+    {
+      snapshot,
+      ids: z.array(z.number().int().nonnegative()).max(128).optional(),
+      fields: z
+        .array(
+          z.enum(["origin_start", "operations", "predecessors", "successors"]),
+        )
+        .optional(),
+      ...page,
+    },
+    "blocks",
+    "await rea.blocks({snapshot:s,ids:[op.block]})",
+  ),
+  slice: definition(
+    "Compose a bounded backward definition traversal from explicit operation IDs. Native default operation rows are compact summaries; request detailed fields or use paged operand relations for wide nodes. Available over a complete recovered graph or bounded call-rooted projection. Missing defining bodies retain frontier reasons; complete use sets and CFG are unavailable for a projection. Default load/phi cutoffs keep boundaries visible; choose stop_kinds:[] to traverse beyond them. Complete reachability and memory aliases remain unknown.",
+    {
+      snapshot,
+      seeds: z.array(z.string()).min(1).max(64),
+      max_operations: z.number().int().min(1).max(256).default(48),
+      stop_at: z.array(z.string()).max(128).default([]),
+      stop_kinds: z.array(kinds).default(["phi", "load"]),
+      fields,
+    },
+    "slice",
+    'await rea.slice({snapshot:s,seeds:callIds,stop_kinds:["load"],fields:["opcode","inputs","output","constants"]})',
+  ),
+  code: definition(
+    "Read reconstructed code from the same retained function capture, even when its graph facet is unavailable. offset is zero-based. A large limit reads the complete code inside Code Mode for local filtering. The bridge byte budget bounds the result. No additional decompilation.",
+    {
+      snapshot,
+      offset: page.offset,
+      limit: z.number().int().positive().default(64),
+    },
+    "code",
+    "await rea.code({snapshot:s,offset:100,limit:64})",
+  ),
+  evidence: definition(
+    "Read complete retained native source in UTF-16-code-unit pages by evidence_id; no reanalysis. component:source preserves exact original wire JSON. component:recovered_ir reads the decoded graph source, including lossless compressed graphs; source hash remains the original capture and view has its own hash/byte count. This is source inspection, not a claim that malformed graph facts are valid. Available until close_binary.",
+    {
+      evidence_id: z.string().startsWith("ev_"),
+      component: z.enum(["source", "recovered_ir"]).default("source"),
+      offset: page.offset,
+      limit: z.number().int().min(1).max(16384).default(4096),
+    },
+    "evidence",
+    "await rea.evidence({evidence_id:id,offset:0,limit:4096})",
+  ),
+};
+
+methods.bytes.input = methods.bytes.input.superRefine((value, context) => {
+  // Pattern failures need not prevent object refinements from running in Zod.
+  // Only the named window contract may admit values to exact u64 arithmetic.
+  const windows = value.windows.flatMap((window) => {
+    const parsed = byteWindow.safeParse(window);
+    return parsed.success ? [parsed.data] : [];
+  });
+  if (
+    windows.reduce((sum, window) => sum + window.length, 0) >
+    MAX_BYTE_BATCH_BYTES
+  )
+    context.addIssue({
+      code: "custom",
+      path: ["windows"],
+      message:
+        "Select at most 64 KiB of bytes per batch; use additional windows/pages in later calls",
+    });
+  if (
+    windows.some(
+      (window) =>
+        BigInt(window.address) + BigInt(window.length) > 0xffffffffffffffffn,
+    )
+  )
+    context.addIssue({
+      code: "custom",
+      path: ["windows"],
+      message: "Byte range end exceeds the native u64 address format",
+    });
+});
+methods.references.input = methods.references.input.superRefine(
+  (value, context) => {
+    if (value.direction === "incoming" && value.scope !== "address")
+      context.addIssue({
+        code: "custom",
+        path: ["scope"],
+        message: "Incoming references require exact address scope",
+      });
+  },
+);
+export class AtomicSession {
+  #owner;
+  #tail = Promise.resolve();
+  constructor(factory = KunaFactOwner.open) {
+    this.factory = factory;
+  }
+  #admit(work) {
+    const pending = this.#tail.then(work, work);
+    this.#tail = pending.catch(() => {});
+    return pending;
+  }
+  get context() {
+    return this.#owner?.context ?? null;
+  }
+  abort() {
+    this.#owner?.cancel();
+  }
+  open(path) {
+    return this.#admit(async () => {
+      if (this.#owner) await this.#owner.close();
+      this.#owner = undefined;
+      this.#owner = await this.factory(path);
+      return {
+        ok: true,
+        context: this.context,
+        methods: Object.keys(methods),
+        evidence_lifetime: "current target lease",
+      };
+    });
+  }
+  close() {
+    return this.#admit(async () => {
+      if (this.#owner) await this.#owner.close();
+      this.#owner = undefined;
+      return { ok: true, closed: true };
+    });
+  }
+  call(
+    name,
+    input,
+    signal = new AbortController().signal,
+    lease = this.context?.lease,
+  ) {
+    return this.#admit(async () => {
+      try {
+        if (signal.aborted)
+          throw new FactFailure(
+            "cancelled",
+            "Request was cancelled before admission",
+          );
+        if (!this.#owner)
+          throw new FactFailure(
+            "no_target",
+            "Call open_binary before analysis",
+          );
+        if (lease !== this.context.lease)
+          throw new FactFailure(
+            "stale_context",
+            "Target lease changed; do not join facts across leases",
+          );
+        if (!Object.hasOwn(methods, name))
+          throw new FactFailure(
+            "unknown_method",
+            `Unknown API method: ${name}`,
+          );
+        const contract = methods[name];
+        const parsed = contract.input.parse(input);
+        const data = await this.#owner[contract.method](parsed, signal);
+        if (signal.aborted)
+          throw new FactFailure(
+            "cancelled",
+            "Request was cancelled after observations were collected",
+            { partial_observation: data },
+          );
+        return { ok: true, context: this.context, data };
+      } catch (error) {
+        return {
+          ok: false,
+          context: this.context,
+          error: {
+            code:
+              error instanceof z.ZodError
+                ? "invalid_input"
+                : (error.code ?? "analysis_failed"),
+            message: error.message,
+            details: error.details ?? {},
+          },
+        };
+      }
+    });
+  }
+}
+
+export const reference = createReference(methods);

experiments/atomic-codemode/api.test.mjs

@@ -0,0 +1,145 @@
+import assert from "node:assert/strict";
+import test from "node:test";
+import Ajv2020 from "ajv/dist/2020.js";
+import { AtomicSession, methods, reference } from "./api.mjs";
+
+const factory = () => {
+  let sequence = 0;
+  const calls = [];
+  const create = async (path) => ({
+    context: { lease: `lease-${++sequence}`, selected_path: path },
+    close: async () => {
+      calls.push("close");
+    },
+    cancel: () => {
+      calls.push("cancel");
+    },
+    strings: async (input) => {
+      calls.push(input);
+      return { items: [], source: { evidence_id: "ev_real-owner" } };
+    },
+  });
+  return { create, calls };
+};
+test("relation identity and operand-only selectors agree in runtime and advertised schemas", () => {
+  const validate = new Ajv2020({ strict: false }).compile(
+    reference("relations").methods[0].input_schema,
+  );
+  for (const [request, valid] of [
+    [{ kind: "operands", seeds: ["0x1:0"], input_indices: [0, 155] }, true],
+    [{ kind: "operands", seeds: ["0x1:0"], input_indices: [] }, true],
+    [{ kind: "uses", seeds: ["0x1:0"] }, true],
+    [
+      {
+        kind: "cases",
+        seeds: ["0x1:0"],
+        case_values: ["0xffffffffffffffff"],
+        case_indices: [0, 155],
+      },
+      true,
+    ],
+    [{ kind: "cases", seeds: [0] }, false],
+    [{ kind: "cases", seeds: ["0x1:0"], case_values: ["0X1"] }, false],
+    [{ kind: "incoming", seeds: [0] }, true],
+    [{ kind: "incoming", seeds: ["0"] }, false],
+    [{ kind: "operands", seeds: [0] }, false],
+    [{ kind: "uses", seeds: ["0x1:0"], input_indices: [0] }, false],
+    [{ kind: "operands", seeds: ["0x1:0"], input_indices: [-1] }, false],
+  ]) {
+    const input = { snapshot: "snap_test", ...request };
+    assert.equal(methods.relations.input.safeParse(input).success, valid);
+    assert.equal(validate(input), valid, JSON.stringify(validate.errors));
+  }
+});
+test("named contracts reject extra keys and stale leases before provider admission", async () => {
+  const provider = factory();
+  const session = new AtomicSession(provider.create);
+  await session.open("first");
+  const lease = session.context.lease;
+  const invalid = await session.call("strings", { query: "x", guessed: true });
+  assert.equal(invalid.error.code, "invalid_input");
+  const wrongIdentity = await session.call("relations", {
+    snapshot: "snap_test",
+    kind: "incoming",
+    seeds: ["0"],
+  });
+  assert.equal(wrongIdentity.error.code, "invalid_input");
+  assert.equal(provider.calls.length, 0);
+  await session.open("second");
+  const stale = await session.call("strings", { query: "x" }, undefined, lease);
+  assert.equal(stale.error.code, "stale_context");
+  assert.deepEqual(provider.calls, ["close"]);
+  await session.close();
+});
+
+test("cancellation after a completed read retains observations under a cancelled outcome", async () => {
+  const controller = new AbortController();
+  const session = new AtomicSession(async () => ({
+    context: { lease: "late-cancel" },
+    close: async () => {},
+    strings: async () => {
+      controller.abort();
+      return {
+        source: { evidence_id: "ev_observed" },
+        items: [{ value: "retained" }],
+      };
+    },
+  }));
+  await session.open("target");
+  const result = await session.call(
+    "strings",
+    { query: "x" },
+    controller.signal,
+  );
+  assert.equal(result.ok, false);
+  assert.equal(result.error.code, "cancelled");
+  assert.equal(
+    result.error.details.partial_observation.items[0].value,
+    "retained",
+  );
+  assert.equal(result.context.lease, "late-cancel");
+  await session.close();
+});
+test("close drains admitted reads and a failed cleanup retains ownership for retry", async () => {
+  let release;
+  let reads = 0;
+  let closes = 0;
+  const session = new AtomicSession(async () => ({
+    context: { lease: "owned" },
+    strings: () => {
+      reads++;
+      return new Promise((resolve) => {
+        release = resolve;
+      });
+    },
+    close: () => {
+      if (++closes === 1) throw new Error("cleanup failed");
+    },
+    cancel() {},
+  }));
+  await session.open("target");
+  const read = session.call("strings", { query: "x" });
+  await new Promise((resolve) => setImmediate(resolve));
+  const close = session.close();
+  assert.equal(closes, 0);
+  assert.equal(reads, 1);
+  release({ items: [] });
+  assert.equal((await read).ok, true);
+  await assert.rejects(close, /cleanup failed/u);
+  assert.equal(session.context.lease, "owned");
+  await session.close();
+  assert.equal(session.context, null);
+});
+test("cancellation before admission makes no provider query; advertised reference schemas compile", async () => {
+  const provider = factory();
+  const session = new AtomicSession(provider.create);
+  await session.open("target");
+  const stop = new AbortController();
+  stop.abort();
+  const cancelled = await session.call("strings", { query: "x" }, stop.signal);
+  assert.equal(cancelled.error.code, "cancelled");
+  assert.deepEqual(provider.calls, []);
+  const ajv = new Ajv2020({ strict: false });
+  for (const method of reference().methods) ajv.compile(method.input_schema);
+  await session.close();
+});

experiments/atomic-codemode/boundary-oracle.md

@@ -0,0 +1,53 @@
+# Calls and values surviving full-graph capacity
+
+The question is `sqlite-boundary-question.txt`. Fix it and this acceptance oracle
+before either real-model run. Both lanes use the same complete fourteen-method
+API, preserved engine, model/effort request, common guidance and resource limits.
+This task requires VM decompilation once per fresh lease; it is different work
+from the earlier shallow-reference question. Do not compare their task times as
+an adapter speedup. Actual client JavaScript dispatch is available in both lanes.
+
+Artifact: Debian `libsqlite3-0:amd64` version `3.40.1-2+deb12u2`,
+`libsqlite3.so.0.8.6`, SHA-256
+`2e6eef9a727f081f0d453b4e5e6cbd8b9ef8b6f86cbf7681cbad444d3b0b55c8`.
+The selected VM implementation entry is `0xea6d0`, not its same-name PLT entry.
+
+Pre-model public MCP capture with the new exporter establishes:
+
+- Code is available. Full graph is unavailable with `atomic operation capacity
+  exceeded`. Calls remain available: 416 retained block-member operations,
+  consisting of 395 CALL and 21 CALLIND. The 395 direct targets are known, all
+  21 indirect targets remain unknown. Call-boundary values are available (1,294),
+  covering targets, arguments and outputs only, with 21,349 operations scanned
+  over 1,229 recovered blocks. These do not supply a complete definition/use graph.
+- At `0xf115c`, recovered call ID `0xf115c:6a84` targets `ram:0x296c0`.
+  Input zero is `v1000118d6`, `fspec:0x102f4`, annotation true, width 8.
+  Three recovered inputs use passing storage `register:0x38/8`, `0x30/8`,
+  `0x10/8` (RDI, RSI, RDX in this verified x86-64 lane).
+- At `0xf238b`, call ID `0xf238b:7130` has the same resolved target.
+  Input zero is `v100012a80`, `fspec:0x1030c`, annotation true, width 8.
+  Two recovered inputs use passing storage `register:0x38/8`, `0x30/8`.
+- Both first arguments are the same snapshot-local SSA value `v30000004b`,
+  `register:0x38/8`, `is_input:true`, definition null. Other retained argument
+  values have defining operation IDs outside this boundary view. Exact SSA
+  identity is supported; source parameter ordinals and general aliases are not.
+
+Independent GNU objdump decoding from the exported VM entry verifies the two
+machine CALL targets and register setup. The earlier complete direct-call pair
+oracle contains both. Of the 396 physical direct immediate calls, the only one
+absent from the recovered 395 is `0xf2493 -> 0x29370` (stack check failure).
+The question does not require rediscovering this extra call; the recovered-call
+count must not be claimed as complete machine/runtime coverage.
+
+Required final facts: correct selected entry and facet availability/scope; both
+call-count groups and target-unknown distinction; both resolved targets separated
+from annotation offsets; argument slots/locations; shared first-input identity;
+selected machine checks; explicit unreadable definition/use and alias/runtime
+limits; real Evidence and subject digest. Do not require the model to report
+every additional native metadata counter. Check that its API source arguments
+show only one VM analysis and all relevant reads reuse its snapshot/Evidence.
+
+Raw public verification is retained as `facets-v3-mcp.json` and its Evidence
+directory under `.cache/atomic-codemode/`. The ordinary VM record (C, mappings,
+variables and other ordinary metadata) equals the prior captured engine output.
+The independent oracle and precomputed answer are absent from the model workspace.

experiments/atomic-codemode/byte-facts.mjs

@@ -0,0 +1,103 @@
+// Native loaded-image data is not instruction decoding or original-file mapping.
+import { z } from "zod";
+import { FactFailure } from "./fact-error.mjs";
+import { MAX_BYTE_BATCH_BYTES, NATIVE_OUTPUT_BYTES } from "./limits.mjs";
+
+const hex = z.string().regex(/^0x(?:0|[1-9a-f][0-9a-f]{0,15})$/u);
+const payload = z.string().regex(/^(?:[0-9a-f]{2})*$/u);
+const header = z.object({
+  binary: z.string(),
+  kind: z.literal("data"),
+  start_hex: hex,
+  end_hex: hex,
+  count: z
+    .number()
+    .int()
+    .nonnegative()
+    .max(MAX_BYTE_BATCH_BYTES / 16),
+  bytes: z.number().int().nonnegative().max(MAX_BYTE_BATCH_BYTES),
+  hex: payload,
+  truncated: z.boolean(),
+  notes: z.array(z.string()),
+});
+const row = z.object({
+  address_hex: hex,
+  size: z.number().int().min(1).max(16),
+  bytes: payload,
+});
+const requireFact = (valid, message) => {
+  if (!valid) throw new FactFailure("invalid_provider_output", message);
+};
+
+export function parseByteFacts(text, inputPath, request) {
+  try {
+    requireFact(
+      Buffer.byteLength(text) <= NATIVE_OUTPUT_BYTES,
+      "Byte source exceeds capture capacity",
+    );
+    const raw = JSON.parse(text);
+    requireFact(
+      Array.isArray(raw?.rows) && raw.rows.length <= MAX_BYTE_BATCH_BYTES / 16,
+      "Byte rows exceed capacity or have invalid shape",
+    );
+    const doc = header.parse(raw);
+    requireFact(
+      doc.binary === inputPath && doc.start_hex === request.address,
+      "Byte source names a different target/window",
+    );
+    requireFact(
+      doc.bytes <= request.length && doc.count === raw.rows.length,
+      "Byte source exceeds the request or disagrees with row count",
+    );
+    let end = BigInt(request.address);
+    const pieces = raw.rows.map((value, index) => {
+      const item = row.parse(value);
+      requireFact(
+        BigInt(item.address_hex) === end && item.bytes.length === item.size * 2,
+        "Byte row address/length is inconsistent",
+      );
+      requireFact(
+        index === raw.rows.length - 1 || item.size === 16,
+        "Nonfinal byte row is not the declared native width",
+      );
+      end += BigInt(item.size);
+      return item.bytes;
+    });
+    requireFact(
+      end === BigInt(doc.end_hex) &&
+        end - BigInt(request.address) === BigInt(doc.bytes) &&
+        pieces.join("") === doc.hex &&
+        doc.hex.length === doc.bytes * 2,
+      "Byte rows, payload and extent disagree",
+    );
+    return {
+      hex: doc.hex,
+      returned_bytes: doc.bytes,
+      end_address: doc.end_hex,
+      coverage: {
+        authority: "native loaded-image bytes",
+        requested_bytes: request.length,
+        status:
+          doc.bytes === request.length
+            ? "complete-for-requested-range"
+            : "partial",
+        provider_truncated: doc.truncated,
+        notes: doc.notes,
+        original_file_mapping: "unknown",
+        runtime_memory: "not-observed",
+      },
+    };
+  } catch (error) {
+    if (error instanceof FactFailure) throw error;
+    throw new FactFailure(
+      "invalid_provider_output",
+      "Native byte window does not match the declared format",
+      {
+        reason:
+          error instanceof z.ZodError
+            ? error.issues[0]?.message.slice(0, 256)
+            : String(error.message).slice(0, 256),
+      },
+    );
+  }
+}

experiments/atomic-codemode/byte-facts.test.mjs

@@ -0,0 +1,104 @@
+import test from "node:test";
+import assert from "node:assert/strict";
+import { parseByteFacts } from "./byte-facts.mjs";
+import { AtomicSession } from "./api.mjs";
+
+const make = () => ({
+  binary: "/input",
+  kind: "data",
+  start_hex: "0x20000000000001",
+  end_hex: "0x20000000000005",
+  count: 1,
+  bytes: 4,
+  hex: "00417fff",
+  truncated: false,
+  notes: [],
+  rows: [
+    {
+      address_hex: "0x20000000000001",
+      size: 4,
+      bytes: "00417fff",
+    },
+  ],
+});
+const request = { address: "0x20000000000001", length: 4 };
+test("byte facts keep exact u64 extents, null/nontext bytes and loaded-image authority", () => {
+  const facts = parseByteFacts(JSON.stringify(make()), "/input", request);
+  assert.equal(facts.hex, "00417fff");
+  assert.equal(facts.end_address, "0x20000000000005");
+  assert.equal(facts.coverage.status, "complete-for-requested-range");
+  assert.equal(facts.coverage.original_file_mapping, "unknown");
+  assert.equal(facts.coverage.runtime_memory, "not-observed");
+  const clipped = parseByteFacts(JSON.stringify(make()), "/input", {
+    ...request,
+    length: 5,
+  });
+  assert.equal(clipped.coverage.status, "partial");
+  const flagged = make();
+  flagged.truncated = true;
+  assert.equal(
+    parseByteFacts(JSON.stringify(flagged), "/input", request).coverage.status,
+    "complete-for-requested-range",
+  );
+});
+
+test("wrong byte subject, contradictory duplicated payload, gaps and counts fail before publishing facts", () => {
+  for (const mutate of [
+    (d) => (d.binary = "/other"),
+    (d) => (d.kind = "code"),
+    (d) => (d.hex = "00417f00"),
+    (d) => (d.rows[0].address_hex = "0x20000000000000"),
+    (d) => (d.rows[0].size = 3),
+    (d) => (d.end_hex = "0x20000000000006"),
+    (d) => (d.count = 2),
+    (d) => (d.rows = Array(4097).fill(null)),
+  ]) {
+    const document = make();
+    mutate(document);
+    assert.throws(
+      () => parseByteFacts(JSON.stringify(document), "/input", request),
+      { code: "invalid_provider_output" },
+    );
+  }
+  const empty = make();
+  Object.assign(empty, {
+    hex: "",
+    rows: [],
+    count: 0,
+    bytes: 0,
+    end_hex: request.address,
+  });
+  assert.equal(
+    parseByteFacts(JSON.stringify(empty), "/input", request).coverage.status,
+    "partial",
+  );
+});
+
+test("byte input capacity, malformed addresses and u64 end overflow reject before native admission", async () => {
+  let reads = 0;
+  const session = new AtomicSession(async () => ({
+    context: { lease: "bytes" },
+    bytes: () => {
+      reads += 1;
+    },
+    close: () => {},
+  }));
+  await session.open("/input");
+  try {
+    for (const windows of [
+      [{ address: "invalid", length: 1 }],
+      [{ address: "0xffffffffffffffff", length: 1 }],
+      [
+        { address: "0x1000", length: 65536 },
+        { address: "0x2000", length: 1 },
+      ],
+      [{ address: "0x1000", length: 0 }],
+    ]) {
+      const result = await session.call("bytes", { windows });
+      assert.equal(result.error.code, "invalid_input", JSON.stringify(result));
+    }
+    assert.equal(reads, 0);
+  } finally {
+    await session.close();
+  }
+});

experiments/atomic-codemode/call-oracle.md

@@ -0,0 +1,64 @@
+# Cross-function context experiment: pinned question and oracle
+
+This is an experiment acceptance oracle, unavailable to the isolated model.
+The question is `sqlite-call-question.txt`. Fix this question and the required
+facts before launching either model lane. Both adapters expose the same complete
+fourteen-method fact API; Code Mode has four MCP tools, direct has seventeen.
+Use the same pinned engine, artifact, model request, reasoning effort, common
+envelope guidance, fresh target leases and resource limits. Do not change runtime
+sources or HEAD during the model batch. One sample per lane cannot establish a
+general advantage, and the client itself already offers JavaScript dispatch in
+both lanes. This comparison isolates the server calling adapter, not all forms of
+composition or a traditional provider workflow.
+
+The artifact is the local Debian ELF `libsqlite3.so.0.8.6`, SHA-256
+`2e6eef9a727f081f0d453b4e5e6cbd8b9ef8b6f86cbf7681cbad444d3b0b55c8`.
+Independent GNU objdump decoding from the function entries covers `0x64060`
+through `0x65400`; existing earlier captures independently pin the same flow.
+No target execution or source-code answer is used.
+
+Required facts for an equal completed answer:
+
+- Wrapper entry `0x65180`; direct machine CALL `0x65211` targets internal helper
+  `0x64060`. The recovered call operation has an fspec annotation as input zero;
+  that annotation's offset is not the target address.
+- Caller LEA `0x65201` passes post-prologue `RSP+0x20` in RDI. Structured passing
+  storage is `register:0x38`, width 8. The SSA argument and the callee's retained
+  input are separate snapshot-local identities; matching their storage supplies a
+  candidate connection. The direct machine transfer and pointer setup support
+  this particular connection. Do not globally infer source parameter ordinal,
+  alias completeness or runtime binding from a slot number or a name.
+- The helper preserves the input context pointer in RBP. Its 32-bit field at
+  context offset `0x30` is incremented at `0x64693`/`0x64696`/`0x64699` for an
+  opening parenthesis and decremented at `0x64c92` for a closing parenthesis.
+  The parser's token/branch context must support the parenthesis interpretation;
+  an ADD/SUB instruction alone does not identify the meaning of a field.
+- In the wrapper this is `[RSP+0x50]`, initialized to zero with the other context
+  tail fields. After helper return zero, LOAD `0x65268` and TEST/JZ
+  `0x65270`/`0x65272` check it. Zero skips the malformed block; nonzero frees the
+  expression at `0x65274`, clears its output pointer at `0x65279`, then reaches
+  the malformed diagnostic at `0x65281` through formatter call `0x65290`.
+- Return-code-one errors can also reach the malformed block, via
+  `0x65239`/`0x6523c`; return code 18 selects a separate depth-limit diagnostic
+  at `0x65330`. No demonstrated runtime-feasibility or full memory-alias claim.
+
+The producer verifier additionally checks two formatter call targets:
+`0x5f031 -> 0x2adc0` (free PLT) and `0x5f062 -> 0x27e50` (vmprintf PLT), whereas
+their input-zero fspec offsets are `0x10000`/`0x10001`. Resolving these PLT entries
+to runtime implementations is outside this task.
+
+Native producer fidelity is checked against previously retained captures of all
+three functions: C, ordinary metadata/line mappings, blocks, operations, values
+and coverage are identical. Only schema version and structured call records are
+added. Older sources used pretty JSON; their smaller/newer encoded sizes must
+not be attributed to the call exporter. The earlier compact-writer change is a
+separate experiment. Unsupported old schema graphs remain unavailable with valid
+C retained, rather than acquiring guessed call records.
+
+The first paired question runs at `01e5c5955` both explain the primary state flow
+correctly but omit the oracle's separate return-1/return-18 distinctions. Because
+the initial question did not explicitly enumerate those codes, the next round
+clarifies that requirement in `sqlite-call-question.txt` before either lane
+starts. The required facts and independent oracle stay unchanged. Earlier
+questions and costs remain pinned to their original commits; neither incomplete
+sample is counted as a full-oracle success.

experiments/atomic-codemode/cli.mjs

@@ -0,0 +1,20 @@
+import { readFile } from "node:fs/promises";
+import { AtomicSession } from "./api.mjs";
+import { executeJavaScript } from "./executor.mjs";
+
+const [binary, script] = process.argv.slice(2);
+if (!binary || !script)
+  throw new Error("Usage: node cli.mjs BINARY JAVASCRIPT_FILE");
+const session = new AtomicSession();
+try {
+  await session.open(binary);
+  const result = await executeJavaScript(
+    session,
+    await readFile(script, "utf8"),
+  );
+  console.log(JSON.stringify(result));
+  process.exitCode = result.ok ? 0 : 1;
+} finally {
+  session.abort();
+  await session.close();
+}

experiments/atomic-codemode/client-wait-study.md

@@ -0,0 +1,51 @@
+# Client waiting and whole-task cost
+
+The [selected-case pair](switch-study.md) exposes one-second client polls after
+the snapshot execution yields. Direct makes six waits (three one-second then
+three ten-second requests); Code Mode makes three one-second requests. Five
+direct and two Code Mode wait responses still report a running cell. Initial
+execution yields after about 31 seconds in both lanes. This is observed client
+behavior, separate from MCP attempts and native source-capture time.
+
+Native latency remains variable, including in sequential scripted MCP runs
+without models. The [overlap canary](overlap-study.md) did not establish a cause.
+No analysis acceleration or cache invalidation change is proposed here.
+
+## Frozen diagnostic
+
+The opt-in runner flag `--guide-long-waits` appends exactly the same client wait
+policy in both lanes: resume an already-running cell with a 30,000 ms wait,
+respect the advertised schema, and do not sleep or duplicate analysis merely
+to check progress. It changes neither the initial execution yield policy nor
+the provider/Code Mode deadlines. It supplies no analysis facts. The policy,
+flag and instruction digest are recorded in each trial's settings. Older runs
+without this flag remain a separate stratum.
+
+Run Code Mode first, then direct, with the unchanged
+[question](sqlite-switch-question.txt) and [oracle](switch-oracle.md), final
+`a9ca8a8a...` engine, complete four/eighteen-tool catalogs and the same exact
+discovery/envelope guidance as the first pair. Use fresh leases and isolated
+model workspaces, requested `gpt-6.1-sol/high`, Codex 0.161.0 and the same one-CPU,
+4 GiB hard / 3 GiB early-stop resource policy. No runtime edits during the pair.
+Local records use `sqlite-switch-{codemode,direct}-v2*`.
+
+Review complete answers against the frozen independent machine oracle and
+retained source, including selector widths, all requested cases, signed target
+arithmetic, case/edge identities, native default versus machine out-of-range
+guard, provenance and unknown runtime behavior. Require one VM capture and
+focused reads; a lower token count without a correct answer is not a win.
+
+Record actual wait arguments and running/completed responses, model responses,
+all input/cached/uncached/output/reasoning tokens, MCP/API attempts, wall time,
+family CPU, kernel family peak, native source-capture time, identity/freshness
+and cleanup/OOM status. Count discovery, guidance and unsuccessful waits in
+whole-task usage. Fewer polls would establish adherence in this client, not
+causally lower tokens or native latency. A before/after pair remains subject to
+model choices, prompt caches, host contention and run order; do not subtract
+native or polling intervals to manufacture an adapter speedup.
+
+The [official Codex CLI reference](https://learn.chatgpt.com/docs/cli/reference)
+documents client configuration but does not establish the `wait` contract used
+here. The actual captured client calls and subsequent run establish behavior;
+no configuration feature flag is treated as proof of the available surface.
+The native/API contracts are unchanged. Results are pending.

experiments/atomic-codemode/control-dependence.mjs

@@ -0,0 +1,119 @@
+// Immediate postdominators on the reversed recovered CFG, using linear retained
+// state. Control dependence is derived for finite paths to recovered exits.
+import { FactFailure } from "./fact-error.mjs";
+export function controlDependencies(
+  blocks,
+  { workLimit = 2_000_000, dependents } = {},
+) {
+  const exit = Symbol("recovered-exit");
+  const successors = new Map(
+    blocks.map((block) => [
+      block.id,
+      block.successors.map((edge) => edge.target),
+    ]),
+  );
+  const reverse = new Map([
+    [exit, []],
+    ...blocks.map((block) => [block.id, []]),
+  ]);
+  const tick = (() => {
+    let work = 0;
+    return () => {
+      if (++work > workLimit)
+        throw new FactFailure(
+          "capacity",
+          "Control-dependence work budget exceeded",
+        );
+    };
+  })();
+  for (const [from, targets] of successors) {
+    if (!targets.length) targets.push(exit);
+    for (const to of targets) reverse.get(to).push(from);
+  }
+  const { parent, visited } = postdominatorParents(
+    exit,
+    successors,
+    reverse,
+    tick,
+  );
+  const relations = [];
+  for (const block of blocks) {
+    if (!parent.has(block.id)) continue;
+    for (const edge of block.successors) {
+      let runner = edge.target;
+      while (
+        parent.has(runner) &&
+        runner !== parent.get(block.id) &&
+        runner !== exit
+      ) {
+        tick();
+        if (!dependents || dependents.has(runner))
+          relations.push({ controller: block.id, dependent: runner, edge });
+        runner = parent.get(runner);
+      }
+    }
+  }
+  return {
+    relations,
+    coverage: {
+      scope:
+        "derived control dependence in the recovered graph; zero-successor blocks treated as exits; finite paths only",
+      status:
+        visited.size === blocks.length + 1
+          ? "complete-for-recovered-finite-exit-paths"
+          : "partial",
+      exit_unreachable_blocks: blocks
+        .filter((block) => !visited.has(block.id))
+        .map((block) => block.id),
+      nontermination: "unknown",
+      runtime_reachability: "unknown",
+    },
+  };
+}
+
+function postdominatorParents(exit, successors, reverse, tick) {
+  const visited = new Set([exit]);
+  const stack = [[exit, 0]],
+    postorder = [];
+  while (stack.length) {
+    tick();
+    const frame = stack.at(-1),
+      children = reverse.get(frame[0]);
+    if (frame[1] < children.length) {
+      const child = children[frame[1]++];
+      if (!visited.has(child)) {
+        visited.add(child);
+        stack.push([child, 0]);
+      }
+    } else {
+      postorder.push(frame[0]);
+      stack.pop();
+    }
+  }
+  const order = postorder.reverse();
+  const position = new Map(order.map((id, i) => [id, i]));
+  const parent = new Map([[exit, exit]]);
+  const intersect = (a, b) => {
+    while (a !== b) {
+      tick();
+      if (position.get(a) > position.get(b)) a = parent.get(a);
+      else b = parent.get(b);
+    }
+    return a;
+  };
+  let changed = true;
+  while (changed) {
+    changed = false;
+    for (const node of order.slice(1)) {
+      tick();
+      const known = successors.get(node).filter((id) => parent.has(id));
+      if (!known.length) continue;
+      const next = known.slice(1).reduce(intersect, known[0]);
+      if (parent.get(node) !== next) {
+        parent.set(node, next);
+        changed = true;
+      }
+    }
+  }
+  return { parent, visited };
+}

experiments/atomic-codemode/evaluation.md

@@ -0,0 +1,261 @@
+# Task coverage before cost comparison
+
+The callable projection now also has a completed [PHI-format model pair](trials.md).
+Both answers reconstruct all four alternatives with exact retained identities,
+independent machine/string checks and partial-scope limits. An observed reference
+lookup mismatch adds work in the direct lane; its costs remain recorded and that
+discovery boundary needs correction before another same-question comparison.
+
+Code Mode can only compose facts the underlying API actually supplies. A task
+cost is comparable after reviewing the completed answer and its evidence. A
+successful tool call, fewer response bytes or more graph records is insufficient.
+No claim of complete reverse-engineering capability follows from these tasks.
+
+The [boundary-facet question/oracle](boundary-oracle.md) now also has a completed
+real-model pair. Both answers correctly use calls and boundary values despite
+full-graph capacity failure, reuse one VM analysis and preserve definition/use
+and runtime uncertainty. Code Mode is faster in this sample and uses slightly
+less cumulative input, while uncached input and maximum response context are
+higher. See the full costs and independent source checks in [trials.md](trials.md).
+This verifies the narrower recovered-call capability; it does not supply selected
+definition bodies when the full graph is unavailable.
+
+The next completed pair reuses that exact question and oracle after adding compact
+call counts to the shared snapshot facet. Both adapters receive the same new
+metadata, derived after native validation from the complete retained call set;
+the engine, fact methods and selection semantics are unchanged. The hypothesis
+is that avoiding full-set paging for these counts reduces task costs. Count
+coverage remains recovered block-member calls, not complete machine calls or
+runtime binding. Both answers satisfy the same oracle and avoid call-set paging.
+Cumulative input is lower in both new samples, but elapsed/uncached costs improve
+only for direct; direct now beats Code Mode within the pair. These observations
+support useful shared aggregates, without establishing a general efficiency
+effect. Selected rows/full source Evidence remain available for composition and
+audit; all native VM function records equal the earlier captures.
+
+The atomic direct and Code Mode lanes share the same method registry, provider
+owner, selectors, failures and retained sources. The traditional lane selects
+relevant production tools, including the discovery route advertised by its
+errors; its selection and extra discovery guidance are recorded. Use the same
+target, engine, authority and question. Freeze runtime files for each model batch.
+Keep full original errors, retries and partial answers in the record. Never show
+this oracle document or repository sources to a measured model.
+
+Review task facts separately from answer length. A confident false physical
+branch target fails the value-origin task even if its reconstructed IR equation
+is right. An answer stopping at an import stub does not complete a task whose
+other lane traces the formatter implementation. A bounded missing graph is not
+an empty graph. Mark unsupported facts unknown and report incompletion when the
+question actually needs them.
+
+Cache regimes are distinct: fresh analysis captures, a reused capture in the
+same lease, a validated persistent project in a new session, OS file cache and
+the model's reported prompt-cache tokens. These experiments do not clear the OS
+cache. The prototype has no cross-session result cache. Whole-model cold and
+reused-project trials therefore remain separate from the API-only same-lease
+cache benchmark. Report cumulative input, included cached input, uncached input,
+output and included reasoning; do not invent server compute or price savings.
+
+All resource trials are serial with one CPU, 4 GiB hard memory, early stop at
+3 GiB, and an observed cleanup receipt. Kernel family memory includes charged
+file cache. A single stochastic trial is an observation, not a reliable ranking;
+API-only replay cannot replace real model trials or establish their token cost.
+
+The CLI transcripts establish that all lanes have a client JavaScript `exec`
+surface, even with `code_mode=false` requested. Count actual model responses,
+client dispatch/wait calls and MCP requests separately. This comparison is not
+an ablation against a client incapable of code composition. The optional
+`agent-trial.py --normalize-mcp-guidance` adds identical envelope/stopping guidance
+to any lane: select `structuredContent` or parsed text once, avoid printing both
+encodings, stop when supported and close the target. It is recorded explicitly;
+the initial SQLite and VM rounds do not have that later guidance. Their native
+analysis commands/profiles also differ (selected functions versus production
+project and facet workflows); engine equality alone does not isolate Code Mode.
+
+The pinned SQLite artifact is `/usr/lib/x86_64-linux-gnu/libsqlite3.so.0.8.6`,
+SHA-256 `2e6eef9a727f081f0d453b4e5e6cbd8b9ef8b6f86cbf7681cbad444d3b0b55c8`.
+The malformed-MATCH oracle and earlier unequal-depth results are retained in
+[trials.md](trials.md). The next VM task is fixed in `sqlite-vm-question.txt`
+before model trials. Its oracle is checked with GNU `objdump -d -M intel` over
+the shipped executable, independently of both model answers and Kuna IR. Dynamic
+symbol inventory (`nm -D --defined-only`) identifies `sqlite3VdbeExec` at
+`0xea6d0`. Reconstructed C and printer line mappings help locate checks, but do
+not replace the physical instruction evidence:
+
+| Required local fact                                                         | Independently checked anchors                                                                                                                                                                   |
+| --------------------------------------------------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
+| Open and release strings are distinct                                       | `0x12c500` and `0x12c538`; loads at `0xf2381` and `0xf2437`                                                                                                                                     |
+| Operation selector zero takes the open path                                 | Load `[op+4]` at `0xedbe5`, test `0xedbf0`, `JNE 0xf0340` at `0xedbf2`                                                                                                                          |
+| Open is rejected for a positive signed counter                              | Load `[db+0xd8]` at `0xedbf8`, test `0xedbff`, `JG 0xf2372` at `0xedc01`                                                                                                                        |
+| Release first resolves the named savepoint, then tests counter and selector | Chain `[db+0x2f0]` at `0xf0340`, links `+0x18` at `0xf0358`, name comparison `0xf036f`; signed `JLE` at `0xf0382` bypasses rejection, selector `==1` at `0xf0384` / `0xf0389` selects `0xf242d` |
+| Both use the same emission/error tail                                       | Open falls to `sqlite3VdbeError@plt` call `0xf238b`; release jumps there at `0xf2446`; local status becomes `5` at `0xf239d`, then jumps to `0xea828` at `0xf23b8`                              |
+| Missing name differs                                                        | Null chain tests `0xf0347` / `0xf035f` reach `0xf1139`; separate format `"no such savepoint: %s"` and call `0xf115c`                                                                            |
+| Static authority and coverage remain explicit                               | Counter/selector names are interpretations of offsets; runtime feasibility, actual dynamic binding and the final result after shared error processing require additional evidence               |
+
+The VM's real native-IR exporter exceeds its operation capacity. It retains
+3,654 lines of reconstructed C and complete source Evidence. This task tests
+whether a model can use those available facts plus selected physical instruction
+windows without inventing a recovered graph. Its local status `5` must not be
+silently promoted to an unconditional final function return: shared error
+processing can still change behavior. No target execution is authorized here.
+
+The post-serialization-fix Pragma task is fixed in `sqlite-pragma-question.txt`.
+It tests a previously inaccessible code facet rather than treating a byte-limit
+failure as a fast completed answer. The two experimental calling modes use the
+same twelve methods and new engine digest `2290e8b1…`, with common normalization
+guidance enabled in both. There is no new traditional cost cell for this task.
+For the same pinned SQLite artifact, independent GNU instruction checks establish:
+
+- `sqlite3Pragma` entry `0xb90e0`; value-pointer test `0xbae56` / branch `0xbae5c`
+  distinguishes the read path from change handling at `0xb9b8c`.
+- `0xb9b90` tests connection byte `+0x65`; zero branches at `0xb9b94` to diagnostic
+  selection at `0xbc58a`. This precedes the additional selector-one bypass at
+  `0xb9b9a` / `0xb9b9f`, so their ordering must be preserved.
+- Allowed local update computes `(value+1)&7`, substitutes one for zero, and
+  stores the selected record's byte `+0x10` at `0xb9bdc`; byte `+0x11` becomes one
+  at `0xb9be3`. The input interpretation and source member names are derivations.
+- The actual diagnostic call is `0xbc596` through `sqlite3ErrorMsg@plt` at
+  `0x2a1e0`; `0xbc59b` jumps to common cleanup `0xb9293`.
+- Cleanup releases both temporary strings, checks the stack canary, and returns
+  at `0xb92d9`. Do not invent a fixed numeric return from this reconstructed
+  void routine. No SQL-level runtime trigger is demonstrated.
+
+Pragma's complete retained code and mappings are checked equal before/after the
+serialization fix. Its recovered graph remains explicitly unavailable at the
+encoded-byte capacity; that is a coverage limitation in both experimental lanes.
+
+The next seeded table-composition task is fixed in `sqlite-table-question.txt`.
+Both experimental modes have the same thirteen methods including `bytes`; a
+traditional comparison is not claimed because its selected catalog/available
+Kuna byte capability has not been established for this task. GNU disassembly of
+the pinned artifact shows `sqlite3GetToken` at `0xe1480`: unsigned input byte load
+at `0xe1480`, class-table base `0x12bcc0` at `0xe1484`, bound `0x1e` at `0xe1492`,
+class reload `0xe1498`, jump-table base `0x12b600` at `0xe149c`, sign-extending
+32-bit table load at `0xe14a3`, base addition `0xe14a7`, indirect jump `0xe14aa`.
+The out-of-range path begins `0xe14f8`.
+
+Independent direct file reads are valid here because `readelf -lW` binds this
+artifact's selected addresses to the same file offsets in its read-only LOAD
+segment (`0x11a000..0x154bfc`). A Python `struct.unpack_from('<i', ...)` check of
+the actual table bytes gives these initial dispatch facts, before model trials:
+
+| Input byte | Class | Signed displacement | Initial machine destination |
+| ---------- | ----: | ------------------: | --------------------------- |
+| `0x2d`     |    11 |             -302928 | `0xe16b0`                   |
+| `0x2f`     |    16 |             -302144 | `0xe19c0`                   |
+| `0x27`     |     8 |             -302480 | `0xe1870`                   |
+| `0x22`     |     8 |             -302480 | `0xe1870`                   |
+| `0x3f`     |     6 |             -302272 | `0xe1940`                   |
+| `0x80`     |    27 |             -303440 | `0xe14b0`                   |
+
+This oracle establishes static table/dispatch arithmetic, not final token
+semantics, future runtime table contents or a feasible SQL input. Models must
+retain those boundaries. This document is not supplied to measured models.
+
+## Cross-function facts work; adapter gains remain mixed
+
+The structured-call experiment closes an observed native export gap: the call's
+fspec annotation does not identify its target code address, and an argument's
+SSA storage is not its passing location. Reusing the native call-spec owner now
+allows a model to connect a caller value to a callee input without parsing a
+display string. Real SQLite answers support this connection with machine
+instructions, trace a shared parenthesis counter across two functions, and
+preserve the limits of recovered slots, aliases and runtime feasibility.
+
+The first question pair omitted oracle-required alternative return paths in both
+answers. Its wording was less explicit than its oracle, so that result stays a
+partial-coverage record. A new question clarification was committed before the
+next pair; both then satisfy the task. The failed engine-pin setup launch also
+remains recorded. No sample is silently converted into a completed task.
+
+In the correctness-equal pair, Code Mode uses 596,978 cumulative input tokens and
+156.821 seconds versus direct's 611,321 and 147.424 seconds. Uncached input,
+output, largest observed response input and family CPU are higher for Code Mode;
+family memory peak is lower. Both use the same fourteen fact methods and a client
+with JavaScript dispatch. The evidence supports usable composable call/input
+facts, without proving that this server adapter is faster or cheaper. Broader
+complex-software superiority and provider parity remain unproven.
+
+Two remaining boundaries deserve independent experiments. First, the initial call view lived inside full recovered IR, so graph capacity
+could hide a small otherwise useful call view. The next exporter revision now
+captures calls and their boundary values independently; real VM verification
+retains both while the graph exceeds capacity. Second, shallow outgoing references should serve a callee/reference list before
+requesting deeper argument/value recovery. The subsequent implementation now
+exposes Kuna's existing `xrefs --from` through the shared parser, with native kind
+filters, explicit scope and paging. Real VM known-target references match an
+independent disassembly pair set, and remain available when the full IR facet
+exceeds capacity. The next pre-pinned model question measures this use directly. Keeping those costs
+and coverage distinctions explicit is more useful than treating a smaller tool
+catalog as proof of better analysis.
+
+The shallow-reference question now completes through both real-model adapters.
+Each composes 396 VM reference rows into 171 exact destinations and 16 calls to
+the selected PLT address, then distinguishes a real stack-check call absent from
+optimized IR. Both keep unresolved indirect/discovery/runtime limits. Native
+reference pairs from both model sessions match the complete independent set of
+396 direct immediate machine calls; the separate 21 indirect instructions are
+not silently included in those known-target totals.
+
+In this one pair, Code Mode takes 93.837 seconds and 264,680 cumulative input
+versus direct's 106.183 and 285,086. Their maximum observed response input
+contexts are almost equal. This is useful positive evidence for selected facts
+plus local aggregation, while prior completed tasks show mixed adapter costs.
+It supports continuing the experiment, not a universal speed/token/provider
+superiority claim. Tool coverage and representation semantics were prerequisites
+for this comparison: a recovered-call-only view would have missed a real call,
+and VM full-IR capacity cannot establish that its shallow references are absent.
+
+An additional shared CLI/API run with the unpatched stock Kuna 1.833 executable
+also returns the same 396/171/16 VM aggregates, with fresh owned HOME/cache. The
+outgoing-reference capability uses native support, independently of the opt-in
+IR exporter. It takes 3.044 family wall / 2.695 CPU seconds and 182.6 MiB peak,
+with cleanup verified and OOM 0; these scripted costs are not model-token data or
+an adapter comparison. Stock external-spec freshness is still outside the engine
+binary digest. Full repository typechecking remains resource-limited as recorded
+in `native-engine.md`; source build, focused tests, changed-test typing and full
+lint/module checks pass.
+
+The bounded call-rooted definition projection now supports a local VM PHI task
+that full-graph capacity previously prevented. Four real-model answers over two
+revisions correctly recover both PHIs, all four alternatives and machine
+transfers, with source identities and global partial coverage retained. This is
+selected static-task coverage, not general complex-software superiority.
+
+The discovery correction removes an observed name-list miss and fixes defaulted
+input schemas, but the second pair does not show stable cost improvement. Direct
+has a 120-second native timeout before a successful retry; Code Mode changes its
+query strategy and costs more than its first run. The complete costs and failure
+remain in [the ledger](trials.md#phi-alternatives-after-named-discovery-correction).
+Fairness requires sufficient facts, correct discovery and honest retry reporting.
+
+The wide-PHI experiment provides a more precise candidate atom: an input slot is
+an individually selectable relationship, not just an array inside an operation.
+Its identity includes the operation and slot index; several relationships can
+share one value while preserving different predecessor edges. A value definition
+can then be composed with selected operation/value reads. Both measured models
+recover those distinctions and separate retained-node absence from unretained
+membership without fetching all 156 slots.
+
+This validates the shared query boundary, not adapter superiority. Code Mode
+groups the same seven fact requests, but four discovery lookups erase the
+MCP-request reduction at the observed model-response/client-dispatch level. Its
+95.699 seconds versus direct's 91.817, nearly equal cumulative input and lower
+uncached input remain a mixed single sample. Discovery is itself part of the
+analyst workflow and deserves a separate test. Broader behavior reconstruction,
+complete large-function CFG/use coverage, aliases and a universal RE basis are
+still unproven. See [the complete scoped results](trials.md#selected-wide-phi-input-relationships).
+
+The same question after adding a static method index again completes correctly
+through both adapters. Code Mode now uses two discovery calls instead of four,
+without changing the seven fact reads. Observed model responses decrease by one,
+but total time increases slightly; direct also takes longer and uses fewer input
+tokens without changing query counts. The second pair is approximately 98.8
+seconds each, with more cumulative and less uncached input for Code Mode. This
+supports the index's discoverability, while leaving whole-task cost mixed.
+
+There is a further distinction to test: an atomic fact API does not itself make
+native computation atomic. A fresh selected-slot question still incurs an entire
+function decompilation and C rendering. Phase costs are unmeasured, so neither
+removing C nor introducing persistent state is yet justified as a speed fix.
+Measure that boundary before claiming analysis-time improvement from narrower
+fact responses.

experiments/atomic-codemode/executor.mjs

@@ -0,0 +1,241 @@
+import { getQuickJS } from "quickjs-emscripten";
+import { methods } from "./api.mjs";
+import { MAX_BRIDGE_BYTES as MAX_BRIDGE } from "./limits.mjs";
+
+const MAX_OUTPUT = 32768;
+const failure = (code, message) => ({ ok: false, error: { code, message } });
+const projectGuestError = (code, error) => {
+  const message =
+    typeof error?.message === "string" ? error.message : String(error);
+  // Check raw size first: JSON escaping expands each code unit at most sixfold,
+  // so the second check has a 6×32 KiB string bound plus a fixed envelope.
+  if (
+    Buffer.byteLength(message) > MAX_OUTPUT ||
+    Buffer.byteLength(JSON.stringify({ code, message })) > MAX_OUTPUT
+  )
+    return {
+      code: "output_budget",
+      message:
+        "JavaScript error exceeds the 32 KiB output budget; its message is omitted. Native Evidence remains retained.",
+      details: {
+        original_error_code: code,
+        original_message_code_units: message.length,
+        original_message_omitted: true,
+      },
+    };
+  return { code, message };
+};
+export const executeJavaScript = async (
+  session,
+  code,
+  {
+    signal = new AbortController().signal,
+    timeoutMs = 120000,
+    cpuBudgetMs = 2000,
+  } = {},
+) => {
+  if (Buffer.byteLength(code) > 32768)
+    return failure("code_budget", "JavaScript source exceeds 32 KiB");
+  const started = performance.now();
+  const context = session.context;
+  const lease = context?.lease;
+  const stop = new AbortController();
+  const combined = AbortSignal.any([signal, stop.signal]);
+  const deadline = Date.now() + timeoutMs;
+  const timer = setTimeout(() => stop.abort(), timeoutMs);
+  const QuickJS = await getQuickJS();
+  const runtime = QuickJS.newRuntime();
+  runtime.setMemoryLimit(64 * 1024 * 1024);
+  runtime.setMaxStackSize(512 * 1024);
+  const budget = { used: 0, deadline: 0, exhausted: false };
+  const runGuest = (work) => {
+    budget.deadline = Date.now() + Math.max(1, cpuBudgetMs - budget.used);
+    const before = performance.now();
+    try {
+      return work();
+    } finally {
+      budget.used += performance.now() - before;
+    }
+  };
+  runtime.setInterruptHandler(() => {
+    budget.exhausted ||= Date.now() >= budget.deadline;
+    return budget.exhausted || combined.aborted || Date.now() >= deadline;
+  });
+  const vm = runtime.newContext();
+  const pending = new Set();
+  const calls = [];
+  const state = { disposed: false, evidence: new Set() };
+  const pump = () => {
+    if (!state.disposed) {
+      const result = runGuest(() => runtime.executePendingJobs());
+      if (result.error) {
+        result.error.dispose();
+        stop.abort();
+      }
+    }
+  };
+  installBridge(vm, session, { combined, lease, calls, pending, pump, state });
+  let outcome;
+  try {
+    const text = await evaluateGuest(vm, code, { runGuest, pump, combined });
+    if (Buffer.byteLength(text) > MAX_OUTPUT)
+      outcome = {
+        ok: false,
+        error: {
+          code: "output_budget",
+          message:
+            "Return a smaller projection (32 KiB UTF-8 output budget); native Evidence remains retained",
+        },
+      };
+    else
+      outcome = {
+        ok: true,
+        value: text === "undefined" ? null : JSON.parse(text),
+      };
+  } catch (error) {
+    outcome = {
+      ok: false,
+      error: projectGuestError(
+        budget.exhausted
+          ? "cpu_budget"
+          : combined.aborted
+            ? "cancelled"
+            : "javascript_error",
+        budget.exhausted
+          ? "Guest synchronous execution budget exceeded"
+          : error,
+      ),
+    };
+  } finally {
+    clearTimeout(timer);
+    stop.abort();
+    await Promise.allSettled(pending);
+    state.disposed = true;
+    vm.dispose();
+    runtime.dispose();
+  }
+  return {
+    ...outcome,
+    context,
+    evidence_ids: [...state.evidence],
+    calls,
+    elapsed_ms: performance.now() - started,
+  };
+};
+
+const retainEvidence = (response, evidence) => {
+  const observation =
+    response.data ?? response.error?.details?.partial_observation;
+  for (const row of [observation, ...(observation?.results ?? [])]) {
+    const id =
+      row?.source?.evidence_id ??
+      row?.evidence_id ??
+      row?.error?.details?.evidence_id;
+    if (id) evidence.add(id);
+  }
+  if (response.error?.details?.evidence_id)
+    evidence.add(response.error.details.evidence_id);
+};
+
+const installBridge = (
+  vm,
+  session,
+  { combined, lease, calls, pending, pump, state },
+) => {
+  const bridge = vm.newFunction("analysis", (methodHandle, inputHandle) => {
+    const name = vm.getString(methodHandle);
+    const inputText = vm.getString(inputHandle);
+    if (calls.length >= 64 || Buffer.byteLength(inputText) > MAX_BRIDGE)
+      throw new Error("Host call or argument budget exceeded");
+    const promise = vm.newPromise();
+    const index = calls.length;
+    calls.push({
+      method: name,
+      input: JSON.parse(inputText),
+      status: "pending",
+    });
+    const work = Promise.resolve()
+      .then(async () => {
+        const before = performance.now();
+        const response = await session.call(
+          name,
+          JSON.parse(inputText),
+          combined,
+          lease,
+        );
+        retainEvidence(response, state.evidence);
+        calls[index].status = response.ok ? "ok" : "error";
+        if (!response.ok) calls[index].error = response.error;
+        calls[index].elapsed_ms = performance.now() - before;
+        let serialized = JSON.stringify(response);
+        if (Buffer.byteLength(serialized) > MAX_BRIDGE)
+          serialized = JSON.stringify({
+            ok: false,
+            error: {
+              code: "bridge_budget",
+              message:
+                "Select fewer fields or a smaller page; full source remains available via evidence",
+            },
+          });
+        if (!state.disposed) {
+          const value = vm.newString(serialized);
+          promise.resolve(value);
+          value.dispose();
+          pump();
+        }
+      })
+      .catch((error) => {
+        if (!state.disposed) {
+          const value = vm.newString(
+            JSON.stringify({
+              ok: false,
+              error: { code: "bridge_failed", message: error.message },
+            }),
+          );
+          promise.resolve(value);
+          value.dispose();
+          pump();
+        }
+      })
+      .finally(() => pending.delete(work));
+    pending.add(work);
+    return promise.handle;
+  });
+  vm.setProp(vm.global, "__analysis", bridge);
+  bridge.dispose();
+};
+
+const evaluateGuest = async (vm, code, { runGuest, pump, combined }) => {
+  const bootstrap = `globalThis.rea=Object.freeze(Object.fromEntries(${JSON.stringify(Object.keys(methods))}.map(name=>[name,async(input={})=>JSON.parse(await __analysis(name,JSON.stringify(input)))])));`;
+  vm.unwrapResult(runGuest(() => vm.evalCode(bootstrap))).dispose();
+  const promiseHandle = vm.unwrapResult(
+    runGuest(() =>
+      vm.evalCode(
+        `(async()=>JSON.stringify(await(async()=>{\n${code}\n})()))()`,
+      ),
+    ),
+  );
+  try {
+    const resolved = vm.resolvePromise(promiseHandle);
+    pump();
+    // A never-settling guest Promise must not hang the MCP request forever.
+    const aborted = new Promise((_, reject) => {
+      if (combined.aborted) {
+        reject(new Error("Execution cancelled"));
+        return;
+      }
+      combined.addEventListener(
+        "abort",
+        () => reject(new Error("Execution cancelled or timed out")),
+        { once: true },
+      );
+    });
+    const result = await Promise.race([resolved, aborted]);
+    const value = vm.unwrapResult(result);
+    const text = vm.getString(value);
+    value.dispose();
+    return text;
+  } finally {
+    promiseHandle.dispose();
+  }
+};

experiments/atomic-codemode/executor.test.mjs

@@ -0,0 +1,120 @@
+import test from "node:test";
+import assert from "node:assert/strict";
+import { executeJavaScript } from "./executor.mjs";
+
+const fake = {
+  context: { lease: "test" },
+  call: async (name, input) => ({ ok: true, data: { name, input } }),
+};
+test("real QuickJS composes async API calls and does not expose host globals", async () => {
+  const result = await executeJavaScript(
+    fake,
+    'const a=await rea.strings({query:"x"}); const b=await rea.operations({ids:[a.data.name]}); return {a,b,host:[typeof process,typeof require,typeof fetch]};',
+  );
+  assert.equal(result.ok, true, JSON.stringify(result));
+  assert.deepEqual(result.value.host, ["undefined", "undefined", "undefined"]);
+  assert.equal(result.calls.length, 2);
+  assert.equal(result.value.b.data.input.ids[0], "strings");
+});
+test("infinite guest execution and never-resolving promises terminate", async () => {
+  const loop = await executeJavaScript(fake, "while(true){}", {
+    timeoutMs: 30,
+  });
+  assert.equal(loop.ok, false);
+  const stalled = await executeJavaScript(
+    fake,
+    "await new Promise(()=>{});return 1",
+    { timeoutMs: 30 },
+  );
+  assert.equal(stalled.ok, false);
+});
+test("output budgets and failures survive without silently clipping evidence", async () => {
+  const result = await executeJavaScript(fake, 'return "x".repeat(40000)');
+  assert.equal(result.error.code, "output_budget");
+  const failure = await executeJavaScript(
+    fake,
+    'throw new Error("explicit failure")',
+  );
+  assert.equal(failure.ok, false);
+});
+
+test("guest CPU budget is independent from the native-analysis wall deadline", async () => {
+  const before = performance.now();
+  const result = await executeJavaScript(fake, "while(true){}", {
+    timeoutMs: 5000,
+    cpuBudgetMs: 25,
+  });
+  assert.equal(result.error.code, "cpu_budget");
+  assert.ok(performance.now() - before < 2000);
+});
+
+test("cancelled reads keep partial-observation Evidence in the execution receipt", async () => {
+  const result = await executeJavaScript(
+    {
+      context: { lease: "partial" },
+      call: async () => ({
+        ok: false,
+        error: {
+          code: "cancelled",
+          details: {
+            partial_observation: { source: { evidence_id: "ev_partial" } },
+          },
+        },
+      }),
+    },
+    "return await rea.strings({query:'x'})",
+  );
+  assert.deepEqual(result.evidence_ids, ["ev_partial"]);
+  assert.equal(result.value.error.code, "cancelled");
+});
+
+test("failed compositions keep completed facts and their provenance after draining", async () => {
+  const owner = {
+    context: { lease: "retained" },
+    call: async () => ({ ok: true, data: { evidence_id: "ev_retained" } }),
+  };
+  for (const ending of [
+    'throw new Error("after analysis")',
+    'return "x".repeat(40000)',
+  ]) {
+    const result = await executeJavaScript(
+      owner,
+      `await rea.strings({}); ${ending}`,
+    );
+    assert.equal(result.ok, false);
+    assert.deepEqual(result.evidence_ids, ["ev_retained"]);
+    assert.deepEqual(result.context, owner.context);
+    assert.equal(result.calls[0].status, "ok");
+  }
+});
+
+test("guest exceptions obey output capacity, including JSON expansion, and preserve observations", async () => {
+  for (const code of [
+    'throw new Error("x".repeat(40000))',
+    'throw new Error("\\0".repeat(6000))',
+  ]) {
+    const result = await executeJavaScript(
+      {
+        context: { lease: "error-output" },
+        call: async () => ({
+          ok: true,
+          data: { evidence_id: "ev_before_error" },
+        }),
+      },
+      `await rea.strings({}); ${code}`,
+    );
+    assert.equal(result.ok, false);
+    assert.equal(result.error.code, "output_budget");
+    assert.equal(result.error.details.original_error_code, "javascript_error");
+    assert.equal(result.error.details.original_message_omitted, true);
+    assert.ok(Buffer.byteLength(JSON.stringify(result.error)) <= 32768);
+    assert.deepEqual(result.evidence_ids, ["ev_before_error"]);
+    assert.equal(result.context.lease, "error-output");
+  }
+  const ordinary = await executeJavaScript(
+    fake,
+    'throw new Error("x".repeat(10000))',
+  );
+  assert.equal(ordinary.error.code, "javascript_error");
+  assert.ok(ordinary.error.message.includes("x".repeat(10000)));
+});

experiments/atomic-codemode/fact-error.mjs

@@ -0,0 +1,24 @@
+import { projectAnalysisError } from "../../dist/domain/analysisErrorProjection.js";
+
+// Shared expected-failure algebra; execution adapters do not depend on native protocols.
+export class FactFailure extends Error {
+  constructor(code, message, details = {}) {
+    super(message);
+    this.code = code;
+    this.details = details;
+  }
+}
+
+export const requireResult = (result, code = "provider_failure") => {
+  if (!result.ok) {
+    const failure = result.error?._tag
+      ? projectAnalysisError(result.error)
+      : result.error;
+    throw new FactFailure(
+      failure?.code ?? code,
+      String(failure?.message ?? failure?.reason ?? failure),
+      { failure },
+    );
+  }
+  return result.value;
+};

experiments/atomic-codemode/fact-reads.mjs

@@ -0,0 +1,69 @@
+// Shared read projections and partial batch semantics; no provider protocols.
+// Parsed JSON facts are owned by their cache. SDK callers may retain views,
+// but must not be able to change later reads under the same Evidence identity.
+export function freezeFacts(value) {
+  if (value && typeof value === "object") {
+    for (const child of Object.values(value)) freezeFacts(child);
+    Object.freeze(value);
+  }
+  return value;
+}
+
+export const sourceSummary = ({ evidence_id, sha256, bytes, elapsed_ms }) => ({
+  evidence_id,
+  sha256,
+  bytes,
+  elapsed_ms,
+});
+
+export const projectFact = (row, fields, identity) =>
+  fields
+    ? Object.fromEntries(
+        [...new Set([...identity, ...fields])]
+          .filter((key) => Object.hasOwn(row, key))
+          .map((key) => [key, row[key]]),
+      )
+    : row;
+
+export const pageFacts = (rows, input, identity = []) => ({
+  items: rows
+    .slice(input.offset, input.offset + input.limit)
+    .map((row) => projectFact(row, input.fields, identity)),
+  total: rows.length,
+  next_offset:
+    input.offset + input.limit < rows.length
+      ? input.offset + input.limit
+      : null,
+});
+
+export async function readFactBatch(requests, capture, read) {
+  const results = [];
+  for (const request of requests) {
+    let source;
+    try {
+      const captured = await capture(request);
+      source = captured.source;
+      const facts = await read(captured, request);
+      results.push({
+        ...facts,
+        ok: true,
+        address: request.address,
+        source: sourceSummary(source),
+      });
+    } catch (error) {
+      results.push({
+        ok: false,
+        address: request.address,
+        error: {
+          code: error.code ?? "provider_failure",
+          message: error.message,
+          details: {
+            ...error.details,
+            ...(source ? { evidence_id: source.evidence_id } : {}),
+          },
+        },
+      });
+    }
+  }
+  return { results };
+}

experiments/atomic-codemode/family-counters.mjs

@@ -0,0 +1,21 @@
+// Read-only kernel counters for the currently bounded diagnostic payload.
+import assert from "node:assert/strict";
+import { readFile } from "node:fs/promises";
+
+export async function familyCounters() {
+  const group = (await readFile("/proc/self/cgroup", "utf8"))
+    .split("\n")
+    .find((row) => row.startsWith("0::"))
+    ?.slice(3);
+  assert.ok(group?.includes("rea-efficiency-"), "Run under resource-run.py");
+  const cpu = (await readFile(`/sys/fs/cgroup${group}/cpu.stat`, "utf8")).match(
+    /^usage_usec (\d+)$/mu,
+  );
+  assert.ok(cpu, "Missing cgroup CPU counter");
+  return {
+    cpu: Number(cpu[1]),
+    peak: Number(
+      (await readFile(`/sys/fs/cgroup${group}/memory.peak`, "utf8")).trim(),
+    ),
+  };
+}

experiments/atomic-codemode/function-facts.mjs

@@ -0,0 +1,33 @@
+import { parseKunaInventory } from "../../dist/kuna/KunaProjectFiles.js";
+import { requireResult } from "./fact-error.mjs";
+import { freezeFacts, pageFacts } from "./fact-reads.mjs";
+
+export const parseFunctionFacts = (text) =>
+  freezeFacts(
+    requireResult(parseKunaInventory(text), "invalid_provider_output"),
+  );
+
+export function selectFunctionFacts(functions, input) {
+  const addresses = input.addresses && new Set(input.addresses);
+  const normalize = (name) =>
+    input.case_sensitive ? name : name.toLowerCase();
+  const query = normalize(input.query);
+  const matches = functions.filter(
+    (entry) =>
+      (!addresses || addresses.has(entry.address)) &&
+      [entry.name, ...entry.aliases].some((name) =>
+        input.exact
+          ? normalize(name) === query
+          : normalize(name).includes(query),
+      ),
+  );
+  return {
+    ...pageFacts(matches, input, ["address"]),
+    coverage: {
+      scope: "native callable-entry inventory, including import stubs",
+      discovery: "unknown",
+      extents: "upper bounds clipped at the next entry; zero means unmeasured",
+      symbol_binding_at_runtime: "unknown",
+    },
+  };
+}

experiments/atomic-codemode/function-facts.test.mjs

@@ -0,0 +1,93 @@
+import test from "node:test";
+import assert from "node:assert/strict";
+import { parseFunctionFacts, selectFunctionFacts } from "./function-facts.mjs";
+import { readFactBatch } from "./fact-reads.mjs";
+import { FactFailure } from "./fact-error.mjs";
+
+test("function names retain multiple entries and aliases instead of choosing a PLT or implementation", () => {
+  const facts = parseFunctionFacts(
+    JSON.stringify({
+      functions: [
+        {
+          name: "format_error",
+          address_hex: "0x1000",
+          aliases: ["format_error@plt"],
+          size: 16,
+        },
+        {
+          name: "format_error",
+          address_hex: "0x2000",
+          aliases: ["FormatError"],
+          size: 0,
+        },
+      ],
+    }),
+  );
+  const query = {
+    query: "format_error",
+    exact: true,
+    case_sensitive: true,
+    offset: 0,
+    limit: 20,
+  };
+  assert.deepEqual(
+    selectFunctionFacts(facts, query).items.map((item) => item.address),
+    ["0x1000", "0x2000"],
+  );
+  assert.equal(
+    selectFunctionFacts(facts, { ...query, query: "format_error@plt" }).items[0]
+      .address,
+    "0x1000",
+  );
+  assert.equal(
+    selectFunctionFacts(facts, {
+      ...query,
+      query: "FORMAT_ERROR",
+      case_sensitive: false,
+    }).total,
+    2,
+  );
+  assert.deepEqual(
+    selectFunctionFacts(facts, {
+      ...query,
+      addresses: ["0x2000"],
+      fields: ["size"],
+    }).items,
+    [{ address: "0x2000", size: 0 }],
+  );
+  assert.throws(() => {
+    facts[0].aliases.push("foreign");
+  });
+});
+
+test("batch transport and decoder failures preserve successful items and captured source identity", async () => {
+  const attempts = [];
+  const result = await readFactBatch(
+    [{ address: "0x1" }, { address: "0x2" }, { address: "0x3" }],
+    async (request) => {
+      attempts.push(request.address);
+      if (request.address === "0x2")
+        throw new FactFailure("capacity", "transport bound");
+      return {
+        source: {
+          evidence_id: `ev_${request.address}`,
+          sha256: "digest",
+          bytes: 3,
+          elapsed_ms: 1,
+        },
+      };
+    },
+    async (_, request) => {
+      if (request.address === "0x3")
+        throw new FactFailure(
+          "invalid_provider_output",
+          "decoder rejected source",
+        );
+      return { items: [{ value: 1 }] };
+    },
+  );
+  assert.deepEqual(attempts, ["0x1", "0x2", "0x3"]);
+  assert.equal(result.results[0].items[0].value, 1);
+  assert.equal(result.results[1].error.code, "capacity");
+  assert.equal(result.results[2].error.details.evidence_id, "ev_0x3");
+});

experiments/atomic-codemode/instruction-facts.mjs

@@ -0,0 +1,109 @@
+// The machine-decoding boundary is separate from optimized recovered IR.
+import { z } from "zod";
+import { FactFailure } from "./fact-error.mjs";
+import { NATIVE_OUTPUT_BYTES } from "./limits.mjs";
+
+const hex = z.string().regex(/^0x(?:0|[1-9a-f][0-9a-f]{0,15})$/u);
+const rowSchema = z.object({
+  address_hex: hex,
+  size: z.number().int().positive().max(65536),
+  bytes: z.string().regex(/^(?:[0-9a-f]{2})+$/u),
+  mnemonic: z.string(),
+  operands: z.string(),
+  text: z.string(),
+});
+const envelopeSchema = z.object({
+  binary: z.string(),
+  kind: z.literal("code"),
+  start_hex: hex,
+  end_hex: hex,
+  count: z.number().int().nonnegative().max(4096),
+  bytes: z.number().int().nonnegative(),
+  truncated: z.boolean(),
+  notes: z.array(z.string()),
+});
+const requireFact = (condition, message) => {
+  if (!condition) throw new FactFailure("invalid_provider_output", message);
+};
+
+export function parseInstructions(
+  text,
+  inputPath,
+  request,
+  fields = ["bytes", "text"],
+) {
+  try {
+    requireFact(
+      Buffer.byteLength(text) <= NATIVE_OUTPUT_BYTES,
+      "Instruction source exceeds transport capacity",
+    );
+    const raw = JSON.parse(text);
+    requireFact(
+      Array.isArray(raw?.instructions) && raw.instructions.length <= 4096,
+      "Instruction window exceeds decoded-row capacity",
+    );
+    const doc = envelopeSchema.parse(raw);
+    requireFact(
+      doc.binary === inputPath && doc.start_hex === request.address,
+      "Instruction source does not match the selected input/window",
+    );
+    requireFact(
+      doc.count === raw.instructions.length && doc.count <= request.count,
+      "Instruction count does not match the requested window",
+    );
+    let next = BigInt(request.address);
+    const items = raw.instructions.map((value) => {
+      const row = rowSchema.parse(value);
+      requireFact(
+        BigInt(row.address_hex) === next && row.bytes.length === row.size * 2,
+        "Instruction bytes/size or sequential address mismatch",
+      );
+      next += BigInt(row.size);
+      const fact = {
+        address: row.address_hex,
+        size: row.size,
+        status: row.mnemonic === ".byte" ? "undecoded" : "decoded",
+        bytes: row.bytes,
+        mnemonic: row.mnemonic,
+        operands: row.operands,
+        text: row.text,
+      };
+      return Object.fromEntries(
+        [...new Set(["address", "size", "status", ...fields])].map((key) => [
+          key,
+          fact[key],
+        ]),
+      );
+    });
+    requireFact(
+      next === BigInt(doc.end_hex) &&
+        next - BigInt(request.address) === BigInt(doc.bytes),
+      "Instruction window end/byte coverage mismatch",
+    );
+    return {
+      items,
+      resume_address: doc.end_hex,
+      coverage: {
+        authority: "machine bytes and native linear decoding; not recovered IR",
+        request,
+        listed_count: items.length,
+        truncated: doc.truncated,
+        notes: doc.notes,
+        alignment:
+          "caller-selected start; runtime execution and instruction alignment are not established",
+      },
+    };
+  } catch (error) {
+    if (error instanceof FactFailure) throw error;
+    throw new FactFailure(
+      "invalid_provider_output",
+      "Native instruction window does not match the declared format",
+      {
+        reason:
+          error instanceof z.ZodError
+            ? error.issues[0]?.message.slice(0, 256)
+            : String(error.message).slice(0, 256),
+      },
+    );
+  }
+}

experiments/atomic-codemode/instruction-facts.test.mjs

@@ -0,0 +1,92 @@
+import test from "node:test";
+import assert from "node:assert/strict";
+import { parseInstructions } from "./instruction-facts.mjs";
+
+const document = (address = "0xd97e") => ({
+  binary: "/private/input",
+  kind: "code",
+  start_hex: address,
+  end_hex: `0x${(BigInt(address) + 6n).toString(16)}`,
+  count: 1,
+  bytes: 6,
+  truncated: true,
+  notes: ["selected window"],
+  instructions: [
+    {
+      address_hex: address,
+      size: 6,
+      bytes: "0f87e9080000",
+      mnemonic: "JA",
+      operands: "0xe26d",
+      text: "JA 0xe26d",
+    },
+  ],
+});
+const parse = (doc, address = doc.start_hex) =>
+  parseInstructions(JSON.stringify(doc), "/private/input", {
+    address,
+    count: 1,
+  });
+
+test("instruction facts preserve exact bytes/addresses and do not substitute an optimized CFG target", () => {
+  const result = parse(document());
+  assert.equal(result.items[0].address, "0xd97e");
+  assert.equal(result.items[0].text, "JA 0xe26d");
+  assert.equal(result.items[0].bytes, "0f87e9080000");
+  assert.equal(result.items[0].status, "decoded");
+  assert.deepEqual(result.coverage.notes, ["selected window"]);
+  const high = document("0x20000000000001");
+  high.instructions[0].address = Number(BigInt(high.start_hex));
+  assert.equal(parse(high).items[0].address, "0x20000000000001");
+});
+
+test("unknown bytes remain undecoded and projection retains byte-range identity", () => {
+  const doc = document();
+  doc.instructions[0].mnemonic = ".byte";
+  const result = parseInstructions(
+    JSON.stringify(doc),
+    "/private/input",
+    { address: "0xd97e", count: 1 },
+    ["mnemonic"],
+  );
+  assert.deepEqual(result.items[0], {
+    address: "0xd97e",
+    size: 6,
+    status: "undecoded",
+    mnemonic: ".byte",
+  });
+});
+
+test("wrong target/window, gaps, extent and byte-size contradictions fail at decoding boundary", () => {
+  for (const mutate of [
+    (doc) => {
+      doc.binary = "/other/input";
+    },
+    (doc) => {
+      doc.instructions[0].address_hex = "0xd97f";
+    },
+    (doc) => {
+      doc.instructions[0].size = 5;
+    },
+    (doc) => {
+      doc.end_hex = "0xd985";
+    },
+    (doc) => {
+      doc.count = 2;
+    },
+  ]) {
+    const doc = document();
+    mutate(doc);
+    assert.throws(
+      () => parse(doc),
+      (error) => error.code === "invalid_provider_output",
+    );
+  }
+  assert.throws(() => parse(document(), "0xd970"), /window/u);
+});
+
+test("decoded-row capacity rejects oversized malformed arrays before expanding errors", () => {
+  const doc = document();
+  doc.instructions = Array.from({ length: 4097 }, () => ({}));
+  assert.throws(() => parse(doc), /capacity/u);
+});

experiments/atomic-codemode/kuna-atomic-timing-diagnostic.patch

@@ -0,0 +1,86 @@
+--- a/decompiler/crates/kuna-console/src/project.rs
++++ b/decompiler/crates/kuna-console/src/project.rs
+@@ -693,6 +693,17 @@
+         flow_ovr.extend(seed.flow_overrides.iter().cloned());
+         let replay = crate::assertions::has_symbol_scoped(prog, &name, single_target);
+         prog.arch_mut().kuna_record_names = replay;
++        let mut profile_clock = (want_atomic_facts
++            && std::env::var_os("KUNA_EXPERIMENT_ATOMIC_TIMING").as_deref()
++                == Some(std::ffi::OsStr::new("1")))
++            .then(std::time::Instant::now);
++        let mut profile_phase = |phase: &str| {
++            if let Some(clock) = profile_clock.as_mut() {
++                let now = std::time::Instant::now();
++                eprintln!("{{\"kind\":\"rea_atomic_phase\",\"address\":\"0x{byte_address:x}\",\"phase\":\"{phase}\",\"wall_ns\":{}}}", now.duration_since(*clock).as_nanos());
++                *clock = std::time::Instant::now();
++            }
++        };
+         let step = crate::decompile_step::decompile_one(
+             prog.arch_mut(),
+             &name,
+@@ -751,6 +762,7 @@
+             }
+             other => other,
+         };
++        profile_phase("decompile_drive");
+         match result {
+             Ok(mut fd) => {
+                 // (kuna `--assert`) The flow overrides the follower REFUSED are only
+@@ -813,6 +825,7 @@
+                 kuna_decomp::kuna_elemptr::record(prog.arch_mut(), &fd);
+                 prog.arch_mut().print_mut().set_record_pointer_arguments(want_pointer_arguments);
+                 let rendered = std::panic::catch_unwind(std::panic::AssertUnwindSafe(|| {
++                    profile_phase("pre_render_metadata");
+                     // Trim the surrounding newlines the same way `kuna decompile`
+                     // does (`decompile.rs::trim_newlines`), so the per-function
+                     // `code` is byte-identical to the single-shot path.
+@@ -826,6 +839,7 @@
+                     };
+                     let pointerargs = prog.arch_mut().print_mut().take_pointer_arguments(&untrimmed);
+                     let code = untrimmed.trim_matches('\n').to_string();
++                    profile_phase("render_c");
+                     let globals = kuna_decomp::decompile_drive::extract_global_objects(prog.arch());
+                     let field_calls = kuna_decomp::kuna_calltargettype::field_declarations(prog.arch(), &fd);
+                     let mut variables =
+@@ -848,15 +862,20 @@
+                             prog, &fd, &var_refs, &raw_tokens, &untrimmed, &code, &variables,
+                         ))
+                     });
++                    profile_phase("metadata_extract");
+                     if want_atomic_facts {
+                         let detail = detail.get_or_insert_with(Default::default);
+                         let membership = kuna_decomp::atomic_membership::inspect(&fd);
++                        profile_phase("membership");
+                         detail.atomic_calls = Some(membership.as_ref().map_err(Clone::clone)
+                             .and_then(|m| kuna_decomp::atomic_calls::collect_all(&fd, m)));
++                        profile_phase("calls");
+                         detail.atomic_facts = Some(membership.as_ref().map_err(Clone::clone)
+                             .and_then(|m| kuna_decomp::atomic_facts::collect(prog.arch(), &fd, m)));
++                        profile_phase("full_graph");
+                         detail.atomic_projection = Some(membership.as_ref().map_err(Clone::clone)
+                             .and_then(|m| kuna_decomp::atomic_projection::collect(prog.arch(), &fd, m)));
++                        profile_phase("projection");
+                     }
+                     for variable in &mut variables {
+                         for address in &mut variable.addresses {
+@@ -881,6 +900,7 @@
+                         .as_ref()
+                         .map(|ctx| ctx.scan(&fd, byte_address))
+                         .unwrap_or_default();
++                    profile_phase("metadata_tail");
+                     (code, variables, types, (globals, field_calls), proto, line_mappings, callee_hints, detail, pointerargs, stack_objects)
+                 }));
+                 match rendered {
+--- a/docs/spec/00-overview.md
++++ b/docs/spec/00-overview.md
+@@ -3496,3 +3496,10 @@
+ runtime reachability. Collection uses the same live function after C rendering;
+ no second decompilation is performed. Complete source Evidence keeps the native
+ frontier while callers may select a compact summary and individual facts.
++
++An experiment-only timing overlay recognizes `KUNA_EXPERIMENT_ATOMIC_TIMING=1`
++only with atomic capture enabled. It records sequential per-function wall times
++to stderr for the decompile drive, pre-render metadata, C rendering, metadata
++extraction, membership, calls, full graph, projection and metadata tail. It does
++not change analysis parameters or JSON facts. Loading, setup and final JSON
++encoding are outside these phase intervals; phase CPU costs are not measured.

experiments/atomic-codemode/kuna-graph-scale-diagnostic.patch

@@ -0,0 +1,28 @@
+--- a/decompiler/crates/kuna-decomp/src/substrate/atomic_facts.rs
++++ b/decompiler/crates/kuna-decomp/src/substrate/atomic_facts.rs
+@@ -64,10 +64,10 @@
+ pub fn collect(arch: &Architecture, fd: &Funcdata) -> Result<FunctionFacts, String> {
+     // Bounds cover allocations before JSON serialization, independently of the
+     // transport byte budget. Overflow is an unavailable facet, not clipped IR.
+-    const MAX_OPERATIONS: usize = 12_000;
+-    const MAX_VALUES: usize = 36_000;
+-    const MAX_INPUTS: usize = 50_000;
+-    const MAX_DISPLAY_BYTES: usize = 2 * 1024 * 1024;
++    const MAX_OPERATIONS: usize = 32_000;
++    const MAX_VALUES: usize = 72_000;
++    const MAX_INPUTS: usize = 128_000;
++    const MAX_DISPLAY_BYTES: usize = 16 * 1024 * 1024;
+     if fd.bblocks_get_size() <= 0 {
+         return Err("recovered basic-block graph is unavailable".into());
+     }
+--- a/decompiler/crates/kuna-decomp/src/substrate/atomic_values.rs
++++ b/decompiler/crates/kuna-decomp/src/substrate/atomic_values.rs
+@@ -7,7 +7,7 @@
+ use slotmap::Key;
+ use std::collections::BTreeSet;
+
+-pub const MAX_TEXT_BYTES: usize = 2 * 1024 * 1024;
++pub const MAX_TEXT_BYTES: usize = 16 * 1024 * 1024;
+ #[derive(Debug, Clone, Serialize)]
+ pub struct ValueFact {
+     pub id: String,

experiments/atomic-codemode/limits.mjs

@@ -0,0 +1,5 @@
+// Shared application resource policy, independent of native wire formats.
+export const NATIVE_OUTPUT_BYTES = 8 * 1024 * 1024;
+export const MAX_BRIDGE_BYTES = 2 * 1024 * 1024;
+// Aggregate selected bytes keep hex at 128 KiB, with headroom for provenance.
+export const MAX_BYTE_BATCH_BYTES = 64 * 1024;

experiments/atomic-codemode/measure-native-encoding.py

@@ -0,0 +1,70 @@
+"""Lossless encoding feasibility on retained native JSON; no provider implementation."""
+
+import argparse
+import base64
+import gzip
+import hashlib
+import json
+from pathlib import Path
+import time
+
+
+def measure(source, output):
+    with source.open('rb') as stream:
+        raw = stream.read(32 * 1024 * 1024 + 1)
+    if len(raw) > 32 * 1024 * 1024:
+        raise ValueError('Input exceeds the diagnostic 32 MiB source bound')
+    document = json.loads(raw)
+    if len(document['functions']) != 1:
+        raise ValueError('Select one retained native function')
+    function = document['functions'][0]
+    graph = function['atomic_facts']
+    graph_json = json.dumps(graph, separators=(',', ':'), ensure_ascii=False).encode()
+    wall, cpu = time.perf_counter(), time.process_time()
+    compressed = gzip.compress(graph_json, compresslevel=1, mtime=0)
+    base64_data = base64.b64encode(compressed)
+    encode_cost = {'wall_ms': (time.perf_counter() - wall) * 1000,
+                   'cpu_ms': (time.process_time() - cpu) * 1000}
+    wall, cpu = time.perf_counter(), time.process_time()
+    restored = gzip.decompress(base64.b64decode(base64_data, validate=True))
+    if restored != graph_json:
+        raise ValueError('Lossless byte roundtrip failed')
+    decode_cost = {'wall_ms': (time.perf_counter() - wall) * 1000,
+                   'cpu_ms': (time.process_time() - cpu) * 1000}
+    results = {}
+    for name, data in [('gzip-base64', base64_data.decode()),
+                       ('gzip-hex', compressed.hex())]:
+        function['atomic_facts'] = {
+            'schema_version': 4, 'encoding': name,
+            'raw_bytes': len(graph_json), 'data': data,
+        }
+        results[name] = len(json.dumps(document, separators=(',', ':'),
+                                      ensure_ascii=False).encode())
+    report = {
+        'source': str(source), 'source_sha256': hashlib.sha256(raw).hexdigest(),
+        'original_document_bytes': len(raw), 'graph_json_bytes': len(graph_json),
+        'graph_statement_bytes': sum(len(op['statement'].encode()) for op in graph['operations']),
+        'graph_operations': len(graph['operations']),
+        'graph_values': len(graph['values']), 'graph_blocks': len(graph['blocks']),
+        'gzip_level': 1, 'gzip_bytes': len(compressed),
+        'candidate_document_bytes': results,
+        'public_transport_budget_bytes': 8 * 1024 * 1024,
+        'lossless_graph_byte_roundtrip': True,
+        'python_gzip_and_base64_encoding': encode_cost,
+        'python_base64_and_gzip_decoding': decode_cost,
+        'scope': 'Offline retained-source format feasibility, not a native/provider/model '
+                 'implementation. Timings are Python only, not Rust/Node or agent savings. '
+                 'Whole-run CPU/peak family memory are separate resource-run receipts.',
+    }
+    with output.open('x') as stream:
+        json.dump(report, stream, indent=2)
+        stream.write('\n')
+    print(json.dumps(report))
+
+
+if __name__ == '__main__':
+    parser = argparse.ArgumentParser(description=__doc__)
+    parser.add_argument('source', type=Path)
+    parser.add_argument('output', type=Path)
+    args = parser.parse_args()
+    measure(args.source, args.output)

experiments/atomic-codemode/measure-native-projection.py

@@ -0,0 +1,186 @@
+"""Size diagnostic SQLite SSA projections; does not implement a provider API."""
+
+import argparse
+from collections import deque
+import hashlib
+import json
+from pathlib import Path
+
+
+COLUMNS = {
+    'blocks': ['id', 'start', 'operations', 'predecessors', 'successors'],
+    'operations': ['id', 'address', 'block', 'opcode', 'statement', 'is_dead', 'inputs', 'output'],
+    'inputs': ['input_index', 'value_id', 'phi_predecessor', 'phi_predecessor_edge'],
+    'successors': ['id', 'target', 'index', 'polarity', 'condition_value', 'condition_test'],
+    'values': ['id', 'space', 'offset', 'size_bytes', 'is_constant', 'is_input',
+               'is_annotation', 'definition', 'type_name'],
+}
+
+
+def encoded(value):
+    return json.dumps(value, ensure_ascii=True, separators=(',', ':')).encode()
+
+
+def bounded_source(path, budget):
+    with path.open('rb') as source:
+        raw = source.read(budget + 1)
+    assert len(raw) <= budget, 'Use a bounded native capture'
+    return raw
+
+
+def verify_capture(golden, path):
+    raw = bounded_source(path, 8 * 1024 * 1024)
+    document = json.loads(raw)
+    assert len(document['functions']) == 1
+    candidate = document['functions'][0]
+    ordinary = lambda fn: {key: row for key, row in fn.items()
+                           if not key.startswith('atomic_')}
+    assert ordinary(candidate) == ordinary(golden)
+    assert candidate['atomic_calls'] == golden['atomic_calls']
+    projection = candidate['atomic_projection']
+    assert projection['schema_version'] == 1
+    assert projection['coverage']['operation_budget'] == 4096
+    operations = {row['id']: row for row in golden['atomic_facts']['operations']}
+    values = {row['id']: row for row in golden['atomic_facts']['values']}
+    assert all(row == operations[row['id']] for row in projection['operations'])
+    assert all(row == values[row['id']] for row in projection['values'])
+    roots = [row['id'] for row in golden['atomic_calls']['items']]
+    assert projection['roots'] == roots
+    queue, queued, expected = deque(roots), set(roots), []
+    while queue and len(expected) < 4096:
+        key = queue.popleft()
+        if key not in operations:
+            continue
+        expected.append(key)
+        for operand in operations[key]['inputs']:
+            value = values.get(operand['value_id'])
+            definition = value['definition'] if value else None
+            if definition is not None and definition not in queued:
+                queued.add(definition)
+                queue.append(definition)
+    actual = [row['id'] for row in projection['operations']]
+    assert actual == expected
+    referenced = {row['output'] for row in projection['operations'] if row['output'] is not None} | {
+        i['value_id'] for row in projection['operations'] for i in row['inputs']
+        if i['value_id'] is not None}
+    assert len(projection['values']) == len(referenced)
+    assert {row['id'] for row in projection['values']} == referenced
+    missing = {row['definition'] for row in projection['values']
+               if row['definition'] is not None} - set(actual)
+    expected_frontier = {key: 'operation-budget' if key in operations
+                         else 'outside-recovered-membership' for key in missing}
+    actual_frontier = {row['definition']: row['reason']
+                       for row in projection['coverage']['frontier']}
+    assert len(actual_frontier) == len(projection['coverage']['frontier'])
+    assert actual_frontier == expected_frontier
+    return {'scope': 'Retained-source comparison, not independent machine/runtime equivalence',
+            'source': str(path), 'source_sha256': hashlib.sha256(raw).hexdigest(),
+            'ordinary_record_and_call_facet_equal': True,
+            'all_operation_and_value_records_equal': True,
+            'breadth_first_budget_and_frontier_match': True,
+            'operations': len(actual), 'values': len(referenced), 'frontier': len(missing)}
+
+
+def pack(table, rows):
+    result = []
+    for row in rows:
+        assert set(row) == set(COLUMNS[table]), 'Do not silently omit unknown fields'
+        result.append([pack(key, row[key]) if key in ('inputs', 'successors') else row[key]
+                       for key in COLUMNS[table]])
+    return result
+
+
+def unpack(table, rows):
+    return [{key: unpack(key, value) if key in ('inputs', 'successors') else value
+             for key, value in zip(COLUMNS[table], row, strict=True)} for row in rows]
+
+
+def projection(operations, values, roots, depth, cap):
+    retained, wave = set(roots), set(roots)
+    exhausted = False
+    for _ in range(depth):
+        following = {values[i['value_id']]['definition'] for key in wave
+                     for i in operations[key]['inputs'] if i['value_id'] is not None
+                     and values[i['value_id']]['definition'] in operations} - retained
+        if len(retained | following) > cap:
+            exhausted = True
+            break
+        retained |= following
+        wave = following
+    ids = {operations[key]['output'] for key in retained
+           if operations[key]['output'] is not None} | {
+               i['value_id'] for key in retained for i in operations[key]['inputs']
+               if i['value_id'] is not None}
+    frontier = {values[i['value_id']]['definition'] for key in retained
+                for i in operations[key]['inputs'] if i['value_id'] is not None
+                and values[i['value_id']]['definition'] is not None} - retained
+    payload = {'operations': [operations[key] for key in sorted(retained)],
+               'values': [values[key] for key in sorted(ids)]}
+    return {'requested_definition_depth': depth, 'operation_budget': cap,
+            'operation_budget_exhausted': exhausted, 'operations': len(retained),
+            'values': len(ids), 'encoded_bytes': len(encoded(payload)),
+            'unexpanded_definition_frontier': len(frontier)}
+
+
+def main():
+    parser = argparse.ArgumentParser(description=__doc__)
+    parser.add_argument('native_json', type=Path)
+    parser.add_argument('report_json', type=Path)
+    parser.add_argument('--verify-capture', type=Path,
+                        help='Also compare a public projection capture with this full diagnostic')
+    args = parser.parse_args()
+    budget = 32 * 1024 * 1024
+    raw = bounded_source(args.native_json, budget)
+    document = json.loads(raw)
+    assert len(document['functions']) == 1
+    function = document['functions'][0]
+    assert function['address_hex'] == '0xea6d0', 'This is the seeded SQLite VM experiment'
+    graph = function['atomic_facts']
+    assert graph['schema_version'] == 3
+    assert len(graph['operations']) <= 32000 and len(graph['values']) <= 72000
+    tables = ['blocks', 'operations', 'values']
+    assert set(graph) == {'schema_version', 'coverage', *tables}
+    packed = {'encoding': 'diagnostic-tuples-proposal-not-public',
+              'schema_version': graph['schema_version'], 'columns': COLUMNS,
+              'coverage': graph['coverage'],
+              **{key: pack(key, graph[key]) for key in tables}}
+    restored = {'schema_version': packed['schema_version'], 'coverage': packed['coverage'],
+                **{key: unpack(key, packed[key]) for key in tables}}
+    assert restored == graph
+    operations = {row['id']: row for row in graph['operations']}
+    values = {row['id']: row for row in graph['values']}
+    assert len(operations) == len(graph['operations'])
+    assert len(values) == len(graph['values'])
+    calls = function['atomic_calls']['items']
+    selected = [row['id'] for row in calls
+                if row['origin_address'] in ('0xf115c', '0xf238b')]
+    assert len(selected) == 2
+    report = {
+        'scope': 'Offline diagnostic sizes, not public-provider or model results. '
+                 'Tuple roundtrip is lossless, but no native codec is implemented. '
+                 'Projections follow retained SSA operands only; no CFG, aliases, '
+                 'source equivalence or runtime proof. Bytes are not token/time savings.',
+        'source': str(args.native_json), 'source_sha256': hashlib.sha256(raw).hexdigest(),
+        'whole_native_document_bytes': len(raw), 'full_graph_bytes': len(encoded(graph)),
+        'lossless_tuple_graph_bytes': len(encoded(packed)), 'tuple_roundtrip_equal': True,
+        'estimated_whole_tuple_document_bytes': len(raw) - len(encoded(graph)) + len(encoded(packed)),
+        'operations': len(operations), 'values': len(values), 'blocks': len(graph['blocks']),
+        'inputs': sum(len(row['inputs']) for row in operations.values()),
+        'cfg_edges': sum(len(row['successors']) for row in graph['blocks']),
+        'statement_utf8_bytes': sum(len(row['statement'].encode()) for row in operations.values()),
+        'value_type_utf8_bytes': sum(len(row['type_name'].encode()) for row in values.values()),
+        'selected_call_projections': [projection(operations, values, selected, depth, 512)
+                                      for depth in range(1, 6)],
+        'all_call_projections': [projection(operations, values, [r['id'] for r in calls], depth, 4096)
+                                 for depth in range(1, 5)],
+    }
+    if args.verify_capture:
+        report['verified_projection'] = verify_capture(function, args.verify_capture)
+    with args.report_json.open('x') as output:
+        json.dump(report, output, indent=2)
+        output.write('\n')
+    print(json.dumps(report, indent=2))
+
+
+if __name__ == '__main__':
+    main()

experiments/atomic-codemode/native-call-contract.mjs

@@ -0,0 +1,57 @@
+import { z } from "zod";
+import { nativeCall } from "./native-calls.mjs";
+import {
+  ordinal,
+  value,
+  rowParser,
+  readNativeFacet,
+  requireFact,
+} from "./native-contract.mjs";
+
+const callFacet = z.strictObject({
+  schema_version: z.literal(1),
+  items: z.array(nativeCall).max(12000),
+  values: z.array(value).max(36000),
+  coverage: z.strictObject({
+    scope: z.string(),
+    blocks_examined: ordinal.min(1),
+    operations_examined: z
+      .number()
+      .int()
+      .nonnegative()
+      .max(Number.MAX_SAFE_INTEGER),
+    value_scope: z.string(),
+    machine_call_equivalence: z.literal("unknown"),
+  }),
+});
+const parseCall = rowParser(nativeCall, {
+  arguments: (row) => nativeCall.shape.arguments.element.parse(row),
+});
+const parseFacet = rowParser(callFacet, {
+  items: parseCall,
+  values: (row) => value.parse(row),
+});
+export const readCallFacet = (fn) =>
+  readNativeFacet(fn, "atomic_calls", "recovered calls", (raw) => {
+    requireFact(
+      raw && typeof raw === "object" && raw.schema_version === 1,
+      "Unsupported native call facet schema; expected 1",
+    );
+    for (const [name, limit] of [
+      ["items", 12000],
+      ["values", 36000],
+    ])
+      requireFact(
+        Array.isArray(raw[name]) && raw[name].length <= limit,
+        `Native call ${name} exceeds capacity or has invalid shape`,
+      );
+    requireFact(
+      raw.items.reduce(
+        (sum, row) =>
+          sum + (Array.isArray(row?.arguments) ? row.arguments.length : 0),
+        0,
+      ) <= 50000,
+      "Native call inputs exceed capacity",
+    );
+    return parseFacet(raw);
+  });

experiments/atomic-codemode/native-call-views.mjs

@@ -0,0 +1,162 @@
+import { callCoverage } from "./native-calls.mjs";
+import { requireFact, requireSameNativeValue } from "./native-contract.mjs";
+import { uniqueNative, readNativeIdentities } from "./native-reads.mjs";
+import { freezeFacts } from "./fact-reads.mjs";
+
+export function indexCallFacet(facts) {
+  const calls = uniqueNative(facts.items),
+    values = uniqueNative(facts.values),
+    used = new Set();
+  requireFact(
+    calls.size <= facts.coverage.operations_examined,
+    "Native call count exceeds examined operation coverage",
+  );
+  const reference = (id) => {
+    requireFact(
+      id === null || values.has(id),
+      "Dangling native call-boundary value",
+    );
+    if (id !== null) used.add(id);
+  };
+  for (const row of calls.values()) {
+    requireFact(
+      row.id.split(":")[0] === row.origin_address,
+      "Native call origin/identity mismatch",
+    );
+    reference(row.target_value_id);
+    reference(row.output);
+    requireFact(
+      row.output === null || values.get(row.output).definition === row.id,
+      "Native call output/definition mismatch",
+    );
+    for (const [i, a] of row.arguments.entries()) {
+      reference(a.value_id);
+      requireFact(
+        a.input_index === i + 1 &&
+          (a.storage === null) === (a.storage_authority === null),
+        "Native call input/storage binding mismatch",
+      );
+    }
+    requireFact(
+      row.spec_available
+        ? row.prototype !== null
+        : row.target === null &&
+            row.target_name === null &&
+            row.prototype === null &&
+            row.stack_offset === null &&
+            row.arguments.every((a) => a.storage === null),
+      "Native missing call specification has invented facts",
+    );
+  }
+  requireFact(
+    used.size === values.size,
+    "Native call boundary contains unrelated values",
+  );
+  for (const value of values.values())
+    if (value.definition !== null && calls.has(value.definition))
+      requireFact(
+        calls.get(value.definition).output === value.id,
+        "Native boundary value/known-call definition mismatch",
+      );
+  return { calls, values };
+}
+export function validateCallGraph(callIndex, graphIndex) {
+  const { calls, values } = callIndex,
+    { operations } = graphIndex;
+  for (const row of calls.values()) {
+    const op = operations.get(row.id);
+    requireFact(
+      op &&
+        op.address === row.origin_address &&
+        op.block === row.block &&
+        op.opcode === row.opcode &&
+        op.is_dead === row.is_dead &&
+        op.output === row.output,
+      "Native call/graph ownership mismatch",
+    );
+    requireFact(
+      row.target_value_id === (op.inputs[0]?.value_id ?? null) &&
+        row.arguments.length === Math.max(0, op.inputs.length - 1) &&
+        row.arguments.every((a, i) => a.value_id === op.inputs[i + 1].value_id),
+      "Native call/graph operand mismatch",
+    );
+  }
+  for (const value of values.values())
+    requireSameNativeValue(
+      value,
+      graphIndex.values.get(value.id),
+      "Native call/graph value mismatch",
+    );
+  for (const op of operations.values())
+    requireFact(
+      !["CPUI_CALL", "CPUI_CALLIND"].includes(op.opcode) || calls.has(op.id),
+      "Native call operation omitted from call coverage",
+    );
+}
+const summarizeCalls = (calls, coverage) => {
+  const counts = {
+    CPUI_CALL: { known_target_count: 0, unknown_target_count: 0 },
+    CPUI_CALLIND: { known_target_count: 0, unknown_target_count: 0 },
+  };
+  for (const row of calls.values())
+    counts[row.opcode][
+      row.target === null ? "unknown_target_count" : "known_target_count"
+    ]++;
+  return freezeFacts({
+    count: calls.size,
+    by_opcode: Object.entries(counts).map(([opcode, row]) => ({
+      opcode,
+      ...row,
+    })),
+    coverage,
+  });
+};
+export function createCallViews(facts, indexes) {
+  const { calls, values } = indexes;
+  freezeFacts(facts);
+  const coverage = freezeFacts({
+    ...callCoverage,
+    ...facts.coverage,
+    definition_operations: "not collected in call-boundary facet",
+  });
+  const valueCoverage = freezeFacts({
+    ...coverage,
+    scope: facts.coverage.value_scope,
+  });
+  return {
+    calls: (input) => ({
+      ...readNativeIdentities(
+        calls,
+        {
+          fields: [
+            "spec_available",
+            "opcode",
+            "is_dead",
+            "target",
+            "target_name",
+            "arguments",
+            "output",
+          ],
+          ...input,
+          ...(input.fields && {
+            fields: ["spec_available", "opcode", "is_dead", ...input.fields],
+          }),
+        },
+        (r) =>
+          !input.origin_addresses ||
+          input.origin_addresses.includes(r.origin_address),
+      ),
+      coverage,
+    }),
+    values: (input) => ({
+      ...readNativeIdentities(
+        values,
+        input,
+        (r) => input.is_input === undefined || r.is_input === input.is_input,
+      ),
+      coverage: valueCoverage,
+    }),
+    summary: summarizeCalls(calls, coverage),
+    value_count: values.size,
+  };
+}