Merge branch 'cursor/rea-limit-inventory-01f4' into cursor/setting-effect-field-01f4

Keep the file-level Ghidra bridge context assertion. Main removed the empty suite.

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

@@ -25,7 +25,7 @@ jobs:
           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 >> "$GITHUB_ENV"
-      # The bridge probe imports only source-owned ARM/Thumb bytes and never
+      # The bridge probe imports only source-owned ARM, Thumb and MIPS bytes and never
       # executes target code. All SDK/project state stays in runner storage.
       - name: Acquire checksum-verified Ghidra SDK in isolated runner storage
         shell: bash
@@ -46,7 +46,7 @@ jobs:
           java_home_variable="JAVA_HOME_${GHIDRA_JAVA_MAJOR}_X64"
           test -n "${!java_home_variable}"
           JAVA_HOME="${!java_home_variable}" npm run verify:ghidra:bridge-sources -- --verified-sdk
-      - name: Inspect real ARM and Thumb instruction context through the bridge
+      - name: Inspect real ARM, Thumb and MIPS instruction context through the bridge
         shell: bash
         run: |
           java_home_variable="JAVA_HOME_${GHIDRA_JAVA_MAJOR}_X64"
@@ -58,10 +58,21 @@ jobs:
             -import "${probe_root}/arm-thumb.bin" -loader BinaryLoader -processor ARM:LE:32:v7 \
             -noanalysis -readOnly -deleteProject \
             -scriptPath "${GITHUB_WORKSPACE}/tests/conformance/ghidra;${GITHUB_WORKSPACE}/bridge/ghidra" \
-            -postScript ReaArmContextProbe.java "${GITHUB_WORKSPACE}/bridge/ghidra" \
+            -postScript ReaInstructionContextProbe.java "${GITHUB_WORKSPACE}/bridge/ghidra" \
             > "${probe_root}/output.log" 2>&1 || { cat "${probe_root}/output.log"; exit 1; }
           cat "${probe_root}/output.log"
           grep -F 'REA_ARM_CONTEXT_OK arm=0 thumb=1 unavailable=null (injected provider value)' "${probe_root}/output.log"
+          mkdir -p "${probe_root}/mips-project"
+          # MIPS32 addiu v0,a0,3; MIPS16 li v0,90. These bytes are never executed.
+          printf '\003\000\202\044\132\152\000\000' > "${probe_root}/mips.bin"
+          timeout 180s "${GHIDRA_INSTALL_DIR}/support/analyzeHeadless" "${probe_root}/mips-project" MipsContext \
+            -import "${probe_root}/mips.bin" -loader BinaryLoader -processor MIPS:LE:32:default \
+            -noanalysis -readOnly -deleteProject \
+            -scriptPath "${GITHUB_WORKSPACE}/tests/conformance/ghidra;${GITHUB_WORKSPACE}/bridge/ghidra" \
+            -postScript ReaInstructionContextProbe.java "${GITHUB_WORKSPACE}/bridge/ghidra" mips \
+            > "${probe_root}/mips-output.log" 2>&1 || { cat "${probe_root}/mips-output.log"; exit 1; }
+          cat "${probe_root}/mips-output.log"
+          grep -F 'REA_MIPS_CONTEXT_OK standard=0 mips16=1' "${probe_root}/mips-output.log"
       # Each probe imports source-owned bytes at 0x1000, away from the Cortex-M
       # vector symbols that ARMCortex.pspec defines at 0, and decodes them from
       # the language's default context, so M-profile Thumb state must hold at

CHANGELOG.md

@@ -1,5 +1,7 @@
 # Changelog
 
+The next release changes analysis-setting output (#9885). `update_analysis_settings`, `inspect_analysis_settings`, and limit errors no longer include `rea_restart_required`, `restart_required`, `restartRequired`, or `effect`. Read `update_mode` (`runtime`, `restart`, or `parameter`) and `pending_restart`. `runtime` means the setter can change the active value. `restart` means this process keeps the value it started with. `parameter` means this call supplied the value, so `update_analysis_settings` cannot change it; artifact reader limits, web-network capture budgets, and an explicit `analysis_timeout_ms`, `max_heap_mb`, or `command_timeout_ms` use it. There is no compatibility alias. Release Please should copy this migration into that release's breaking-changes section.
+
 ## [6.8.0](https://github.com/morluto/rea/compare/rea-agents-6.7.0...rea-agents-6.8.0) (2026-10-25)
 
 ### Highlights

docs/.vitepress/documentation.ts

@@ -358,6 +358,11 @@ export const documentationSections = [
         source: "design/limits-and-recovery.md",
         route: "/design/limits-and-recovery",
       },
+      {
+        text: "REA limit inventory",
+        source: "design/rea-limit-inventory.md",
+        route: "/design/rea-limit-inventory",
+      },
       {
         text: "MCP tool design",
         source: "development/tool-design.md",

docs/README.md

@@ -111,6 +111,7 @@ product pages. The VitePress documentation source is this `docs/` tree.
 ## Development and verification
 
 - [Settings and recovery](design/limits-and-recovery.md)
+- [REA limit inventory](design/rea-limit-inventory.md)
 - [Flutter budget calibration](verification/flutter-budget-calibration.md)
 - [Source-map budget calibration](verification/source-map-budget-calibration.md)
 - [Source-location delivery qualification](verification/source-location-delivery.md)

docs/design/limits-and-recovery.md

@@ -5,7 +5,9 @@ who owns each value, when an update takes effect, how providers enforce it, and
 how to extend the system. The public tool contract is in
 [MCP contracts](../reference/mcp-contracts.md#editable-analysis-settings); recorded runtime
 checks and their limitations are in the
-[verification report](../verification/runtime-analysis-settings.md).
+[verification report](../verification/runtime-analysis-settings.md). The
+[limit inventory](rea-limit-inventory.md) lists every numeric `REA_` limit, which
+of them are runtime settings, and which stay restart-only.
 
 ## Problem and design boundary
 
@@ -255,8 +257,8 @@ are runtime-wide, not scoped to a target or an individual connection.
 ### Failure and retry
 
 An ASAR admission error supplies `details.limit` with the exact setting,
-effective value, units, bounds, scope, application point, restart requirement
-and, when known, required bytes and value source. The agent can immediately call:
+effective value, units, bounds, scope, application point, `update_mode`,
+`pending_restart` and, when known, required bytes and value source. The agent can immediately call:
 
 ```json
 {
@@ -273,13 +275,27 @@ that argument; raising a runtime default cannot override it.
 
 The resource owner constructs `details.limit` from the registry plus the budget
 it actually enforced. Metadata includes `agent_editable: true`,
-`scope: "rea-runtime"`, and `rea_restart_required: false`. Optional
+`scope: "rea-runtime"`, `update_mode`, and `pending_restart`. `update_mode`
+is `runtime` when `update_analysis_settings` can change the active value,
+`restart` when this process keeps the value it started with, and `parameter`
+when this call supplied the value. An explicit `analysis_timeout_ms`,
+`max_heap_mb`, or `command_timeout_ms`, plus artifact reader limits and
+web-network capture budgets, use `parameter`. `pending_restart` is false when
+that reported value is the one this process is using. Optional
 `required_value`, `source`, and `parameter` fields are supplied only where known
 or applicable. Do not infer a source because a value happens to equal a default.
 Error copying, attribution, redaction, and delivery must preserve these facts.
 Human-readable remediation explains the same failure; clients should not need
 to parse that sentence to identify an editable setting.
 
+This output is breaking. Results and errors no longer carry the previous
+restart boolean or a single effect field, and there is no compatibility alias.
+Read `update_mode` to see whether the setter can change the active value, and
+`pending_restart` to see whether a configured value is waiting. A configuration
+error for a Windows ChangeTime setting carries those two fields on
+`details.settings`, not only in the constraint sentence. The removed field names
+are listed in the changelog.
+
 `scope` and `applies_at` describe a continuing caller, such as an MCP
 connection. A one-shot invocation, such as a CLI command, has released its
 provider session and exited its runtime before its caller reads the failure,
@@ -288,8 +304,10 @@ so its adapter projects with an explicit `"invocation"` caller lifetime.
 `scope: "rea-invocation"`, `applies_at: "next_invocation"` and the launch
 `environment` seed when one exists. `agent_editable: true` still holds, because
 the caller chooses that environment value or per-call `parameter` for the new
-invocation; `rea_restart_required: false` means no running REA process needs a
-restart. A producer keeps guidance that holds only while its caller retains the
+invocation; `update_mode: "runtime"` and `pending_restart: false` mean the
+reported value is active in this process. `update_mode: "restart"` means
+`update_analysis_settings` does not change it. `update_mode: "parameter"` means
+this call supplied the value. A producer keeps guidance that holds only while its caller retains the
 session, such as "the Ghidra session remains usable" or a same-session runtime
 update, in `continuingCallerGuidance`; the shared presentation adds it only for
 a continuing caller. The producer's own message and remediation must stay true
@@ -499,7 +517,10 @@ own `expanded_bytes` controls its independently enforced bundle expansion.
 
 Failures preserve `details.reader_limit` with the exact parameter, effective
 value, units, supported range, required value when known, next-call scope and
-`can_increase`. When `can_increase` is true, raise that field and retry the same
+`can_increase`. `update_mode` is `parameter`: the caller selects the value on
+the next call, and `update_analysis_settings` does not accept it.
+`pending_restart` is false. Web-network capture budgets use the same mode.
+When `can_increase` is true, raise that field and retry the same
 complete artifact. At the supported maximum, or when the required representation
 exceeds it, the failure is not retryable by increasing that budget; inspect the
 declared dimensions and input validity, or explicitly choose a suitable reader.

docs/design/rea-limit-inventory.md

@@ -0,0 +1,229 @@
+# REA limit inventory
+
+Audit for [#9881](https://github.com/morluto/rea/issues/9881). Counts below are
+from this revision of `src`. A numeric `REA_` limit is a name under `src` (tests
+excluded) that carries a byte, time, count, heap, budget, deadline, or maximum.
+`scripts/verify/rea-numeric-limits.mjs` fails the static checks when one of those
+names is neither an environment seed of
+[`ANALYSIS_SETTINGS`](https://github.com/morluto/rea/blob/main/src/domain/analysisSettingDefinitions.ts) nor an
+entry in [`numericLimitRegistry.ts`](https://github.com/morluto/rea/blob/main/src/domain/numericLimitRegistry.ts)
+with a reason.
+
+`update_mode` is `runtime` when `update_analysis_settings` can change the active
+value, `restart` when this process keeps the value it started with, and
+`parameter` when this call supplied the value.
+`pending_restart` is true only when a configured value is not the one this
+process is using. A rejected Windows ChangeTime launch value is not active, so
+that failure reports `pending_restart: true`. Every other value reported here
+is the one in force, so `pending_restart` is false. Prose that only says to
+restart REA is marked `prose`. `mcp_server_restart_required` is a
+server-identity alignment state in `src/serverIdentity.ts`, not a setting field.
+
+Three numeric limits are absent from this tree and are being moved into the
+runtime system on their own PRs. This audit does not change them:
+
+| Name                                         | Tracking |
+| -------------------------------------------- | -------- |
+| `REA_NTFS_MFT_MAX_FILE_BYTES`                | #9868    |
+| `REA_REGF_HIVE_MAX_FILE_BYTES`               | #9865    |
+| `REA_WEB_BUNDLE_ENDPOINT_LITERAL_CANDIDATES` | #9836    |
+
+When those names land in `src`, the static check requires them to be analysis
+settings. They are not on the restart-only list.
+
+## Counts
+
+| Category                                                              |                                                                        Count |
+| --------------------------------------------------------------------- | ---------------------------------------------------------------------------: |
+| Runtime analysis settings                                             |                  17 (14 have an `REA_` seed; 3 are per-call parameters only) |
+| Per-call numeric `REA_` limits still read from the launch environment |                                                                           11 |
+| Restart-only numeric `REA_` limits                                    |                                                                           15 |
+| Numeric `REA_` names discovered in `src`                              |                                                                           40 |
+| Non-numeric `REA_` settings kept restart-only                         | 53 caller settings, plus process-protocol names that are not caller settings |
+| Hardcoded ceilings left for follow-up                                 |                        grouped below; none converted in the inventory change |
+
+## Runtime analysis settings
+
+Already adjustable with `update_analysis_settings`. Each call reads the snapshot
+accepted for that operation. Launch source is `default` or `setting`; a runtime
+override is `source: "runtime"`. The structured field is `update_mode: "runtime"` and `pending_restart: false`.
+`applies_at` is the finer point
+(`next_operation`, `next_worker`, or `next_startup`).
+
+| Setting                                      | Environment seed                                 | Owner                                                         | Default                                   |            Floor |                                Maximum | Read                  |
+| -------------------------------------------- | ------------------------------------------------ | ------------------------------------------------------------- | ----------------------------------------- | ---------------: | -------------------------------------: | --------------------- |
+| `asar.max_header_bytes`                      | `REA_ASAR_MAX_HEADER_BYTES`                      | `src/domain/analysisResourceLimits.ts`                        | 16777216                                  |                1 |                              268435456 | per operation         |
+| `asar.max_listing_bytes`                     | `REA_ASAR_MAX_LISTING_BYTES`                     | `src/domain/analysisResourceLimits.ts`                        | 67108864                                  |                1 |              `Number.MAX_SAFE_INTEGER` | per operation         |
+| `ghidra.value_flow_max_operations`           | `REA_GHIDRA_VALUE_FLOW_MAX_OPERATIONS`           | `src/domain/analysisResourceLimits.ts`                        | 3000                                      |                1 |                             2147483647 | per operation         |
+| `ghidra.value_flow_max_inputs_per_operation` | `REA_GHIDRA_VALUE_FLOW_MAX_INPUTS_PER_OPERATION` | `src/domain/analysisResourceLimits.ts`                        | 64                                        |                1 |                             2147483647 | per operation         |
+| `ghidra.value_flow_max_edges`                | `REA_GHIDRA_VALUE_FLOW_MAX_EDGES`                | `src/domain/analysisResourceLimits.ts`                        | 12000                                     |                1 |                             2147483647 | per operation         |
+| `ghidra.decompile_max_payload_mib`           | `REA_GHIDRA_DECOMPILE_MAX_PAYLOAD_MIB`           | `src/domain/analysisResourceLimits.ts`                        | unset (`null`, unbounded payload policy)  |                1 |                                   2047 | per operation         |
+| `ghidra.startup_timeout_ms`                  | `REA_GHIDRA_STARTUP_TIMEOUT_MS`                  | `src/domain/analysisResourceLimits.ts`                        | 330000                                    |                1 |                             2147483647 | next provider startup |
+| `hopper.startup_timeout_ms`                  | `REA_HOPPER_STARTUP_TIMEOUT_MS`                  | `src/domain/analysisResourceLimits.ts`                        | 120000                                    |                1 |                             2147483647 | next provider startup |
+| `ghidra.decompile_timeout_seconds`           | `REA_GHIDRA_DECOMPILE_TIMEOUT_SECONDS`           | `src/domain/analysisResourceLimits.ts`                        | unset (`null`, decompile stays unbounded) |                1 | `MAX_GHIDRA_DECOMPILE_TIMEOUT_SECONDS` | per operation         |
+| `jadx.operation_timeout_ms`                  | `REA_JADX_OPERATION_TIMEOUT_MS`                  | `src/domain/analysisResourceLimits.ts`                        | 120000                                    |                1 |                             2147483647 | per operation         |
+| `regex.matching_deadline_ms`                 | `REA_REGEX_MATCHING_DEADLINE_MS`                 | `src/domain/analysisResourceLimits.ts`                        | 5000                                      |                1 |                             2147483647 | per operation         |
+| `regex.worker_heap_request_mib`              | `REA_REGEX_WORKER_HEAP_REQUEST_MIB`              | `src/domain/analysisResourceLimits.ts`                        | 128                                       |                7 |                                  16384 | next worker           |
+| `capture.decoder_old_generation_mib`         | `REA_CAPTURE_DECODER_OLD_GENERATION_MIB`         | `src/domain/analysisResourceLimits.ts`                        | 192                                       | semi-space floor |                                  16384 | next worker           |
+| `javascript.analysis_timeout_ms`             | none (`analysis_timeout_ms`)                     | `src/domain/javascript/javascriptAnalysisResourceControls.ts` | owner default                             |                1 |                             2147483647 | per operation         |
+| `javascript.max_heap_mb`                     | none (`max_heap_mb`)                             | `src/domain/javascript/javascriptAnalysisResourceControls.ts` | owner default                             |      owner floor |                          owner maximum | next worker           |
+| `wasm.command_timeout_ms`                    | none (`command_timeout_ms`)                      | `src/domain/wasm/wasmArtifact.ts`                             | owner default                             |                1 |                          owner maximum | per operation         |
+| `native.matching_representation_bytes`       | `REA_NATIVE_MATCHING_REPRESENTATION_BYTES`       | `src/domain/analysisResourceLimits.ts`                        | 100663296                                 |                1 |                             1073741824 | per operation         |
+
+An out-of-range `update_analysis_settings` value is rejected by the setting's
+schema. Invalid launch seeds that the global environment parser accepts as
+provider-local stay inspectable with `source: "invalid"`.
+
+## Per-call limits still outside the runtime system
+
+These are checked when the work runs. They keep the `REA_` variable, default,
+floor, and maximum, and the migration reads the accepted snapshot instead of a
+value copied at process start. Until that lands they are `policy: "per-call"` in
+the registry so the static check still requires a reason.
+
+| Environment                                | Owner                                       |                 Default | Floor |    Maximum | Read today                                                                            | Restart field                                                                   |
+| ------------------------------------------ | ------------------------------------------- | ----------------------: | ----: | ---------: | ------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------- |
+| `REA_BRIDGE_MAX_REQUEST_BYTES`             | `src/config/bridgeMaxRequestBytes.ts`       |                16777216 |  7464 |  536870887 | each bridge request, from the launch value                                            | prose                                                                           |
+| `REA_DEBUGGER_COMMAND_DEADLINE_MS`         | `src/domain/analysisResourceLimits.ts`      |                  120000 |     1 | 2147483647 | copied when the debugger manager is constructed; applied to each command              | `update_mode: "restart"` and `pending_restart: false`                           |
+| `REA_DEBUGGER_MAX_COMMAND_OUTPUT_BYTES`    | `src/domain/analysisResourceLimits.ts`      |                16777216 |     1 |   89303718 | copied when the debugger manager is constructed; applied to each command              | `update_mode: "restart"` and `pending_restart: false`                           |
+| `REA_HOPPER_CALLER_SCAN_BUDGET_MS`         | `src/domain/analysisResourceLimits.ts`      |                    2000 |     1 | 2147483647 | each caller-scan operation, from config resolved at launch                            | prose                                                                           |
+| `REA_JEB_MCP_TIMEOUT_MS`                   | `src/jeb/JebMcpConnection.ts`               | 60000 (MCP SDK default) |     1 | 2147483647 | captured when the JEB connection is created; applied to discovery and each tools/call | prose                                                                           |
+| `REA_KUNA_PROJECT_TIMEOUT_MS`              | `src/kuna/KunaConfiguration.ts`             |                 1800000 |     1 | 2147483647 | project export and whole-program scans, from the launch configuration                 | prose                                                                           |
+| `REA_KUNA_TIMEOUT_MS`                      | `src/kuna/KunaConfiguration.ts`             |                  120000 |     1 | 2147483647 | each function decompile, from the launch configuration                                | prose                                                                           |
+| `REA_PWNTOOLS_MAX_INPUT_BYTES`             | `src/native/pwntools/PwntoolsInputLimit.ts` |                33554432 |     1 | 1073741824 | each pwntools decode, unless the call passes `producer_limits.input_bytes`            | prose                                                                           |
+| `REA_WINDOWS_CHANGE_TIME_DRIFT_MS`         | `src/config/windowsChangeTimeSettle.ts`     |                    2000 |     1 |      60000 | each ChangeTime observation                                                           | `update_mode: "restart"` and `pending_restart: true` on an invalid launch value |
+| `REA_WINDOWS_CHANGE_TIME_SETTLE_POLL_MS`   | `src/config/windowsChangeTimeSettle.ts`     |                      25 |     1 |       1000 | each ChangeTime observation                                                           | `update_mode: "restart"` and `pending_restart: true` on an invalid launch value |
+| `REA_WINDOWS_CHANGE_TIME_SETTLE_MAX_POLLS` | `src/config/windowsChangeTimeSettle.ts`     |                      40 |     1 |       1000 | each ChangeTime observation                                                           | `update_mode: "restart"` and `pending_restart: true` on an invalid launch value |
+
+## Restart-only numeric limits
+
+These stay out of `update_analysis_settings`.
+
+| Environment                         | Owner                                  |             Default |    Floor |                   Maximum | Read                                        | Why restart-only                                                                                            |
+| ----------------------------------- | -------------------------------------- | ------------------: | -------: | ------------------------: | ------------------------------------------- | ----------------------------------------------------------------------------------------------------------- |
+| `REA_DEBUGGER_MAX_PENDING_COMMANDS` | `src/domain/analysisResourceLimits.ts` |                  16 |        1 |                       256 | debugger manager construction               | Queue capacity of the manager created at startup. `update_mode: "restart"` and `pending_restart: false`.    |
+| `REA_DEBUGGER_MAX_SESSIONS`         | `src/domain/analysisResourceLimits.ts` |                   8 |        1 |                        64 | debugger manager construction               | Session-pool size of the manager created at startup. `update_mode: "restart"` and `pending_restart: false`. |
+| `REA_HOPPER_DEMO_DIALOG_TIMEOUT_MS` | `src/domain/hopperStartupFailure.ts`   |               10000 |        1 |                2147483647 | Linux private-display adapter start         | The helper reads the process environment. Prose restart.                                                    |
+| `REA_JADX_ACTIVE_PROCESSOR_COUNT`   | `src/android/JadxConfiguration.ts`     | unset (JVM default) |        1 |                2147483647 | JADX JVM launch                             | `-XX:ActiveProcessorCount` is fixed for that JVM. Prose on invalid values.                                  |
+| `REA_JADX_HEAP_MIB`                 | `src/android/JadxConfiguration.ts`     | unset (JVM default) |        1 | `Number.MAX_SAFE_INTEGER` | JADX JVM launch                             | `-Xmx` is fixed for that JVM. Prose on invalid values.                                                      |
+| `REA_KUNA_CACHE_MAX_BYTES`          | `src/kuna/KunaConfiguration.ts`        |          4294967296 |  1048576 | `Number.MAX_SAFE_INTEGER` | Kuna cache construction                     | The cache is sized once for the provider. Prose restart.                                                    |
+| `REA_KUNA_JOBS`                     | `src/kuna/KunaConfiguration.ts`        |                   1 |        1 |                        32 | Kuna run-slot construction                  | Worker-pool width for the provider. Prose restart.                                                          |
+| `REA_MCP_DOCTOR_DEADLINE_MS`        | `src/config/mcpStartupBudgets.ts`      |               30000 |        1 |                2147483647 | MCP server startup                          | Startup phase of the MCP server itself. Prose.                                                              |
+| `REA_MCP_FIRST_CATALOG_BUDGET_MS`   | `src/config/mcpStartupBudgets.ts`      |               20000 |        1 |                2147483647 | MCP server startup                          | Startup phase of the MCP server itself. Prose.                                                              |
+| `REA_MCP_INITIALIZE_BUDGET_MS`      | `src/config/mcpStartupBudgets.ts`      |               15000 |        1 |                2147483647 | MCP server startup                          | Startup phase of the MCP server itself. Prose.                                                              |
+| `REA_MCP_MAX_INPUT_ELEMENTS`        | `src/config/mcpInputBudget.ts`         |                1024 |        1 | `Number.MAX_SAFE_INTEGER` | MCP server creation                         | Argument admission for the server process. Prose restart.                                                   |
+| `REA_MCP_MAX_RESPONSE_BYTES`        | `src/config/mcpResponseBudget.ts`      | 10485760 when unset | 10485760 | `Number.MAX_SAFE_INTEGER` | MCP server creation                         | Stdio receive buffer and the paired client buffer. Prose restart.                                           |
+| `REA_MCP_TOOLS_LIST_BUDGET_MS`      | `src/config/mcpStartupBudgets.ts`      |                5000 |        1 |                2147483647 | MCP server startup                          | Startup phase of the MCP server itself. Prose.                                                              |
+| `REA_RECONNECT_TIMEOUT_SECONDS`     | `src/hopper/BridgeLauncher.ts`         |                  30 |       30 |                        30 | written into the Hopper bridge helper       | Not a caller setting. It is `HOPPER_SHUTDOWN_TIMEOUT_MS / 1000`.                                            |
+| `REA_SWIFTC_COMPILE_TIMEOUT_MS`     | `src/config/providerStartupTimeout.ts` |              120000 |        1 |                2147483647 | Darwin ownership-helper reader construction | Captured at process start for that helper compile. Prose restart.                                           |
+
+## Restart-only non-numeric settings
+
+No floor or maximum. They are read from the launch environment (startup). None of them set `effect`. Recovery text tells the caller to change the launch
+environment, which takes effect on the next process. They stay
+restart-only because they choose a command, a binary, a filesystem root, the
+browser sandbox, secret patterns, the provider, or the shape of the MCP server
+or logger.
+
+| Name                                 | Owner                                              | Why                                         |
+| ------------------------------------ | -------------------------------------------------- | ------------------------------------------- |
+| `REA_ADB_PATH`                       | `src/adb/AdbConfiguration.ts`                      | Caller-selected executable path.            |
+| `REA_AGENT_ENV_COMMAND`              | `src/application/DesktopBuildDoctor.ts`            | Caller-selected command.                    |
+| `REA_ANALYSIS_PROVIDER`              | `src/config/environment.ts`                        | Provider choice for the process.            |
+| `REA_APKTOOL_COMMAND`                | `src/apktool/ApktoolConfiguration.ts`              | Caller-selected command.                    |
+| `REA_BINWALK_COMMAND`                | `src/application/SetupRegistrationEnvironment.ts`  | Caller-selected command.                    |
+| `REA_BROWSER_EXECUTABLE`             | `src/application/CapabilityInventory.ts`           | Caller-selected browser binary.             |
+| `REA_BROWSER_NO_SANDBOX`             | `src/application/SetupRegistrationEnvironment.ts`  | Browser sandbox policy.                     |
+| `REA_CAPINFOS_COMMAND`               | `src/pcap/PcapConfiguration.ts`                    | Caller-selected command.                    |
+| `REA_CUTTER_BRIDGE_DIR`              | `src/application/SetupRegistrationEnvironment.ts`  | Filesystem root for the Cutter bridge.      |
+| `REA_DARWIN_MODULE_CACHE_PATH`       | `src/application/SetupRegistrationEnvironment.ts`  | Filesystem path for the Swift module cache. |
+| `REA_DARWIN_SWIFTC_PATH`             | `src/application/SetupRegistrationEnvironment.ts`  | Caller-selected compiler path.              |
+| `REA_DCONF_COMMAND`                  | `src/desktop/DconfDatabaseReader.ts`               | Caller-selected command.                    |
+| `REA_DOCKER_COMMAND`                 | `src/application/DesktopBuildDoctor.ts`            | Caller-selected command.                    |
+| `REA_ELECTRON_CAPTURE_ARGV`          | `src/browser/PlaywrightElectronActiveProvider.ts`  | Caller-selected process arguments.          |
+| `REA_EVM_PRLIMIT_COMMAND`            | `src/application/SetupRegistrationEnvironment.ts`  | Caller-selected command.                    |
+| `REA_FIRMWARE_PRLIMIT_COMMAND`       | `src/application/SetupRegistrationEnvironment.ts`  | Caller-selected command.                    |
+| `REA_GDB_COMMAND`                    | `src/application/SetupRegistrationEnvironment.ts`  | Caller-selected debugger binary.            |
+| `REA_GHIDRA_CHECKPOINT_DIR`          | `src/config/environment.ts`                        | Filesystem root.                            |
+| `REA_GHIDRA_COMPILER_SPEC_ID`        | `src/config/environment.ts`                        | Provider analysis choice.                   |
+| `REA_GHIDRA_DISABLED_ANALYZERS`      | `src/config/environment.ts`                        | Provider analysis choice.                   |
+| `REA_GHIDRA_LANGUAGE_ID`             | `src/config/environment.ts`                        | Provider analysis choice.                   |
+| `REA_GHIDRA_NATIVEAOT_JAR`           | `src/config/environment.ts`                        | Caller-selected binary path.                |
+| `REA_GHIDRA_PROGRAM_NAME`            | `src/config/environment.ts`                        | Provider project selection.                 |
+| `REA_GHIDRA_PROJECT_PATH`            | `src/config/environment.ts`                        | Filesystem root.                            |
+| `REA_GHIDRA_SEED_FILE`               | `src/config/environment.ts`                        | Filesystem path.                            |
+| `REA_IDA_MCP_CONFIG`                 | `src/config/environment.ts`                        | Filesystem path of the IDA MCP config.      |
+| `REA_ILSPY_CMD_PATH`                 | `src/config/environment.ts`                        | Caller-selected executable path.            |
+| `REA_JADX_MCP_JAR`                   | `src/android/JadxConfiguration.ts`                 | Caller-selected binary path.                |
+| `REA_JAVASCRIPT_PRLIMIT_COMMAND`     | `src/application/SetupRegistrationEnvironment.ts`  | Caller-selected command.                    |
+| `REA_JEB_MCP_URL`                    | `src/config/environment.ts`                        | Provider endpoint.                          |
+| `REA_KUNA_CACHE_DIR`                 | `src/kuna/KunaConfiguration.ts`                    | Filesystem root.                            |
+| `REA_KUNA_COMMAND`                   | `src/kuna/KunaConfiguration.ts`                    | Caller-selected executable path.            |
+| `REA_LOG_LEVEL`                      | `src/config/environment.ts`                        | Logger configuration for the process.       |
+| `REA_MANAGED_LINUX_HOPPER_LAUNCHER`  | `src/application/HopperLauncherDiscovery.ts`       | Caller-selected launcher path.              |
+| `REA_MCP_INPUT_SCHEMA_PROFILE`       | `src/config/environment.ts`                        | MCP server schema shape.                    |
+| `REA_MCP_OUTPUT_SCHEMA_PROFILE`      | `src/config/environment.ts`                        | MCP server schema shape.                    |
+| `REA_MCP_TOOL_CATALOG_PROFILE`       | `src/config/environment.ts`                        | MCP server catalog shape.                   |
+| `REA_MCP_TOOL_FAMILIES`              | `src/config/environment.ts`                        | Which tools the MCP server registers.       |
+| `REA_MITMDUMP_COMMAND`               | `src/application/SetupRegistrationEnvironment.ts`  | Caller-selected command.                    |
+| `REA_OBJDUMP_COMMAND`                | `src/application/SetupRegistrationEnvironment.ts`  | Caller-selected command.                    |
+| `REA_PWNDBG_GDB`                     | `src/native/pwndbg/PwndbgCoreContext.ts`           | Caller-selected debugger binary.            |
+| `REA_PWNDBG_GDBINIT`                 | `src/native/pwndbg/PwndbgCoreContext.ts`           | Caller-selected file path.                  |
+| `REA_PWNDBG_VENV_PATH`               | `src/native/pwndbg/PwndbgCoreContext.ts`           | Filesystem root.                            |
+| `REA_PWNTOOLS_PYTHON`                | `src/native/pwntools/PwntoolsDecoder.ts`           | Caller-selected interpreter path.           |
+| `REA_REFERENCE_SECRET_PATTERNS_JSON` | `src/config/environment.ts`                        | Secret-pattern policy.                      |
+| `REA_RIZIN_COMMAND`                  | `src/rizin/RizinDebugSessionManager.ts`            | Caller-selected debugger binary.            |
+| `REA_SIGROK_COMMAND`                 | `src/signals/SigrokRecordingProvider.ts`           | Caller-selected command.                    |
+| `REA_SIGROK_PRLIMIT_COMMAND`         | `src/signals/SigrokRecordingProvider.ts`           | Caller-selected command.                    |
+| `REA_TSHARK_COMMAND`                 | `src/pcap/PcapConfiguration.ts`                    | Caller-selected command.                    |
+| `REA_TSHARK_PRLIMIT_COMMAND`         | `src/pcap/PcapNativeBound.ts`                      | Caller-selected command.                    |
+| `REA_UNBLOB_COMMAND`                 | `src/application/SetupRegistrationEnvironment.ts`  | Caller-selected command.                    |
+| `REA_WABT_BIN_DIRECTORY`             | `src/wasm/WabtArtifactProvider.ts`                 | Filesystem root of the WABT tools.          |
+| `REA_WAKARU_COMMAND`                 | `src/application/SetupRegistrationEnvironment.ts`  | Caller-selected command.                    |
+| `REA_WORKFLOW_PROVIDER`              | `src/application/DirectAnalysis.ts`                | Provider choice.                            |
+| `REA_XPROP_COMMAND`                  | `src/application/desktop/DesktopCaptureFailure.ts` | Caller-selected command.                    |
+
+Names that look like `REA_` but are process protocol, not caller settings:
+`REA_BUILD_COMMIT`, `REA_BUILD_COMPUTER_REA_ARCHIVE`, `REA_INSTALL_SCRIPT`,
+`REA_OWNS_PROCESS_LIFETIME`, `REA_PAYLOAD`, `REA_PROCESS_RUN_ID`, `REA_RUN_ID`,
+`REA_SOCKET`, `REA_STATE_FORMAT`, `REA_TARGET_PATH`, `REA_TOKEN`, bridge markers
+such as `REA_COMMAND_COMPLETE_`, and the `ERR_REA_WINDOWS_NATIVE` code.
+`GHIDRA_INSTALL_DIR`, `JAVA_HOME`, `HOPPER_LAUNCHER_PATH`, `HOPPER_TARGET_PATH`,
+`HOPPER_TARGET_KIND`, and `HOPPER_LOADER_ARGS_JSON` are launch configuration
+without the `REA_` prefix. Paths and the provider choice among them stay
+restart-only for the same reason as the table above.
+
+## Hardcoded ceilings left for follow-up
+
+These are checked per call and tell the caller this release has no runtime
+override. They are not `REA_` settings, so the numeric-limit check does not
+list them. Each family has its own owner. Turning one into a setting needs that
+owner's default, floor, maximum, and an error that carries the setting id,
+value, source, and range. Format and protocol ceilings that the
+[limits guide](limits-and-recovery.md) already calls fixed stay fixed. In
+particular, JSON nesting stops at `MAX_JSON_DEPTH` (1000) and the debugger
+control-reply budget stays separate from
+`REA_DEBUGGER_MAX_COMMAND_OUTPUT_BYTES`.
+
+| Family                                           | Examples                                                                                                     | Follow-up                                                                                   |
+| ------------------------------------------------ | ------------------------------------------------------------------------------------------------------------ | ------------------------------------------------------------------------------------------- |
+| Native structural input ceilings                 | PE resources, imports, load config, certificates, exports, initializers, binary hardening, Windows shortcuts | One setting per ceiling, or a shared input budget only where the owners match.              |
+| Native plist                                     | `MAX_NATIVE_PLIST_BYTES`, decode bytes, nesting depth                                                        | Separate from JSON depth.                                                                   |
+| SQLite inspection                                | concurrent inspections, input bytes, diagnostic bytes                                                        | Pool width stays restart-only if it sizes the worker set; byte budgets can become settings. |
+| Producer command allowances                      | fixed command deadline and retained output where the text says there is no runtime override                  | Distinct from per-call `producer_limits`.                                                   |
+| Hopper caller-scan retention and page allowances | `HOPPER_CALLER_SCAN_RETENTION_BYTES`, `HOPPER_CALLER_PAGE_ALLOWANCE_BYTES`                                   | Explicitly not raised by the scan time budget.                                              |
+| Rizin frame protocol                             | `MAX_FRAME_BYTES`, queued frames                                                                             | Protocol bounds, not caller budgets.                                                        |
+| Debugger detach and control replies              | `DETACH_TIMEOUT_MS`, `DEBUGGER_CONTROL_REPLY_BYTES`                                                          | Ownership deadlines, not the shared command deadline.                                       |
+| Web network capture nesting                      | `max_nesting_depth` already has a per-call parameter                                                         | The fixed depth maximum is the follow-up, not the parameter.                                |
+
+## Static check
+
+`npm run lint` runs `scripts/verify/rea-numeric-limits.mjs` after
+dependencies are installed. `scripts/verify/architecture-guards.mjs` runs the
+same check when `typescript6` is already installed; the pre-install CI step
+cannot. The check scans `src` excluding tests,
+keeps names that are numeric limits, and requires each name to be an
+`environment` on an analysis setting or a registry entry with a reason.
+`policy: "per-call"` is temporary. The migration that moves the limit deletes
+the entry and rejects any new `per-call` policy so the durable exceptions are
+only `restart-only`.

docs/specs/README.md

@@ -26,6 +26,7 @@ for new work, roadmap granularity and producer-owned contracts.
 - [Native call source coordinates](native/call-source-locations/spec.md) ([plan](native/call-source-locations/plan.md), [tasks](native/call-source-locations/tasks.md), [verification](native/call-source-locations/verification.md))
 - [Go static string-header literal recovery](native/go-string-literals/spec.md) ([plan](native/go-string-literals/plan.md), [tasks](native/go-string-literals/tasks.md))
 - [Windows Shell Link shortcut inspection](native/windows-shortcut/spec.md) ([plan](native/windows-shortcut/plan.md), [tasks](native/windows-shortcut/tasks.md))
+- [NTFS $MFT and USN Journal inspection](native/ntfs-mft/spec.md) ([plan](native/ntfs-mft/plan.md), [tasks](native/ntfs-mft/tasks.md))
 - [Solidity terminal metadata inspection](solidity/cbor-metadata/spec.md) ([plan](solidity/cbor-metadata/plan.md), [tasks](solidity/cbor-metadata/tasks.md), [verification](solidity/cbor-metadata/verification.md))
 - [Inspect a V8 code cache container](v8/code-cache-inspection/spec.md) ([plan](v8/code-cache-inspection/plan.md), [tasks](v8/code-cache-inspection/tasks.md))
 - [Deno compile executable inventory and extraction](artifacts/deno-standalone/spec.md) ([plan](artifacts/deno-standalone/plan.md), [tasks](artifacts/deno-standalone/tasks.md))

docs/specs/native/ntfs-mft/plan.md

@@ -0,0 +1,39 @@
+# Implementation plan
+
+## Boundary and contract
+
+Follow the existing artifact readers. `src/domain/windows/ntfsMft.ts` owns FILE
+record bytes and coverage; `src/domain/artifactFormatMetadata.ts` owns the
+`ntfs-mft` result schema. `NtfsMftArtifactReader` reads the admitted file
+through `ArtifactFileSource` and exposes its facts through `formatMetadata()`.
+A record stream is its own content, so the root occurrence carries every fact
+and no members are enumerated.
+
+Recognition lives in `hasMftFileRecordSignature` and is applied in
+`classifyArtifactBytes`, next to the other unambiguous magic probes.
+
+## Incremental work and validation
+
+1. Define the first-step contract, limits and observations; add parser fixtures.
+2. Implement the FILE record header and update-sequence fixup parse with
+   explicit limitations.
+3. Add the reader, format descriptor and byte classification.
+4. Decode attribute records, then `$FILE_NAME` attributes so a record's parent
+   reference, names and timestamps become explicit; keep hard links separate.
+5. Parse `$UsnJrnl:$J` records and correlate them to records by full file
+   reference, including sequence numbers, reporting unresolved references.
+6. Document the workflow in a native Windows artifact guide.
+
+## Effects and compatibility
+
+The workflow reads one caller-selected local file. It adds no target execution,
+injection, mutation, protection recovery, network access, volume mount or live
+filesystem access, so no maintainer scope decision is required.
+`inspect_artifact` routing for every existing format is unchanged, and the new
+format id appears only for bytes whose first record carries the `FILE`
+signature.
+
+## References
+
+- https://learn.microsoft.com/en-us/windows/win32/fileio/master-file-table
+- https://github.com/libyal/libfsntfs/blob/main/documentation/New%20Technologies%20File%20System%20(NTFS).asciidoc

docs/specs/native/ntfs-mft/spec.md

@@ -0,0 +1,51 @@
+# NTFS $MFT and USN Journal inspection
+
+This specification records an implementation increment and its original problem
+statement. For current behavior and limitations, see the artifact inventory
+guide.
+
+## Problem and outcome
+
+REA has no reader for acquired NTFS metadata artifacts. An analyst holding a
+captured `$MFT` stream cannot ask which FILE records it contains, whether their
+update-sequence fixups survived acquisition, or which records are directories,
+extensions of another record, or no longer in use. Add offline, read-only
+inventory of a selected `$MFT` file, delivered in steps so each one merges on
+its own.
+
+## Requirements
+
+- Recognize a `$MFT` stream by a plausible first `FILE` record header under
+  any file name. CLI and MCP target admission route it to the artifact reader.
+- Validate each FILE record header: the `FILE` signature, sequence number, flags,
+  used and total byte counts, base record reference, and the update-sequence
+  array's offset, count and array bounds.
+- Verify the update-sequence fixups: every sector trailer the array covers must
+  hold the array's sequence number. A mismatch keeps the record's header facts
+  and is reported per record and in the metadata count.
+- Report a missing, misplaced or short update-sequence array as unverified
+  rather than as a clean record, and a trailing partial record as a byte count.
+- Report the first invalid record signature and the number of invalid slots.
+  Continue at the next slot when the stream declared a supported record size.
+  With a fallback record size, stop at the first invalid signature.
+- Bound the whole file at a fixed 512 MiB ceiling and report the ceiling as a
+  `limit` failure with its boundary.
+- Return the first 32 decoded record headers with their source offsets and the
+  number of omitted headers. Aggregate counts cover the full parsed stream.
+- Parse only the records' headers and fixups. Attributes, filenames, data
+  streams and the `$UsnJrnl:$J` journal are not decoded in this step, so no path,
+  file size, file reference correlation or change history is claimed.
+
+## Acceptance
+
+Synthesized fixtures cover a clean record, a failed fixup, a missing update
+sequence array, an array shorter than the record's sector count, extension and
+directory records, a record whose signature is wrong mid-stream, a trailing
+partial record, an undeclared record size, a 4 KiB record, a stream shorter than
+one header, bytes without the `FILE` signature, and the file-size ceiling.
+Inventory tests cover the root occurrence's metadata and the coverage status
+when a fixup fails, and classification tests cover several file names for the
+same bytes.
+
+Not covered in this step: real `$MFT` files written by a Windows host, USN
+journal records, attribute decoding, and any correlation between the two.

docs/specs/native/ntfs-mft/tasks.md

@@ -0,0 +1,11 @@
+# Tasks
+
+- [x] Define the first-step scope, contract and limits.
+- [x] Parse FILE record headers and update-sequence fixups with synthesized
+      fixtures and explicit limitations.
+- [x] Add the reader, format descriptor and byte classification.
+- [ ] Decode attribute records, then `$FILE_NAME` attributes with parent
+      references, names, hard links and timestamps.
+- [ ] Parse `$UsnJrnl:$J` records and correlate them to records by full file
+      reference.
+- [ ] Verify against real `$MFT` and journal captures from a Windows host.

package.json

@@ -148,7 +148,7 @@
     "build:termux:run": "npm run compile:termux && npm run artifacts:skills",
     "format": "oxfmt .",
     "format:check": "oxfmt --check .",
-    "lint": "node scripts/ci/static-check-report.mjs lint && npm run verify:module-boundaries",
+    "lint": "node scripts/ci/static-check-report.mjs lint && npm run verify:module-boundaries && node scripts/verify/rea-numeric-limits.mjs",
     "lint:details": "oxlint && npm run verify:module-boundaries",
     "lint:fix": "oxlint --fix && npm run verify:module-boundaries",
     "knip": "knip --cache",

scripts/ci/scopes.mjs

@@ -636,7 +636,7 @@ const PATH_SCOPES = {
     "src/ghidra/GhidraJavaProperties.ts",
     "src/ghidra/GhidraReleasePolicy.ts",
     "scripts/verify-ghidra-bridge-sources.mjs",
-    "tests/conformance/ghidra/ReaArmContextProbe.java",
+    "tests/conformance/ghidra/ReaInstructionContextProbe.java",
     "tests/conformance/ghidra/ReaComputedCallCoverageProbe.java",
     "tests/conformance/ghidra/ReaDecodeProbe.java",
     ".github/workflows/ghidra-bridge-sources.yml",

scripts/verify-qlog-captures.mjs

@@ -50,7 +50,7 @@ transport.stderr?.on("data", (bytes) => {
 const ajv = new Ajv2020({ strict: false, validateFormats: false });
 const measurements = [];
 const contract = toolContract("inspect_web_network_capture");
-await withVerifierCleanup(async () => {
+const report = await withVerifierCleanup(async () => {
   try {
     await client.connect(transport);
   } catch (cause) {
@@ -196,7 +196,7 @@ await withVerifierCleanup(async () => {
     });
   }
   await client.ping();
-  const report = {
+  return {
     status: "passed",
     profile: "aioquic-1.5.0-json-0.3",
     measurements,
@@ -206,17 +206,6 @@ await withVerifierCleanup(async () => {
       "No PCAP or keylog is read by this verifier; repeated STREAM coverage is not proof of retransmission or network loss.",
     ],
   };
-  await writeFile(reportPath, JSON.stringify(report, null, 2) + "\n", {
-    flag: "wx",
-    mode: 0o600,
-  });
-  console.log(
-    JSON.stringify({
-      status: report.status,
-      cases: measurements.length,
-      report: reportPath,
-    }),
-  );
 }, [
   async () => {
     await closeVerifierResources([
@@ -225,3 +214,15 @@ await withVerifierCleanup(async () => {
     ]);
   },
 ]);
+
+await writeFile(reportPath, JSON.stringify(report, null, 2) + "\n", {
+  flag: "wx",
+  mode: 0o600,
+});
+console.log(
+  JSON.stringify({
+    status: report.status,
+    cases: measurements.length,
+    report: reportPath,
+  }),
+);

scripts/verify/architecture-guards.mjs

@@ -1,3 +1,5 @@
+import { spawnSync } from "node:child_process";
+import { existsSync } from "node:fs";
 import { readFile, readdir } from "node:fs/promises";
 import { join, relative, sep } from "node:path";
 import { fileURLToPath } from "node:url";
@@ -193,4 +195,23 @@ if (failures.length > 0) {
   console.error(`Architecture guards failed:\n\n${failures.join("\n\n")}\n`);
   process.exit(1);
 }
+// CI runs this file before `npm ci`. The numeric-limit parser needs
+// typescript6, so that check waits until dependencies exist. `npm run lint`
+// runs it after install.
+const typescript6 = new URL(
+  "../../node_modules/typescript6/package.json",
+  import.meta.url,
+);
+if (!existsSync(typescript6)) {
+  console.log(
+    "Numeric REA_ limit check waits until dependencies are installed.",
+  );
+} else {
+  const numericLimits = spawnSync(
+    process.execPath,
+    [fileURLToPath(new URL("./rea-numeric-limits.mjs", import.meta.url))],
+    { stdio: "inherit" },
+  );
+  if (numericLimits.status !== 0) process.exit(numericLimits.status ?? 1);
+}
 console.log(`Architecture guards passed across ${files.length} source files.`);

scripts/verify/rea-numeric-limits.d.mts

@@ -0,0 +1,23 @@
+/** Relative path of the registry. Its text is a declaration, not a producer. */
+export const NUMERIC_LIMIT_DECLARATION: "src/domain/numericLimitRegistry.ts";
+
+/** Limits that must become analysis settings and cannot be registry exceptions. */
+export const REQUIRED_RUNTIME_LIMITS: readonly [
+  "REA_NTFS_MFT_MAX_FILE_BYTES",
+  "REA_REGF_HIVE_MAX_FILE_BYTES",
+  "REA_WEB_BUNDLE_ENDPOINT_LITERAL_CANDIDATES",
+];
+
+/** True for a numeric `REA_` limit, including the documented endpoint-candidate count. */
+export function isNumericLimit(name: string): boolean;
+
+/** Producer paths for numeric limits. Registry text, comments, and allowlists are omitted. */
+export function discoverNumericLimitUses(
+  files: readonly { readonly path: string; readonly source: string }[],
+): Map<string, string>;
+
+/** Why a tracked limit is not an analysis setting, when it is present. */
+export function requiredRuntimeLimitFailure(
+  name: string,
+  presence: { readonly inRuntime: boolean; readonly inRegistry: boolean },
+): string | undefined;

scripts/verify/rea-numeric-limits.mjs

@@ -0,0 +1,399 @@
+import { readFile, readdir } from "node:fs/promises";
+import { join, relative, resolve, sep } from "node:path";
+import { fileURLToPath, pathToFileURL } from "node:url";
+import ts from "typescript6";
+
+// Every numeric REA_ limit under src must be an analysis setting or a
+// restart-only exception with a reason and an update_mode. A per-call limit
+// belongs in ANALYSIS_SETTINGS. The three limits moving on #9868, #9865, and
+// #9836 are not in this tree; once their names appear here they must join
+// ANALYSIS_SETTINGS. REA_WEB_BUNDLE_ENDPOINT_LITERAL_CANDIDATES is numeric
+// even though its name has no byte or time suffix.
+
+const root = fileURLToPath(new URL("../../", import.meta.url));
+const definitionsPath = join(root, "src/domain/analysisSettingDefinitions.ts");
+const registryPath = join(root, "src/domain/numericLimitRegistry.ts");
+
+const listFiles = async (dir) => {
+  const output = [];
+  for (const entry of await readdir(dir, { withFileTypes: true })) {
+    const path = join(dir, entry.name);
+    if (entry.isDirectory()) {
+      if (entry.name === "node_modules" || entry.name === "dist") continue;
+      output.push(...(await listFiles(path)));
+    } else if (
+      entry.name.endsWith(".ts") &&
+      !entry.name.endsWith(".test.ts") &&
+      !entry.name.endsWith(".fixture.ts")
+    )
+      output.push(path);
+  }
+  return output;
+};
+
+/**
+ * Count-shaped limits whose names do not use a byte, time, or maximum suffix.
+ * `REA_WEB_BUNDLE_ENDPOINT_LITERAL_CANDIDATES` is the runtime setting on #9836.
+ */
+const DOCUMENTED_COUNT_LIMITS = new Set([
+  "REA_WEB_BUNDLE_ENDPOINT_LITERAL_CANDIDATES",
+]);
+
+/** Registry text is a declaration, not a producer that uses the limit. */
+export const NUMERIC_LIMIT_DECLARATION = "src/domain/numericLimitRegistry.ts";
+
+/**
+ * These limits are moving onto runtime analysis settings on their own PRs.
+ * A producer or a registry entry cannot satisfy that requirement.
+ */
+export const REQUIRED_RUNTIME_LIMITS = [
+  "REA_NTFS_MFT_MAX_FILE_BYTES",
+  "REA_REGF_HIVE_MAX_FILE_BYTES",
+  "REA_WEB_BUNDLE_ENDPOINT_LITERAL_CANDIDATES",
+];
+
+const unwrapExpression = (node) => {
+  let current = node;
+  while (
+    ts.isAsExpression(current) ||
+    ts.isSatisfiesExpression(current) ||
+    ts.isParenthesizedExpression(current) ||
+    ts.isTypeAssertionExpression(current) ||
+    ts.isNonNullExpression(current)
+  )
+    current = current.expression;
+  return current;
+};
+
+const literalText = (node) =>
+  ts.isStringLiteral(node) || ts.isNoSubstitutionTemplateLiteral(node)
+    ? node.text
+    : undefined;
+
+/** Same-file const strings and `setting` fields, so `environment[name]` still resolves. */
+const stringBindings = (sourceFile) => {
+  const bindings = new Map();
+  const visit = (node) => {
+    if (
+      ts.isVariableDeclaration(node) &&
+      ts.isIdentifier(node.name) &&
+      node.initializer !== undefined
+    ) {
+      const initializer = unwrapExpression(node.initializer);
+      const literal = literalText(initializer);
+      if (literal !== undefined) bindings.set(node.name.text, literal);
+      if (ts.isObjectLiteralExpression(initializer)) {
+        for (const property of initializer.properties) {
+          if (
+            !ts.isPropertyAssignment(property) ||
+            (!ts.isIdentifier(property.name) &&
+              !ts.isStringLiteral(property.name))
+          )
+            continue;
+          const value = literalText(unwrapExpression(property.initializer));
+          if (value !== undefined)
+            bindings.set(`${node.name.text}.${property.name.text}`, value);
+        }
+      }
+    }
+    ts.forEachChild(node, visit);
+  };
+  visit(sourceFile);
+  return bindings;
+};
+
+const boundLimitName = (node, bindings) => {
+  const expression = unwrapExpression(node);
+  const literal = literalText(expression);
+  if (literal !== undefined) return literal;
+  if (ts.isIdentifier(expression)) return bindings.get(expression.text);
+  if (
+    ts.isPropertyAccessExpression(expression) &&
+    ts.isIdentifier(expression.expression) &&
+    (ts.isIdentifier(expression.name) || ts.isStringLiteral(expression.name))
+  )
+    return bindings.get(
+      `${expression.expression.text}.${expression.name.text}`,
+    );
+  return undefined;
+};
+
+export const isNumericLimit = (name) =>
+  DOCUMENTED_COUNT_LIMITS.has(name) ||
+  /(?:_BYTES|_MS|_MIB|_SECONDS|_POLLS|_JOBS|_ELEMENTS|_COUNT|_SESSIONS|_COMMANDS)$/u.test(
+    name,
+  ) ||
+  /_MAX_/u.test(name) ||
+  /_(?:TIMEOUT|DEADLINE|BUDGET)(?:_|$)/u.test(name) ||
+  /_HEAP(?:_|$)/u.test(name);
+
+const limitName = (text) =>
+  text !== undefined && /^REA_[A-Z0-9_]+$/u.test(text) && isNumericLimit(text)
+    ? text
+    : undefined;
+
+const remember = (discovered, path, name) => {
+  const limit = limitName(name);
+  if (limit === undefined || discovered.has(limit)) return;
+  discovered.set(limit, path);
+};
+
+/**
+ * Producers are syntax, not text. A read is a property or element access,
+ * including a const string used as the element, a string argument to a call,
+ * an object-binding key (rename and default included), an `in` check, or an
+ * object key that a spread can copy. Comments and array literals are not reads.
+ */
+export const discoverNumericLimitUses = (files) => {
+  const discovered = new Map();
+  for (const file of files) {
+    const relativePath = file.path.split(sep).join("/");
+    if (relativePath === NUMERIC_LIMIT_DECLARATION) continue;
+    const sourceFile = ts.createSourceFile(
+      relativePath,
+      file.source,
+      ts.ScriptTarget.Latest,
+      true,
+      ts.ScriptKind.TS,
+    );
+    const bindings = stringBindings(sourceFile);
+    const visit = (node) => {
+      if (
+        ts.isPropertyAccessExpression(node) ||
+        ts.isPropertyAccessChain?.(node)
+      )
+        remember(discovered, relativePath, node.name.text);
+      if (ts.isElementAccessExpression(node))
+        remember(
+          discovered,
+          relativePath,
+          boundLimitName(node.argumentExpression, bindings),
+        );
+      if (ts.isCallExpression(node) || ts.isNewExpression(node)) {
+        for (const argument of node.arguments ?? [])
+          remember(
+            discovered,
+            relativePath,
+            boundLimitName(argument, bindings),
+          );
+      }
+      if (
+        ts.isBinaryExpression(node) &&
+        node.operatorToken.kind === ts.SyntaxKind.InKeyword
+      )
+        remember(discovered, relativePath, boundLimitName(node.left, bindings));
+      if (
+        ts.isBindingElement(node) &&
+        node.dotDotDotToken === undefined &&
+        ts.isObjectBindingPattern(node.parent)
+      ) {
+        const key = node.propertyName ?? node.name;
+        if (
+          ts.isIdentifier(key) ||
+          ts.isStringLiteral(key) ||
+          ts.isNumericLiteral(key)
+        )
+          remember(discovered, relativePath, key.text);
+      }
+      if (
+        (ts.isPropertyAssignment(node) ||
+          ts.isShorthandPropertyAssignment(node)) &&
+        (ts.isIdentifier(node.name) ||
+          ts.isStringLiteral(node.name) ||
+          ts.isNumericLiteral(node.name))
+      )
+        remember(discovered, relativePath, node.name.text);
+      if (ts.isVariableDeclaration(node) && node.initializer !== undefined)
+        remember(
+          discovered,
+          relativePath,
+          literalText(unwrapExpression(node.initializer)),
+        );
+      if (ts.isPropertyAssignment(node))
+        remember(
+          discovered,
+          relativePath,
+          literalText(unwrapExpression(node.initializer)),
+        );
+      if (
+        ts.isTemplateExpression(node) ||
+        ts.isNoSubstitutionTemplateLiteral(node)
+      ) {
+        const chunks = ts.isNoSubstitutionTemplateLiteral(node)
+          ? [node.text]
+          : [
+              node.head.text,
+              ...node.templateSpans.map((span) => span.literal.text),
+            ];
+        for (const chunk of chunks)
+          for (const token of chunk.match(/REA_[A-Z0-9_]+/gu) ?? [])
+            remember(discovered, relativePath, token);
+      }
+      ts.forEachChild(node, visit);
+    };
+    visit(sourceFile);
+  }
+  return discovered;
+};
+
+/** Tracked limits join ANALYSIS_SETTINGS. The exception registry cannot classify them. */
+export const requiredRuntimeLimitFailure = (
+  name,
+  { inRuntime, inRegistry },
+) => {
+  if (!REQUIRED_RUNTIME_LIMITS.includes(name)) return undefined;
+  if (inRuntime && !inRegistry) return undefined;
+  return `${name} must join ANALYSIS_SETTINGS; numericLimitRegistry cannot classify it`;
+};
+
+const namesIn = (source, pattern) =>
+  new Set([...source.matchAll(pattern)].map((match) => match[1]));
+
+const direct =
+  process.argv[1] !== undefined &&
+  import.meta.url === pathToFileURL(resolve(process.argv[1])).href;
+
+if (direct) {
+  const failures = [];
+  const definitions = await readFile(definitionsPath, "utf8");
+  const registry = await readFile(registryPath, "utf8");
+  const runtime = namesIn(definitions, /environment:\s*"(REA_[A-Z0-9_]+)"/gu);
+  const declaredNames = [
+    ...registry.matchAll(/name:\s*"(REA_[A-Z0-9_]+)"/gu),
+  ].map((match) => match[1]);
+  const entries = [
+    ...registry.matchAll(
+      /name:\s*"(REA_[A-Z0-9_]+)"\s*,\s*update_mode:\s*"(runtime|restart)"\s*,\s*policy:\s*"(restart-only|per-call)"\s*,\s*read:\s*"(startup|per-call)"\s*,\s*reason:\s*"((?:\\.|[^"\\])*)"/gu,
+    ),
+  ];
+  const listed = new Map();
+  for (const match of entries) {
+    const [, name, updateMode, policy, read, reason] = match;
+    if (listed.has(name)) failures.push(`${name} is listed twice`);
+    listed.set(name, { updateMode, policy, read, reason });
+    if (reason.trim().length < 40)
+      failures.push(`${name} needs a reason that states why`);
+    if (updateMode !== "restart")
+      failures.push(
+        `${name} is outside ANALYSIS_SETTINGS, so its update_mode must be restart`,
+      );
+    if (runtime.has(name))
+      failures.push(
+        `${name} is both a runtime setting and on the exception list`,
+      );
+    if (policy === "per-call" && read !== "per-call")
+      failures.push(`${name} is per-call policy but is not read per call`);
+    if (policy === "restart-only" && read !== "startup")
+      failures.push(`${name} is restart-only but is not read at startup`);
+  }
+  for (const name of declaredNames) {
+    if (!listed.has(name))
+      failures.push(`${name} is registered without update_mode`);
+  }
+  const settingBodies = [
+    ...definitions.matchAll(/\n {2}"[^"]+": \{([\s\S]*?)\n {2}\},/gu),
+  ];
+  if (settingBodies.length === 0)
+    failures.push("ANALYSIS_SETTINGS did not parse into setting bodies");
+  for (const [, body] of settingBodies) {
+    if (!/updateMode:\s*"(?:runtime|restart)"/u.test(body))
+      failures.push("an analysis setting is missing updateMode");
+  }
+
+  const discovered = discoverNumericLimitUses(
+    await Promise.all(
+      (await listFiles(join(root, "src"))).map(async (path) => ({
+        path: relative(root, path),
+        source: await readFile(path, "utf8"),
+      })),
+    ),
+  );
+
+  for (const [name, file] of [...discovered].sort(([left], [right]) =>
+    left < right ? -1 : left > right ? 1 : 0,
+  )) {
+    if (!runtime.has(name) && !listed.has(name))
+      failures.push(
+        `${name} in ${file} is neither an analysis setting nor docs/design/rea-limit-inventory.md via src/domain/numericLimitRegistry.ts`,
+      );
+  }
+  for (const name of listed.keys()) {
+    if (!discovered.has(name))
+      failures.push(`${name} is registered but no longer appears in src`);
+  }
+  const present = new Set([...discovered.keys(), ...listed.keys(), ...runtime]);
+  for (const name of REQUIRED_RUNTIME_LIMITS) {
+    if (!present.has(name)) continue;
+    const failure = requiredRuntimeLimitFailure(name, {
+      inRuntime: runtime.has(name),
+      inRegistry: listed.has(name),
+    });
+    if (failure !== undefined) failures.push(failure);
+  }
+
+  const parameterOwners = [
+    "src/artifacts/ArtifactReaderLimits.ts",
+    "src/domain/artifactReaderLimits.ts",
+    "src/domain/webNetworkCapture.ts",
+  ];
+  for (const relativePath of parameterOwners) {
+    const source = await readFile(join(root, relativePath), "utf8");
+    if (/update_mode:\s*"runtime"/u.test(source))
+      failures.push(
+        `${relativePath} is a per-call parameter, so update_mode cannot be runtime`,
+      );
+    if (!/update_mode:\s*"parameter"/u.test(source))
+      failures.push(`${relativePath} must report update_mode parameter`);
+  }
+
+  const forbidden =
+    /(?<![A-Za-z0-9_])(?:rea_restart_required|restartRequired|restart_required)(?![A-Za-z0-9_])|effect:\s*"(?:immediate|next_process)"|SettingChangeEffect/gu;
+  // Docs and verification receipts keep the field names they recorded.
+  // Source, tests, skills, and scripts cannot reintroduce those names.
+  const scanRoots = ["src", "tests", ".agents", "scripts"].map((dir) =>
+    join(root, dir),
+  );
+  const skipDir = new Set([
+    "node_modules",
+    "dist",
+    ".turbo",
+    ".cache",
+    "rea-numeric-limits.mjs",
+  ]);
+  const listText = async (dir) => {
+    const output = [];
+    let entries;
+    try {
+      entries = await readdir(dir, { withFileTypes: true });
+    } catch {
+      return output;
+    }
+    for (const entry of entries) {
+      if (skipDir.has(entry.name)) continue;
+      const path = join(dir, entry.name);
+      if (entry.isDirectory()) output.push(...(await listText(path)));
+      else if (/\.(?:ts|tsx|mjs|js|md|json|py|yml|yaml)$/u.test(entry.name))
+        output.push(path);
+    }
+    return output;
+  };
+  for (const dir of scanRoots) {
+    for (const path of await listText(dir)) {
+      if (path.endsWith(`${sep}rea-numeric-limits.mjs`)) continue;
+      const source = await readFile(path, "utf8");
+      const hit = source.match(forbidden);
+      if (hit)
+        failures.push(
+          `${relative(root, path).split(sep).join("/")} still uses ${[...new Set(hit)].join(", ")}`,
+        );
+    }
+  }
+
+  if (failures.length > 0) {
+    console.error(`Numeric REA_ limit check failed:\n${failures.join("\n")}`);
+    process.exit(1);
+  }
+  console.log(
+    `Numeric REA_ limits: ${runtime.size} runtime, ${[...listed.values()].filter((entry) => entry.policy === "restart-only").length} restart-only, ${[...listed.values()].filter((entry) => entry.policy === "per-call").length} per-call pending migration, ${discovered.size} discovered.`,
+  );
+}

src/android/JadxOperationTimeout.test.ts

@@ -36,7 +36,8 @@ it("names REA_JADX_OPERATION_TIMEOUT_MS when the operation deadline elapses", ()
       applies_at: "next_operation",
       agent_editable: true,
       scope: "rea-runtime",
-      rea_restart_required: false,
+      update_mode: "runtime",
+      pending_restart: false,
     },
   });
   expect(projected).toMatchObject({

src/application/AnalysisSettings.ts

@@ -134,7 +134,7 @@ export class AnalysisSettings {
         };
       }),
       active_work_changed: false as const,
-      rea_restart_required: false as const,
+      pending_restart: false as const,
     });
   }
   #describe(setting: AnalysisSettingId) {
@@ -156,6 +156,8 @@ export class AnalysisSettings {
       minimum: definition.minimum,
       maximum: definition.maximum,
       applies_at: definition.appliesAt,
+      update_mode: definition.updateMode,
+      pending_restart: false as const,
       provider: definition.provider,
       operations: [...definition.operations],
       ...(definition.environment === undefined

src/application/BinaryTargetResolver.ts

@@ -97,6 +97,7 @@ import {
   isDfuSeFile,
 } from "../domain/firmware/dfuse.js";
 import { hasFitImagesNode } from "../domain/firmware/fitImage.js";
+import { hasMftFileRecordHeader } from "../domain/windows/ntfsMft.js";
 import {
   locateEspFlashLayout,
   type EspFlashSelection,
@@ -816,6 +817,7 @@ const detectArtifactFormat = async (
     });
   if (esp === "esp-image") return "esp-image";
   if (hasFitImagesNode(magic)) return "fit";
+  if (hasMftFileRecordHeader(magic)) return "ntfs-mft";
   // Binary plists, including keyed archives, often use other extensions.
   if (hasBinaryPlistSignature(magic)) return "plist";
   const named = namedArtifactFormat(lower);
@@ -899,6 +901,7 @@ const isArchiveFormat = (
     "dfuse",
     "esp-flash-image",
     "fit",
+    "ntfs-mft",
   ].includes(format);
 
 /** Whether the leading bytes begin any executable header family, valid or not. */

src/application/binary/TargetFormatVocabularies.test.ts

@@ -135,6 +135,7 @@ const TARGETS: TargetFixtures = {
     espFlash: { selected: true, flashOffset: null, partitionTableOffset: null },
   },
   fit: { ...archive, format: "fit" },
+  "ntfs-mft": { ...archive, format: "ntfs-mft" },
   plist: { ...artifact, format: "plist" },
   javascript: { ...artifact, format: "javascript" },
   "source-map": { ...artifact, format: "source-map" },

src/application/javascript/JavaScriptArtifactReconstruction.ts

@@ -1,6 +1,7 @@
 import { discoverJavaScriptPreloadPaths } from "./JavaScriptPreloadDiscovery.js";
 import { artifactNameForClassification } from "../../domain/artifactNameSelection.js";
 import type { AnalysisSettingValues } from "../../domain/analysisSettings.js";
+import { selectedAnalysisSetting } from "../../domain/analysisSettingDefinitions.js";
 import { absoluteCallerPath } from "../../filesystem/callerPath.js";
 import {
   DEFAULT_ASAR_HEADER_ADMISSION,
@@ -181,18 +182,24 @@ export const reconstructJavaScriptArtifact = async (
           ? {}
           : { fileNodes: input.max_semantic_nodes_per_file }),
       });
+      const heap = selectedAnalysisSetting(
+        input.max_heap_mb,
+        overrides?.["javascript.max_heap_mb"],
+        JAVASCRIPT_ANALYSIS_RESOURCE_DEFAULTS.heapMb,
+      );
+      const timeout = selectedAnalysisSetting(
+        input.analysis_timeout_ms,
+        overrides?.["javascript.analysis_timeout_ms"],
+        JAVASCRIPT_ANALYSIS_RESOURCE_DEFAULTS.timeoutMs,
+      );
       const sourceLimits = {
         ...(input.semantic_precision === undefined
           ? {}
           : { semanticPrecision: input.semantic_precision }),
-        heapMb:
-          input.max_heap_mb ??
-          overrides?.["javascript.max_heap_mb"] ??
-          JAVASCRIPT_ANALYSIS_RESOURCE_DEFAULTS.heapMb,
-        timeoutMs:
-          input.analysis_timeout_ms ??
-          overrides?.["javascript.analysis_timeout_ms"] ??
-          JAVASCRIPT_ANALYSIS_RESOURCE_DEFAULTS.timeoutMs,
+        heapMb: heap.value,
+        heapSource: heap.source,
+        timeoutMs: timeout.value,
+        timeoutSource: timeout.source,
       };
       const preloads = await discoverJavaScriptPreloadPaths(
         files,

src/application/wasm/WasmArtifactPort.ts

@@ -15,6 +15,8 @@ export type WasmArtifactExecution = Omit<AnalysisExecution, "subject"> & {
 /** Effective operation budget, separate from caller-authored Evidence parameters. */
 export interface WasmArtifactExecutionOptions extends ExecutionOptions {
   readonly commandTimeoutMs?: number;
+  /** Selection owner for `commandTimeoutMs`. Absent when the caller did not record it. */
+  readonly commandTimeoutSource?: "default" | "runtime" | "request" | undefined;
 }
 
 /** Replaceable offline WASM inspection engine, independent of target execution and network clients. */

src/application/wasm/WasmArtifactService.test.ts

@@ -240,6 +240,7 @@ it("gives cancellation precedence over malformed raw input", async () => {
 it("uses runtime defaults only for omitted budgets, preserving explicit input and reset", async () => {
   const settings = new AnalysisSettings({});
   const accepted: number[] = [];
+  const sources: (string | undefined)[] = [];
   const authored: (number | undefined)[] = [];
   const service = new WasmArtifactService(
     {
@@ -248,6 +249,7 @@ it("uses runtime defaults only for omitted budgets, preserving explicit input an
         const value = wasmArtifactFixture();
         value.tool_profile.timeout_ms = options!.commandTimeoutMs!;
         accepted.push(options!.commandTimeoutMs!);
+        sources.push(options!.commandTimeoutSource);
         authored.push(input.command_timeout_ms);
         return ok(wasmArtifactExecution(value));
       },
@@ -275,6 +277,7 @@ it("uses runtime defaults only for omitted budgets, preserving explicit input an
   });
   expect((await settings.run(() => service.inspect({ path }))).ok).toBe(true);
   expect(accepted).toEqual([123000, 42000, 30000]);
+  expect(sources).toEqual(["runtime", "request", "default"]);
   expect(authored).toEqual([undefined, 42000, undefined]);
 });
 

src/application/wasm/WasmArtifactService.ts

@@ -1,4 +1,5 @@
 import type { AnalysisSettingsReader } from "../../domain/analysisSettings.js";
+import { selectedAnalysisSetting } from "../../domain/analysisSettingDefinitions.js";
 import { constants as bufferConstants } from "node:buffer";
 import type { ExecutionOptions } from "../AnalysisProvider.js";
 import {
@@ -78,13 +79,15 @@ export class WasmArtifactService {
           ),
         );
     }
-    const timeout =
-      input.value.command_timeout_ms ??
-      this.settings?.().overrides["wasm.command_timeout_ms"] ??
-      DEFAULT_WASM_COMMAND_TIMEOUT_MS;
+    const timeout = selectedAnalysisSetting(
+      input.value.command_timeout_ms,
+      this.settings?.().overrides["wasm.command_timeout_ms"],
+      DEFAULT_WASM_COMMAND_TIMEOUT_MS,
+    );
     const inspected = await this.provider.inspect(input.value, {
       ...options,
-      commandTimeoutMs: timeout,
+      commandTimeoutMs: timeout.value,
+      commandTimeoutSource: timeout.source,
     });
     if (!inspected.ok) return inspected;
     const execution = inspected.value;
@@ -108,7 +111,7 @@ export class WasmArtifactService {
       !report.success ||
       report.data.artifact.path !== input.value.path ||
       execution.subject.path !== input.value.path ||
-      report.data.tool_profile.timeout_ms !== timeout ||
+      report.data.tool_profile.timeout_ms !== timeout.value ||
       report.data.tool_profile.output_budget_bytes !==
         wasmResourceLimits(input.value).outputBytes ||
       !wasmArtifactMatchesSubject(report.data.artifact, execution.subject) ||

src/artifacts/ArtifactOperationFailure.test.ts

@@ -54,7 +54,8 @@ it("preserves adjustable reader admission facts and gives cleanup priority", ()
         required_value: 2,
         unit: "count",
         applies_at: "next_operation",
-        rea_restart_required: false,
+        update_mode: "parameter",
+        pending_restart: false,
       },
     },
   });

src/artifacts/ArtifactReader.ts

@@ -215,6 +215,7 @@ export interface ArtifactReader {
     | "dfuse"
     | "esp-flash-image"
     | "fit"
+    | "ntfs-mft"
     | "file";
   entries(
     signal?: AbortSignal,

src/artifacts/ArtifactReaderLimits.ts

@@ -40,7 +40,8 @@ export const artifactReaderLimitFailure = (
       applies_at: "next_operation",
       agent_editable: true,
       can_increase: canIncrease,
-      rea_restart_required: false,
+      update_mode: "parameter" as const,
+      pending_restart: false as const,
     },
   });
 };

src/artifacts/DecodedRegionArtifactReader.ts

@@ -18,7 +18,10 @@ import {
   type ArtifactTraversalOptions,
 } from "./ArtifactReader.js";
 import { ArtifactResourceScope } from "./ArtifactResourceScope.js";
-import { assertArtifactReadActive } from "./ArtifactFileOperation.js";
+import {
+  artifactReadCancelledBy,
+  assertArtifactReadActive,
+} from "./ArtifactFileOperation.js";
 import { ArtifactFileSource } from "./ArtifactFileSource.js";
 import { readFileHandleChunks } from "../filesystem/readFileHandleChunks.js";
 import type { StableRegularFileDescriptor } from "../filesystem/RegularFile.js";
@@ -64,12 +67,14 @@ export const regionFileName = (start: number, end: number): string =>
  */
 export class DecodedRegionArtifactReader<Image> implements ArtifactReader {
   readonly #scope = new ArtifactResourceScope();
-  readonly #entries = new Map<string, ArtifactEntry>();
-  readonly #regions = new Map<string, Uint8Array>();
+  #entries = new Map<string, ArtifactEntry>();
+  #regions = new Map<string, Uint8Array>();
   readonly #streams = new Set<Readable>();
   readonly #source: ArtifactFileSource;
   readonly #decoding: RegionDecoding<Image>;
-  #prepared: Promise<void> | undefined;
+  #prepared:
+    | { signal: AbortSignal | undefined; promise: Promise<void> }
+    | undefined;
   /** Whole-file facts only; the region map is the sole owner of region bytes. */
   #described: Omit<DecodedRegions, "regions"> | undefined;
   #closing = false;
@@ -110,8 +115,27 @@ export class DecodedRegionArtifactReader<Image> implements ArtifactReader {
   ): AsyncIterable<ArtifactEntry> {
     await this.#scope.run(async () => {
       this.#assertOpen(signal);
-      this.#prepared ??= this.#prepare(signal);
-      await this.#prepared;
+      for (;;) {
+        const pending = (this.#prepared ??= {
+          signal,
+          promise: this.#prepare(signal),
+        });
+        try {
+          await pending.promise;
+          break;
+        } catch (cause: unknown) {
+          if (this.#prepared === pending) this.#prepared = undefined;
+          if (
+            pending.signal !== signal &&
+            signal?.aborted !== true &&
+            artifactReadCancelledBy(cause, pending.signal)
+          ) {
+            this.#assertOpen(signal);
+            continue;
+          }
+          throw cause;
+        }
+      }
       await this.#source.read(signal);
     });
     const coverage = this.#described?.coverage;
@@ -230,6 +254,8 @@ export class DecodedRegionArtifactReader<Image> implements ArtifactReader {
 
   async #prepare(signal?: AbortSignal): Promise<void> {
     const described = this.#decoding.describe(await this.#decode(signal));
+    const entries = new Map<string, ArtifactEntry>();
+    const regions = new Map<string, Uint8Array>();
     let index = 0;
     for (const region of described.regions) {
       if (index % 256 === 0) await setImmediate();
@@ -248,13 +274,16 @@ export class DecodedRegionArtifactReader<Image> implements ArtifactReader {
         adapterKey: `${this.#decoding.keyPrefix}:${String(index)}`,
         formatMetadata: region.formatMetadata,
       } satisfies ArtifactEntry;
-      this.#entries.set(entry.adapterKey, entry);
-      this.#regions.set(entry.adapterKey, region.bytes);
+      entries.set(entry.adapterKey, entry);
+      regions.set(entry.adapterKey, region.bytes);
       index += 1;
     }
     const { regions: _regions, ...facts } = described;
-    this.#described = facts;
     await this.#source.read(signal);
+    this.#assertOpen(signal);
+    this.#entries = entries;
+    this.#regions = regions;
+    this.#described = facts;
   }
 
   async #decode(signal?: AbortSignal): Promise<Image> {

src/artifacts/FitArtifactReader.test.ts

@@ -1,3 +1,4 @@
+import { artifactPreparationRecoveryCases } from "../../tests/fixtures/artifactReaderLifecycle.js";
 import { getEventListeners, once } from "node:events";
 import { readFile, truncate, writeFile } from "node:fs/promises";
 import { join } from "node:path";
@@ -404,6 +405,38 @@ it("keeps the cancellation reason before admission", async () => {
   }
 });
 
+it("does not publish FIT entries when final source verification fails", async () => {
+  const path = await writeFit("partial-publication.itb", FIT_EMBEDDED);
+  const control = new FitArtifactReader(path);
+  const reader = new FitArtifactReader(path);
+  const expected = await firstEntry(control);
+  const nativeRead = ArtifactFileSource.prototype.read;
+  const physical = Object.assign(
+    new Error("final source verification failed"),
+    { code: "EIO" },
+  );
+  let calls = 0;
+  const read = vi
+    .spyOn(ArtifactFileSource.prototype, "read")
+    .mockImplementation(function (this: ArtifactFileSource, ...args) {
+      calls += 1;
+      if (calls === 4) return Promise.reject(physical);
+      return nativeRead.apply(this, args);
+    });
+  try {
+    await expect(firstEntry(reader)).rejects.toBe(physical);
+    await expect(reader.open(expected)).rejects.toMatchObject({
+      reason: "integrity",
+    });
+    read.mockRestore();
+    expect((await firstEntry(reader)).path).toBe(expected.path);
+  } finally {
+    read.mockRestore();
+    await reader.close();
+    await control.close();
+  }
+});
+
 it("keeps the cancellation reason during hash verification", async () => {
   const reader = new FitArtifactReader(
     await writeFit("hashing.itb", LARGE_FIT),
@@ -430,6 +463,8 @@ it("keeps the cancellation reason during hash verification", async () => {
       reader.entries(controller.signal)[Symbol.asyncIterator]().next(),
     ).rejects.toMatchObject({ reason: "cancelled", cause: reason });
     expect(observer).toHaveBeenCalledWith("sha256");
+    observer.mockRestore();
+    expect((await firstEntry(reader)).kind).toBe("file");
   } finally {
     observer.mockRestore();
     await reader.close();
@@ -569,3 +604,8 @@ it("preserves a final validation failure when cancellation overlaps it", async (
     await reader.close();
   }
 });
+
+artifactPreparationRecoveryCases({
+  image: () => FIT_EMBEDDED,
+  createReader: (path, admitted) => new FitArtifactReader(path, admitted),
+});

src/artifacts/FitArtifactReader.ts

@@ -13,7 +13,10 @@ import {
   type ArtifactTraversalOptions,
 } from "./ArtifactReader.js";
 import { ArtifactResourceScope } from "./ArtifactResourceScope.js";
-import { assertArtifactReadActive } from "./ArtifactFileOperation.js";
+import {
+  artifactReadCancelledBy,
+  assertArtifactReadActive,
+} from "./ArtifactFileOperation.js";
 import { ArtifactFileSource } from "./ArtifactFileSource.js";
 import {
   createFitDigest,
@@ -30,11 +33,13 @@ const FDT_HEADER_BYTES = 40;
 export class FitArtifactReader implements ArtifactReader {
   readonly format = "fit" as const;
   readonly #scope = new ArtifactResourceScope();
-  readonly #entries = new Map<string, ArtifactEntry>();
+  #entries = new Map<string, ArtifactEntry>();
   readonly #streams = new Set<Readable>();
   readonly #path: string;
   readonly #source: ArtifactFileSource;
-  #prepared: Promise<void> | undefined;
+  #prepared:
+    | { signal: AbortSignal | undefined; promise: Promise<void> }
+    | undefined;
   #metadata: FitMetadata | undefined;
   #coverage: ArtifactReaderTraversal | undefined;
   #closing = false;
@@ -66,8 +71,27 @@ export class FitArtifactReader implements ArtifactReader {
   ): AsyncIterable<ArtifactEntry> {
     await this.#scope.run(async () => {
       this.#assertOpen(signal);
-      this.#prepared ??= this.#prepare(signal);
-      await this.#prepared;
+      for (;;) {
+        const pending = (this.#prepared ??= {
+          signal,
+          promise: this.#prepare(signal),
+        });
+        try {
+          await pending.promise;
+          break;
+        } catch (cause: unknown) {
+          if (this.#prepared === pending) this.#prepared = undefined;
+          if (
+            pending.signal !== signal &&
+            signal?.aborted !== true &&
+            artifactReadCancelledBy(cause, pending.signal)
+          ) {
+            this.#assertOpen(signal);
+            continue;
+          }
+          throw cause;
+        }
+      }
       await this.#source.read(signal);
     });
     if (this.#coverage !== undefined)
@@ -207,6 +231,7 @@ export class FitArtifactReader implements ArtifactReader {
       this.#path,
       initial.size,
     );
+    const entries = new Map<string, ArtifactEntry>();
     let index = 0;
     for (const image of structure.images) {
       if (index % 256 === 0) await setImmediate();
@@ -228,15 +253,17 @@ export class FitArtifactReader implements ArtifactReader {
         adapterKey: `fit:${String(image.index)}`,
         formatMetadata: image.metadata,
       } satisfies ArtifactEntry;
-      this.#entries.set(entry.adapterKey, entry);
+      entries.set(entry.adapterKey, entry);
     }
+    await this.#source.read(signal);
+    this.#assertOpen(signal);
+    this.#entries = entries;
     this.#metadata = structure.metadata;
     if (structure.limitations.length > 0)
       this.#coverage = {
         status: "complete",
         limitations: structure.limitations,
       };
-    await this.#source.read(signal);
   }
 
   /** One streamed pass per image feeds every supported digest. */

src/artifacts/HexRecordArtifactReader.test.ts

@@ -1,3 +1,6 @@
+import { artifactPreparationRecoveryCases } from "../../tests/fixtures/artifactReaderLifecycle.js";
+import { ArtifactFileSource } from "./ArtifactFileSource.js";
+import { openRegularFile } from "../filesystem/RegularFile.js";
 import { writeFile } from "node:fs/promises";
 import { join } from "node:path";
 
@@ -159,6 +162,72 @@ it("settles an active region stream when closed", async () => {
   }
 });
 
+it("retries decoded-region preparation cancelled while reading", async () => {
+  const path = await writeImage("retry.hex", INTEL_HEX_I32_GAP);
+  const admitted = await openRegularFile(path, { symlinks: "reject" });
+  const reader = new HexRecordArtifactReader(path, "intel-hex", admitted);
+  const controller = new AbortController();
+  const reason = { stage: "decoded region read" };
+  const nativeRead = admitted.handle.read.bind(admitted.handle);
+  const read = vi.spyOn(admitted.handle, "read").mockImplementation((async (
+    ...args: unknown[]
+  ) => {
+    const result = await Reflect.apply(nativeRead, admitted.handle, args);
+    controller.abort(reason);
+    return result;
+  }) as typeof admitted.handle.read);
+  try {
+    await expect(
+      reader.entries(controller.signal)[Symbol.asyncIterator]().next(),
+    ).rejects.toMatchObject({ reason: "cancelled", cause: reason });
+    read.mockRestore();
+    expect(
+      (await reader.entries()[Symbol.asyncIterator]().next()).value,
+    ).toMatchObject({ kind: "file" });
+  } finally {
+    read.mockRestore();
+    await reader.close();
+    await admitted.handle.close();
+  }
+});
+
+it("does not publish decoded regions when final source verification fails", async () => {
+  const path = await writeImage("partial.hex", INTEL_HEX_I32_GAP);
+  const control = new HexRecordArtifactReader(path, "intel-hex");
+  const reader = new HexRecordArtifactReader(path, "intel-hex");
+  const expected = (await control.entries()[Symbol.asyncIterator]().next())
+    .value!;
+  const nativeRead = ArtifactFileSource.prototype.read;
+  const physical = Object.assign(
+    new Error("final source verification failed"),
+    { code: "EIO" },
+  );
+  let calls = 0;
+  const read = vi
+    .spyOn(ArtifactFileSource.prototype, "read")
+    .mockImplementation(function (this: ArtifactFileSource, ...args) {
+      calls += 1;
+      if (calls === 2) return Promise.reject(physical);
+      return nativeRead.apply(this, args);
+    });
+  try {
+    await expect(reader.entries()[Symbol.asyncIterator]().next()).rejects.toBe(
+      physical,
+    );
+    await expect(reader.open(expected)).rejects.toMatchObject({
+      reason: "integrity",
+    });
+    read.mockRestore();
+    expect(
+      (await reader.entries()[Symbol.asyncIterator]().next()).value,
+    ).toMatchObject({ path: expected.path });
+  } finally {
+    read.mockRestore();
+    await reader.close();
+    await control.close();
+  }
+});
+
 it("honors cancellation before admitting a file with the exact reason", async () => {
   const reader = new HexRecordArtifactReader(
     await writeImage("cancel.hex", INTEL_HEX_I32_GAP),
@@ -337,3 +406,9 @@ it("bounds copied region chunks and isolates cached bytes across opens", async (
     await reader.close();
   }
 });
+
+artifactPreparationRecoveryCases({
+  image: () => Buffer.from(INTEL_HEX_I32_GAP),
+  createReader: (path, admitted) =>
+    new HexRecordArtifactReader(path, "intel-hex", admitted),
+});

src/artifacts/NtfsMftArtifactReader.test.ts

@@ -0,0 +1,211 @@
+import { writeFile } from "node:fs/promises";
+import { join } from "node:path";
+
+import { expect, it, vi } from "vitest";
+
+import { ArtifactFileSource } from "./ArtifactFileSource.js";
+import { assertArtifactReadActive } from "./ArtifactFileOperation.js";
+import { NtfsMftArtifactReader } from "./NtfsMftArtifactReader.js";
+
+import { MFT_DEFAULT_RECORD_BYTES } from "../domain/windows/ntfsMft.js";
+import { createTestTempDirectory } from "../../tests/fixtures/temporaryDirectory.js";
+import { scanCanonicalArtifactInventory } from "../../tests/fixtures/artifactInventory.js";
+
+/** A 1 KiB FILE record whose update sequence array covers both sectors. */
+const recordBytes = (options: {
+  readonly sequenceNumber?: number;
+  readonly flags?: number;
+  readonly index?: number;
+}): Buffer => {
+  const record = Buffer.alloc(MFT_DEFAULT_RECORD_BYTES);
+  record.write("FILE", 0, "latin1");
+  const sequenceNumber = options.sequenceNumber ?? 1;
+  record.writeUInt16LE(48, 4);
+  record.writeUInt16LE(3, 6);
+  record.writeUInt16LE(sequenceNumber, 48);
+  record.writeUInt16LE(0xa5a5, 50);
+  record.writeUInt16LE(0xa5a5, 52);
+  record.writeUInt16LE(sequenceNumber, 16);
+  record.writeUInt16LE(1, 18);
+  record.writeUInt32LE(options.index ?? 0, 44);
+  record.writeUInt16LE(56, 20);
+  record.writeUInt16LE(options.flags ?? 0x0001, 22);
+  record.writeUInt32LE(64, 24);
+  record.writeUInt32LE(0xffff_ffff, 56);
+  record.writeUInt32LE(MFT_DEFAULT_RECORD_BYTES, 28);
+  record.writeUInt16LE(sequenceNumber, 510);
+  record.writeUInt16LE(sequenceNumber, 1022);
+  return record;
+};
+
+const writeMft = async (name: string, bytes: Buffer): Promise<string> => {
+  const directory = await createTestTempDirectory("rea-ntfs-mft-");
+  const path = join(directory, name);
+  await writeFile(path, bytes);
+  return path;
+};
+
+it("reports a $MFT stream's record headers as the root occurrence", async () => {
+  const inventory = await scanCanonicalArtifactInventory(
+    await writeMft(
+      "$MFT",
+      Buffer.concat([
+        recordBytes({}),
+        recordBytes({ flags: 0x0003, index: 1 }),
+      ]),
+    ),
+  );
+  expect(inventory.manifest.root_format).toBe("ntfs-mft");
+  expect(inventory.coverage.status).toBe("complete");
+  // A record stream declares no members, so the root is its whole inventory.
+  expect(inventory.occurrences.map((row) => row.logical_path)).toEqual(["."]);
+  expect(inventory.occurrences[0]).toMatchObject({
+    artifact_format: "ntfs-mft",
+    format_metadata: {
+      kind: "ntfs-mft",
+      record_bytes: MFT_DEFAULT_RECORD_BYTES,
+      record_count: 2,
+      in_use_record_count: 2,
+      directory_record_count: 1,
+      extension_record_count: 0,
+      fixup_mismatch_count: 0,
+      fixup_unverified_record_count: 0,
+      truncated_tail_bytes: 0,
+    },
+  });
+});
+
+it("names a failed fixup in the coverage limitations", async () => {
+  const record = recordBytes({ sequenceNumber: 4 });
+  // A trailer that is not the sequence number marks the sector as not cleanly written.
+  record.writeUInt16LE(0x1234, 510);
+  const inventory = await scanCanonicalArtifactInventory(
+    await writeMft("$MFT", record),
+  );
+  expect(inventory.manifest.root_format).toBe("ntfs-mft");
+  expect(inventory.coverage.status).toBe("unknown");
+  expect(inventory.limitations).toContainEqual(
+    expect.stringMatching(/sector trailer that is not the update sequence/u),
+  );
+});
+
+it("reports an unaligned used-byte count as incomplete coverage", async () => {
+  const record = recordBytes({});
+  record.writeUInt32LE(57, 24);
+  const inventory = await scanCanonicalArtifactInventory(
+    await writeMft("$MFT", record),
+  );
+  expect(inventory.coverage.status).toBe("unknown");
+  expect(inventory.occurrences[0]?.format_metadata).toMatchObject({
+    record_headers: [{ usedBytes: 57, extentValid: false }],
+  });
+});
+
+it("reports incomplete coverage when FILE extents contradict the record slots", async () => {
+  const usedTooLarge = recordBytes({});
+  usedTooLarge.writeUInt32LE(4096, 24);
+  const totalTooSmall = recordBytes({});
+  totalTooSmall.writeUInt32LE(512, 28);
+
+  const inventory = await scanCanonicalArtifactInventory(
+    await writeMft("$MFT", Buffer.concat([usedTooLarge, totalTooSmall])),
+  );
+  expect(inventory.coverage.status).toBe("unknown");
+  expect(inventory.limitations).toContainEqual(
+    "2 FILE record(s) declare inconsistent record extents, beginning at offset 0",
+  );
+  expect(inventory.occurrences[0]?.format_metadata).toMatchObject({
+    record_count: 2,
+    fixup_mismatch_count: 0,
+    fixup_unverified_record_count: 0,
+  });
+});
+
+it("retries after final source verification fails without publishing its metadata", async () => {
+  const reader = new NtfsMftArtifactReader(
+    await writeMft("$MFT", recordBytes({})),
+  );
+  const read = ArtifactFileSource.prototype.read;
+  const spy = vi
+    .spyOn(ArtifactFileSource.prototype, "read")
+    .mockImplementationOnce(function (this: ArtifactFileSource, signal) {
+      return read.call(this, signal);
+    })
+    .mockRejectedValueOnce(new Error("final source verification failed"));
+  try {
+    await expect(
+      reader.entries()[Symbol.asyncIterator]().next(),
+    ).rejects.toThrow("final source verification failed");
+    expect(reader.formatMetadata()).toBeUndefined();
+    await expect(
+      reader.entries()[Symbol.asyncIterator]().next(),
+    ).resolves.toMatchObject({
+      done: true,
+    });
+    expect(reader.formatMetadata()).toMatchObject({
+      kind: "ntfs-mft",
+      record_count: 1,
+    });
+  } finally {
+    spy.mockRestore();
+    await reader.close();
+  }
+});
+
+it("retries a shared preparation cancelled by another traversal", async () => {
+  const reader = new NtfsMftArtifactReader(
+    await writeMft("$MFT", recordBytes({})),
+  );
+  const controller = new AbortController();
+  let entered: () => void = () => {};
+  const started = new Promise<void>((resolve) => {
+    entered = resolve;
+  });
+  const read = ArtifactFileSource.prototype.read;
+  const spy = vi
+    .spyOn(ArtifactFileSource.prototype, "read")
+    .mockImplementationOnce(async function (this: ArtifactFileSource, signal) {
+      entered();
+      await new Promise<void>((resolve) => {
+        signal?.addEventListener("abort", () => resolve(), { once: true });
+      });
+      assertArtifactReadActive(signal);
+      return read.call(this, signal);
+    });
+  try {
+    const cancelled = reader
+      .entries(controller.signal)
+      [Symbol.asyncIterator]()
+      .next();
+    const cancelledResult = expect(cancelled).rejects.toMatchObject({
+      reason: "cancelled",
+    });
+    await started;
+    const other = reader.entries()[Symbol.asyncIterator]().next();
+    controller.abort("cancelled first traversal");
+    await cancelledResult;
+    await expect(other).resolves.toMatchObject({ done: true });
+    expect(reader.formatMetadata()).toMatchObject({ record_count: 1 });
+  } finally {
+    controller.abort();
+    spy.mockRestore();
+    await reader.close();
+  }
+});
+
+it("revalidates the source before reusing a completed header inventory", async () => {
+  const path = await writeMft("$MFT", recordBytes({}));
+  const reader = new NtfsMftArtifactReader(path);
+  try {
+    await reader.entries()[Symbol.asyncIterator]().next();
+    await writeFile(
+      path,
+      Buffer.concat([recordBytes({}), recordBytes({ index: 1 })]),
+    );
+    await expect(
+      reader.entries()[Symbol.asyncIterator]().next(),
+    ).rejects.toMatchObject({ reason: "integrity" });
+  } finally {
+    await reader.close();
+  }
+});

src/artifacts/NtfsMftArtifactReader.ts

@@ -0,0 +1,174 @@
+import type { Readable } from "node:stream";
+
+import {
+  ntfsMftMetadataSchema,
+  type ArtifactFormatMetadata,
+} from "../domain/artifactFormatMetadata.js";
+import type { RegionDecodeFailure } from "../domain/firmware/regionDecoder.js";
+import {
+  createMftDecoder,
+  type MftInventory,
+} from "../domain/windows/ntfsMft.js";
+import {
+  ArtifactReaderFailure,
+  type ArtifactEntry,
+  type ArtifactReader,
+  type ArtifactReaderTraversal,
+  type ArtifactTraversalOptions,
+} from "./ArtifactReader.js";
+import { ArtifactResourceScope } from "./ArtifactResourceScope.js";
+import {
+  artifactReadCancelledBy,
+  assertArtifactReadActive,
+} from "./ArtifactFileOperation.js";
+import { ArtifactFileSource } from "./ArtifactFileSource.js";
+import { readFileHandleChunks } from "../filesystem/readFileHandleChunks.js";
+import type { StableRegularFileDescriptor } from "../filesystem/RegularFile.js";
+
+/**
+ * An MFT stream is its own content: the FILE records describe the source
+ * volume, not members inside the stream, so header facts are attached to
+ * the root occurrence.
+ */
+export class NtfsMftArtifactReader implements ArtifactReader {
+  readonly format = "ntfs-mft" as const;
+  readonly #scope = new ArtifactResourceScope();
+  readonly #source: ArtifactFileSource;
+  #prepared:
+    | { signal: AbortSignal | undefined; promise: Promise<void> }
+    | undefined;
+  #metadata: ArtifactFormatMetadata | undefined;
+  #limitations: readonly string[] = [];
+  #closing = false;
+
+  constructor(path: string, admitted?: StableRegularFileDescriptor) {
+    this.#source = new ArtifactFileSource(path, "$MFT stream", admitted);
+  }
+
+  formatMetadata(): ArtifactFormatMetadata | undefined {
+    return this.#metadata === undefined
+      ? undefined
+      : structuredClone(this.#metadata);
+  }
+
+  async prepareContainer(
+    expectedSha256: string,
+    signal?: AbortSignal,
+  ): Promise<void> {
+    await this.#scope.run(async () => {
+      this.#assertOpen(signal);
+      await this.#source.bindSha256(expectedSha256, signal);
+    });
+  }
+
+  async *entries(
+    signal?: AbortSignal,
+    options?: ArtifactTraversalOptions,
+  ): AsyncIterable<ArtifactEntry> {
+    await this.#scope.run(async () => {
+      this.#assertOpen(signal);
+      for (;;) {
+        const pending = (this.#prepared ??= {
+          signal,
+          promise: this.#prepare(signal),
+        });
+        try {
+          await pending.promise;
+          break;
+        } catch (cause: unknown) {
+          if (this.#prepared === pending) this.#prepared = undefined;
+          if (
+            pending.signal !== signal &&
+            signal?.aborted !== true &&
+            artifactReadCancelledBy(cause, pending.signal)
+          ) {
+            this.#assertOpen(signal);
+            continue;
+          }
+          throw cause;
+        }
+      }
+      this.#assertOpen(signal);
+      await this.#source.read(signal);
+    });
+    // An MFT declares no members, so the root occurrence is the whole
+    // inventory; the coverage callback reports what its parse did not establish.
+    options?.onCoverage?.(coverageOf(this.#limitations));
+  }
+
+  async open(entry: ArtifactEntry, _signal?: AbortSignal): Promise<Readable> {
+    return this.#scope.run(async () => {
+      throw new ArtifactReaderFailure(
+        "unavailable",
+        `$MFT stream declares no member to open for ${entry.path}`,
+      );
+    });
+  }
+
+  async close(): Promise<void> {
+    this.#closing = true;
+    await this.#scope.close();
+    await this.#source.close();
+    this.#metadata = undefined;
+  }
+
+  provenance(): readonly [] {
+    return [];
+  }
+
+  #assertOpen(signal?: AbortSignal): void {
+    assertArtifactReadActive(signal);
+    if (this.#closing)
+      throw new ArtifactReaderFailure(
+        "unavailable",
+        "$MFT stream reader is closed",
+      );
+  }
+
+  async #prepare(signal?: AbortSignal): Promise<void> {
+    const image = await this.#decode(signal);
+    await this.#source.read(signal);
+    this.#metadata = ntfsMftMetadataSchema.parse(image.metadata);
+    this.#limitations = image.limitations;
+  }
+
+  async #decode(signal?: AbortSignal): Promise<MftInventory> {
+    const { handle, initial } = await this.#source.read(signal);
+    const decoder = createMftDecoder(initial.size);
+    if (!decoder.ok) throw this.#failure(decoder.error);
+    if (initial.size > 0)
+      for await (const chunk of readFileHandleChunks(handle, {
+        end: initial.size - 1,
+        signal,
+        cancellationFailure: (reason) =>
+          new ArtifactReaderFailure("cancelled", "$MFT stream read cancelled", {
+            cause: reason,
+          }),
+      }) as AsyncIterable<Buffer>) {
+        this.#assertOpen(signal);
+        const written = decoder.value.write(chunk);
+        if (!written.ok) throw this.#failure(written.error);
+      }
+    const image = decoder.value.finish();
+    if (!image.ok) throw this.#failure(image.error);
+    return image.value;
+  }
+
+  #failure(failure: RegionDecodeFailure): ArtifactReaderFailure {
+    return failure.kind === "format"
+      ? new ArtifactReaderFailure(
+          "format",
+          `$MFT stream is malformed: ${failure.message}`,
+        )
+      : new ArtifactReaderFailure(
+          "limit",
+          `$MFT stream exceeds a fixed limit: ${failure.message}`,
+          failure.boundary === undefined ? {} : { boundary: failure.boundary },
+        );
+  }
+}
+
+const coverageOf = (limitations: readonly string[]): ArtifactReaderTraversal =>
+  limitations.length === 0
+    ? { status: "complete", limitations: [] }
+    : { status: "unknown", limitations };

src/artifacts/UImageArtifactReader.boundaries.test.ts

@@ -1,4 +1,7 @@
-import { artifactReaderLifecycleCases } from "../../tests/fixtures/artifactReaderLifecycle.js";
+import {
+  artifactPreparationRecoveryCases,
+  artifactReaderLifecycleCases,
+} from "../../tests/fixtures/artifactReaderLifecycle.js";
 import { writeFile } from "node:fs/promises";
 import { join } from "node:path";
 import { buffer } from "node:stream/consumers";
@@ -91,6 +94,14 @@ it("preserves cancellation during payload CRC admission", async () => {
         }
       })(),
     ).rejects.toMatchObject({ reason: "cancelled", cause: reason });
+    reads.mockRestore();
+    const recovered = await entries(reader);
+    expect(recovered).toHaveLength(1);
+    expect(
+      (await buffer(await reader.open(recovered[0]!))).equals(
+        Buffer.alloc(128 * 1024, 0x61),
+      ),
+    ).toBe(true);
   } finally {
     await reader.close();
     reads.mockRestore();
@@ -205,3 +216,8 @@ it("reports impossible over-cap component extents as malformed", async () => {
     await reader.close();
   }
 });
+
+artifactPreparationRecoveryCases({
+  image: () => buildUImage({ payload: Buffer.from("payload") }),
+  createReader: (path, admitted) => new UImageArtifactReader(path, admitted),
+});

src/artifacts/UImageArtifactReader.ts

@@ -8,7 +8,10 @@ import type {
   UImageMetadata,
 } from "../domain/artifactFormatMetadata.js";
 import { ArtifactFileSource } from "./ArtifactFileSource.js";
-import { assertArtifactReadActive } from "./ArtifactFileOperation.js";
+import {
+  artifactReadCancelledBy,
+  assertArtifactReadActive,
+} from "./ArtifactFileOperation.js";
 import {
   ArtifactReaderFailure,
   copyArtifactEntry,
@@ -47,7 +50,9 @@ export class UImageArtifactReader implements ArtifactReader {
   readonly #entries = new Map<string, UImageEntry>();
   readonly #streams = new Set<Readable>();
   readonly #source: ArtifactFileSource;
-  #prepared: Promise<void> | undefined;
+  #prepared:
+    | { signal: AbortSignal | undefined; promise: Promise<void> }
+    | undefined;
   #header: UImageMetadata | undefined;
   #closing = false;
 
@@ -75,8 +80,27 @@ export class UImageArtifactReader implements ArtifactReader {
   ): AsyncIterable<ArtifactEntry> {
     await this.#scope.run(async () => {
       this.#assertOpen(signal);
-      this.#prepared ??= this.#prepare(signal);
-      await this.#prepared;
+      for (;;) {
+        const pending = (this.#prepared ??= {
+          signal,
+          promise: this.#prepare(signal),
+        });
+        try {
+          await pending.promise;
+          break;
+        } catch (cause: unknown) {
+          if (this.#prepared === pending) this.#prepared = undefined;
+          if (
+            pending.signal !== signal &&
+            signal?.aborted !== true &&
+            artifactReadCancelledBy(cause, pending.signal)
+          ) {
+            this.#assertOpen(signal);
+            continue;
+          }
+          throw cause;
+        }
+      }
       await this.#source.read(signal);
     });
     this.#assertOpen(signal);

src/artifacts/UnityFsArtifactReader.boundaries.test.ts

@@ -1,3 +1,5 @@
+import { artifactPreparationRecoveryCases } from "../../tests/fixtures/artifactReaderLifecycle.js";
+import { ArtifactFileSource } from "./ArtifactFileSource.js";
 import { truncate, writeFile } from "node:fs/promises";
 import { join } from "node:path";
 import { buffer, text } from "node:stream/consumers";
@@ -43,6 +45,68 @@ const repeatedBlock = (data: Buffer): Buffer => {
   ]);
 };
 
+it("retries UnityFs preparation cancelled during file reading", async () => {
+  const path = await fixture(buildUnityFs([{ path: "a", data: "x" }]));
+  const admitted = await openRegularFile(path, { symlinks: "reject" });
+  const reader = new UnityFsArtifactReader(path, admitted);
+  const controller = new AbortController();
+  const reason = { stage: "header read" };
+  const nativeRead = admitted.handle.read.bind(admitted.handle);
+  const read = vi.spyOn(admitted.handle, "read").mockImplementation((async (
+    ...args: unknown[]
+  ) => {
+    const result = await Reflect.apply(nativeRead, admitted.handle, args);
+    controller.abort(reason);
+    return result;
+  }) as typeof admitted.handle.read);
+  try {
+    await expect(
+      (async () => {
+        for await (const _ of reader.entries(controller.signal)) {
+        }
+      })(),
+    ).rejects.toMatchObject({ reason: "cancelled", cause: reason });
+    read.mockRestore();
+    expect((await entries(reader)).map((entry) => entry.path)).toEqual(["a"]);
+  } finally {
+    read.mockRestore();
+    await reader.close();
+    await admitted.handle.close();
+  }
+});
+
+it("does not publish UnityFs entries when final source verification fails", async () => {
+  const path = await fixture(buildUnityFs([{ path: "a", data: "x" }]));
+  const control = new UnityFsArtifactReader(path);
+  const reader = new UnityFsArtifactReader(path);
+  const expected = (await entries(control))[0]!;
+  const nativeRead = ArtifactFileSource.prototype.read;
+  const physical = Object.assign(
+    new Error("final source verification failed"),
+    { code: "EIO" },
+  );
+  let calls = 0;
+  const read = vi
+    .spyOn(ArtifactFileSource.prototype, "read")
+    .mockImplementation(function (this: ArtifactFileSource, ...args) {
+      calls += 1;
+      if (calls === 2) return Promise.reject(physical);
+      return nativeRead.apply(this, args);
+    });
+  try {
+    await expect(entries(reader)).rejects.toBe(physical);
+    await expect(reader.open(expected)).rejects.toMatchObject({
+      reason: "integrity",
+    });
+    read.mockRestore();
+    expect((await entries(reader)).map((entry) => entry.path)).toEqual(["a"]);
+  } finally {
+    read.mockRestore();
+    await reader.close();
+    await control.close();
+  }
+});
+
 describe("UnityFS allocation and malformed-input boundaries", () => {
   it("rejects an oversized directory extending past the physical file", async () => {
     const bytes = buildUnityFs([{ path: "data", data: "x" }]);
@@ -431,3 +495,8 @@ describe("UnityFS stream capacity and cache ownership", () => {
     }
   });
 });
+
+artifactPreparationRecoveryCases({
+  image: () => buildUnityFs([{ path: "a", data: "x" }]),
+  createReader: (path, admitted) => new UnityFsArtifactReader(path, admitted),
+});

src/artifacts/UnityFsArtifactReader.ts

@@ -12,7 +12,10 @@ import {
   type ArtifactTraversalOptions,
 } from "./ArtifactReader.js";
 import { ArtifactResourceScope } from "./ArtifactResourceScope.js";
-import { assertArtifactReadActive } from "./ArtifactFileOperation.js";
+import {
+  artifactReadCancelledBy,
+  assertArtifactReadActive,
+} from "./ArtifactFileOperation.js";
 import { readArtifactFileBytes } from "./ArtifactFileRange.js";
 import type {
   ArtifactFormatMetadata,
@@ -52,10 +55,12 @@ interface UnityFsEntryState {
 export class UnityFsArtifactReader implements ArtifactReader {
   readonly format = "unityfs" as const;
   readonly #scope = new ArtifactResourceScope();
-  readonly #entries = new Map<string, UnityFsEntryState>();
+  #entries = new Map<string, UnityFsEntryState>();
   readonly #streams = new Map<Readable, AbortController>();
   readonly #source: ArtifactFileSource;
-  #prepared: Promise<void> | undefined;
+  #prepared:
+    | { signal: AbortSignal | undefined; promise: Promise<void> }
+    | undefined;
   #header: UnityFsMetadata | undefined;
   #blocks: readonly UnityFsBlock[] = [];
   #dataStart = 0;
@@ -88,8 +93,27 @@ export class UnityFsArtifactReader implements ArtifactReader {
   ): AsyncIterable<ArtifactEntry> {
     await this.#scope.run(async () => {
       this.#assertOpen(signal);
-      this.#prepared ??= this.#prepare(signal);
-      await this.#prepared;
+      for (;;) {
+        const pending = (this.#prepared ??= {
+          signal,
+          promise: this.#prepare(signal),
+        });
+        try {
+          await pending.promise;
+          break;
+        } catch (cause: unknown) {
+          if (this.#prepared === pending) this.#prepared = undefined;
+          if (
+            pending.signal !== signal &&
+            signal?.aborted !== true &&
+            artifactReadCancelledBy(cause, pending.signal)
+          ) {
+            this.#assertOpen(signal);
+            continue;
+          }
+          throw cause;
+        }
+      }
       await this.#source.read(signal);
     });
     this.#assertOpen(signal);
@@ -229,7 +253,7 @@ export class UnityFsArtifactReader implements ArtifactReader {
     if (header === undefined)
       throw unityFsFormatFailure("no complete UnityFS header");
     const status = header.layout.status;
-    this.#header = {
+    const metadata: UnityFsMetadata = {
       kind: "unityfs",
       signature: "UnityFS",
       format_version: header.formatVersion,
@@ -245,7 +269,12 @@ export class UnityFsArtifactReader implements ArtifactReader {
       data_hash: null,
       directory_status: status,
     };
-    if (status !== "available") return;
+    if (status !== "available") {
+      await this.#source.read(signal);
+      this.#assertOpen(signal);
+      this.#header = metadata;
+      return;
+    }
     const { dataHash, blocks, nodes, dataStart } = await readUnityFsBlocksInfo(
       handle,
       header,
@@ -259,44 +288,55 @@ export class UnityFsArtifactReader implements ArtifactReader {
       if (streamBytes > BigInt(budget))
         throw expansionLimit("declared virtual stream", streamBytes, budget);
     }
-    this.#header = {
-      ...this.#header,
+    const preparedMetadata: UnityFsMetadata = {
+      ...metadata,
       data_hash: dataHash,
       block_count: blocks.length,
       data_start: dataStart,
     };
-    this.#dataStart = dataStart;
     // Keep the source block count above, but do not revisit zero-output rows
     // for every member. Parsed stream/compressed offsets remain authoritative.
-    this.#blocks = blocks.filter((block) => block.uncompressedSize > 0);
-    this.#remainingDecodeBytes = budget;
-    await this.#registerNodes(nodes, budget, signal);
+    const readableBlocks = blocks.filter((block) => block.uncompressedSize > 0);
+    const registered = await this.#registerNodes(
+      nodes,
+      readableBlocks,
+      budget,
+      signal,
+    );
     await this.#source.read(signal);
+    this.#assertOpen(signal);
+    this.#header = preparedMetadata;
+    this.#dataStart = dataStart;
+    this.#blocks = readableBlocks;
+    this.#remainingDecodeBytes = budget;
+    this.#entries = registered.entries;
+    this.#remainingReadOperations = registered.remainingReadOperations;
   }
 
   async #registerNodes(
     nodes: readonly UnityFsNode[],
+    blocks: readonly UnityFsBlock[],
     budget: number,
     signal?: AbortSignal,
-  ): Promise<void> {
+  ): Promise<{
+    entries: Map<string, UnityFsEntryState>;
+    remainingReadOperations: number;
+  }> {
+    const entries = new Map<string, UnityFsEntryState>();
     let readableBytes = 0n;
     for (let index = 0; index < nodes.length; index++) {
       if (index % 256 === 0) await setImmediate();
       this.#assertOpen(signal);
       const node = nodes[index]!;
       const firstBlock =
-        node.size === 0 ? -1 : blockEndingAfter(this.#blocks, node.offset);
+        node.size === 0 ? -1 : blockEndingAfter(blocks, node.offset);
       const lastBlock =
         node.size === 0
           ? -1
-          : blockEndingAfter(this.#blocks, node.offset + node.size - 1);
+          : blockEndingAfter(blocks, node.offset + node.size - 1);
       if (node.size !== 0 && (firstBlock < 0 || lastBlock < firstBlock))
         throw unityFsFormatFailure("member has no block range");
-      const unavailable = firstUnavailableBlock(
-        this.#blocks,
-        firstBlock,
-        lastBlock,
-      );
+      const unavailable = firstUnavailableBlock(blocks, firstBlock, lastBlock);
       const kind =
         (node.flags & 1) !== 0 ? ("directory" as const) : ("file" as const);
       if (kind === "file" && unavailable === undefined) {
@@ -331,7 +371,7 @@ export class UnityFsArtifactReader implements ArtifactReader {
           flags: node.flags,
         },
       } satisfies UnityFsEntry;
-      this.#entries.set(adapterKey, {
+      entries.set(adapterKey, {
         entry,
         firstBlock,
         lastBlock,
@@ -340,14 +380,15 @@ export class UnityFsArtifactReader implements ArtifactReader {
     }
     // Four ordinary passes over admitted blocks, nodes and plaintext chunks,
     // with a small-reader floor. Byte quotas alone cannot bound tiny-block I/O.
-    this.#remainingReadOperations =
+    const remainingReadOperations =
       4 *
       Math.max(
         1024,
-        this.#blocks.length +
+        blocks.length +
           nodes.length +
           Math.ceil(Number(readableBytes) / CHUNK_BYTES),
       );
+    return { entries, remainingReadOperations };
   }
 
   async *#memberChunks(

src/artifacts/WxapkgArtifactReader.boundaries.test.ts

@@ -1,3 +1,5 @@
+import { artifactPreparationRecoveryCases } from "../../tests/fixtures/artifactReaderLifecycle.js";
+import { ArtifactFileSource } from "./ArtifactFileSource.js";
 import { once } from "node:events";
 import { writeFile } from "node:fs/promises";
 import { join } from "node:path";
@@ -26,6 +28,68 @@ const entries = async (reader: WxapkgArtifactReader) => {
   return result;
 };
 
+it("retries Wxapkg preparation cancelled during file reading", async () => {
+  const path = await fixture(buildWxapkg([{ path: "a", data: "x" }]));
+  const admitted = await openRegularFile(path, { symlinks: "reject" });
+  const reader = new WxapkgArtifactReader(path, admitted);
+  const controller = new AbortController();
+  const reason = { stage: "header read" };
+  const nativeRead = admitted.handle.read.bind(admitted.handle);
+  const read = vi.spyOn(admitted.handle, "read").mockImplementation((async (
+    ...args: unknown[]
+  ) => {
+    const result = await Reflect.apply(nativeRead, admitted.handle, args);
+    controller.abort(reason);
+    return result;
+  }) as typeof admitted.handle.read);
+  try {
+    await expect(
+      (async () => {
+        for await (const _ of reader.entries(controller.signal)) {
+        }
+      })(),
+    ).rejects.toMatchObject({ reason: "cancelled", cause: reason });
+    read.mockRestore();
+    expect((await entries(reader)).map((entry) => entry.path)).toEqual(["a"]);
+  } finally {
+    read.mockRestore();
+    await reader.close();
+    await admitted.handle.close();
+  }
+});
+
+it("does not publish Wxapkg entries when final source verification fails", async () => {
+  const path = await fixture(buildWxapkg([{ path: "a", data: "x" }]));
+  const control = new WxapkgArtifactReader(path);
+  const reader = new WxapkgArtifactReader(path);
+  const expected = (await entries(control))[0]!;
+  const nativeRead = ArtifactFileSource.prototype.read;
+  const physical = Object.assign(
+    new Error("final source verification failed"),
+    { code: "EIO" },
+  );
+  let calls = 0;
+  const read = vi
+    .spyOn(ArtifactFileSource.prototype, "read")
+    .mockImplementation(function (this: ArtifactFileSource, ...args) {
+      calls += 1;
+      if (calls === 2) return Promise.reject(physical);
+      return nativeRead.apply(this, args);
+    });
+  try {
+    await expect(entries(reader)).rejects.toBe(physical);
+    await expect(reader.open(expected)).rejects.toMatchObject({
+      reason: "integrity",
+    });
+    read.mockRestore();
+    expect((await entries(reader)).map((entry) => entry.path)).toEqual(["a"]);
+  } finally {
+    read.mockRestore();
+    await reader.close();
+    await control.close();
+  }
+});
+
 describe("wxapkg allocation and malformed-input boundaries", () => {
   it("admits the encoded table budget before reading it", async () => {
     const bytes = buildWxapkg([{ path: "a", data: "x" }]);
@@ -308,3 +372,8 @@ describe("wxapkg stream capacity and freshness", () => {
     }
   });
 });
+
+artifactPreparationRecoveryCases({
+  image: () => buildWxapkg([{ path: "a", data: "x" }]),
+  createReader: (path, admitted) => new WxapkgArtifactReader(path, admitted),
+});

src/artifacts/WxapkgArtifactReader.ts

@@ -12,7 +12,10 @@ import {
   type ArtifactTraversalOptions,
 } from "./ArtifactReader.js";
 import { ArtifactResourceScope } from "./ArtifactResourceScope.js";
-import { assertArtifactReadActive } from "./ArtifactFileOperation.js";
+import {
+  artifactReadCancelledBy,
+  assertArtifactReadActive,
+} from "./ArtifactFileOperation.js";
 import { readArtifactFileBytes } from "./ArtifactFileRange.js";
 import type {
   ArtifactFormatMetadata,
@@ -48,10 +51,12 @@ interface WxapkgEntryState {
 export class WxapkgArtifactReader implements ArtifactReader {
   readonly format = "wxapkg" as const;
   readonly #scope = new ArtifactResourceScope();
-  readonly #entries = new Map<string, WxapkgEntryState>();
+  #entries = new Map<string, WxapkgEntryState>();
   readonly #streams = new Map<Readable, AbortController>();
   readonly #source: ArtifactFileSource;
-  #prepared: Promise<void> | undefined;
+  #prepared:
+    | { signal: AbortSignal | undefined; promise: Promise<void> }
+    | undefined;
   #header: WxapkgMetadata | undefined;
   #remainingReadOperations = 0;
   #admittedReadOperations = 0;
@@ -81,8 +86,27 @@ export class WxapkgArtifactReader implements ArtifactReader {
   ): AsyncIterable<ArtifactEntry> {
     await this.#scope.run(async () => {
       this.#assertOpen(signal);
-      this.#prepared ??= this.#prepare(signal);
-      await this.#prepared;
+      for (;;) {
+        const pending = (this.#prepared ??= {
+          signal,
+          promise: this.#prepare(signal),
+        });
+        try {
+          await pending.promise;
+          break;
+        } catch (cause: unknown) {
+          if (this.#prepared === pending) this.#prepared = undefined;
+          if (
+            pending.signal !== signal &&
+            signal?.aborted !== true &&
+            artifactReadCancelledBy(cause, pending.signal)
+          ) {
+            this.#assertOpen(signal);
+            continue;
+          }
+          throw cause;
+        }
+      }
       await this.#source.read(signal);
     });
     this.#assertOpen(signal);
@@ -201,7 +225,7 @@ export class WxapkgArtifactReader implements ArtifactReader {
     const header = await readWxapkgHeader(handle, initial.size, signal);
     if (header === undefined)
       throw wxapkgFormatFailure("no complete wxapkg header");
-    this.#header = {
+    const metadata: WxapkgMetadata = {
       kind: "wxapkg",
       info: header.layout.status === "available" ? header.info : null,
       index_length:
@@ -212,7 +236,12 @@ export class WxapkgArtifactReader implements ArtifactReader {
         header.layout.status === "available" ? header.fileCount : null,
       directory_status: header.layout.status,
     };
-    if (header.layout.status !== "available") return;
+    if (header.layout.status !== "available") {
+      await this.#source.read(signal);
+      this.#assertOpen(signal);
+      this.#header = metadata;
+      return;
+    }
     const { members } = await readWxapkgFileTable(
       handle,
       header,
@@ -220,6 +249,7 @@ export class WxapkgArtifactReader implements ArtifactReader {
       signal,
     );
     const budget = artifactExpansionBudgetBytes(initial.size);
+    const entries = new Map<string, WxapkgEntryState>();
     let declaredBytes = 0n;
     for (let index = 0; index < members.length; index++) {
       if (index % 256 === 0) await setImmediate();
@@ -234,7 +264,7 @@ export class WxapkgArtifactReader implements ArtifactReader {
           { boundary: "memory" },
         );
       const adapterKey = `wxapkg-entry:${String(index)}`;
-      this.#entries.set(adapterKey, {
+      entries.set(adapterKey, {
         entry: {
           path: member.path,
           sourcePath: member.declaredPath,
@@ -260,14 +290,18 @@ export class WxapkgArtifactReader implements ArtifactReader {
       });
     }
     // Header, table and per-chunk member reads; a small-reader floor applies.
-    this.#admittedReadOperations =
+    const admittedReadOperations =
       4 *
       Math.max(
         1024,
         members.length + Math.ceil(Number(declaredBytes) / CHUNK_BYTES),
       );
-    this.#remainingReadOperations = this.#admittedReadOperations;
     await this.#source.read(signal);
+    this.#assertOpen(signal);
+    this.#header = metadata;
+    this.#entries = entries;
+    this.#admittedReadOperations = admittedReadOperations;
+    this.#remainingReadOperations = admittedReadOperations;
   }
 
   async *#memberChunks(

src/artifacts/artifactFormats.test.ts

@@ -43,7 +43,8 @@ type FormatsWithReaders =
   | "uf2"
   | "dfuse"
   | "esp-flash-image"
-  | "fit";
+  | "fit"
+  | "ntfs-mft";
 
 const readerFormatsMatch: Equal<ArtifactReaderFormat, FormatsWithReaders> =
   true;

src/artifacts/artifactFormats.ts

@@ -40,6 +40,7 @@ import { NodeSeaArtifactReader } from "./NodeSeaArtifactReader.js";
 import type { PyInstallerAdmission } from "./pyinstaller/PyInstallerAdmission.js";
 import { PyInstallerArtifactReader } from "./PyInstallerArtifactReader.js";
 import { Uf2ArtifactReader } from "./Uf2ArtifactReader.js";
+import { NtfsMftArtifactReader } from "./NtfsMftArtifactReader.js";
 import { UImageArtifactReader } from "./UImageArtifactReader.js";
 import { UnityFsArtifactReader } from "./UnityFsArtifactReader.js";
 import { WxapkgArtifactReader } from "./WxapkgArtifactReader.js";
@@ -465,6 +466,14 @@ const NON_ZIP_FORMATS = {
     createReader: (path, options) =>
       new FitArtifactReader(path, sourceOf(options)),
   }),
+  "ntfs-mft": described("ntfs-mft", {
+    ...plain,
+    retainClassifiedSource: true,
+    readerLimitation:
+      "MFT inventory reports whole-stream counts and a bounded sample of the first 32 decoded FILE record headers, with the omitted-header count. Attributes, filenames, data streams and the $UsnJrnl:$J journal are not decoded, so no path, file size or change history is established.",
+    createReader: (path, options) =>
+      new NtfsMftArtifactReader(path, sourceOf(options)),
+  }),
   unknown: described("unknown", plain),
 } satisfies Record<
   Exclude<ArtifactFormatId, ZipPackageFormat>,

src/artifacts/inventory/ArtifactGraphConstruction.test.ts

@@ -576,6 +576,28 @@ it("classifies nested UF2 by block magics and DfuSe by prefix and suffix", () =>
   ).toMatchObject({ format: "file" });
 });
 
+it("classifies a $MFT stream by its FILE record header, under any name", () => {
+  const mft = Buffer.alloc(2_048);
+  mft.write("FILE", 0, "latin1");
+  mft.writeUInt32LE(1024, 28);
+  for (const path of ["evidence/$MFT", "evidence/mft.raw", "evidence/records"])
+    expect(classifyArtifactBytesAtPath(path, mft)).toEqual({
+      kind: "executable",
+      format: "ntfs-mft",
+    });
+});
+
+it("leaves plain content beginning with FILE unclassified as MFT", () => {
+  const text = Buffer.from(
+    "FILE is a plain text prefix, not an NTFS record header",
+  );
+  expect(
+    classifyArtifactBytesAtPath("evidence/readme.txt", text),
+  ).toMatchObject({
+    format: "file",
+  });
+});
+
 it("classifies a FIT by a /images node within the first 4 KiB", () => {
   const images = [{ name: "kernel", data: Buffer.from("kernel") }];
   const late = devicetreeFixture({

src/artifacts/inventory/ArtifactGraphConstruction.ts

@@ -11,6 +11,7 @@ import { recognizeEspImage } from "../../domain/firmware/espImage.js";
 import { hasUf2Block } from "../../domain/firmware/uf2.js";
 import { isDfuSeFile } from "../../domain/firmware/dfuse.js";
 import { hasFitImagesNode } from "../../domain/firmware/fitImage.js";
+import { hasMftFileRecordHeader } from "../../domain/windows/ntfsMft.js";
 import type { ArtifactClassificationEvidence } from "../ArtifactClassificationProbe.js";
 import { asarHeaderAdmissionFailure } from "../AsarHeader.js";
 import { hasDestinationPath } from "../ArtifactPaths.js";
@@ -622,6 +623,7 @@ export const classifyArtifactBytes = (
     return "dfuse";
   if (recognizeEspImage(prefix, fileSize) === "esp-image") return "esp-image";
   if (hasFitImagesNode(prefix)) return "fit";
+  if (hasMftFileRecordHeader(prefix)) return "ntfs-mft";
   if (hasBinaryPlistSignature(prefix)) return "plist";
   if (hasXmlPlistSignature(prefix)) return "plist";
   const fatFormat = classifyFatArtifact(evidence, fileSize);

src/browser/history/HistoricalCaptureBudgets.test.ts

@@ -89,7 +89,8 @@ it.each(["max_input_bytes", "max_reply_bytes", "max_nesting_depth"] as const)(
           parameter: control,
           effective_value: selected,
           source: "request",
-          rea_restart_required: false,
+          update_mode: "parameter",
+          pending_restart: false,
         },
       },
     });

src/config/captureDecoderHeap.test.ts

@@ -103,11 +103,10 @@ it("uses a valid REA_CAPTURE_DECODER_OLD_GENERATION_MIB override", async () => {
     /floor 8 MiB.*maximum 16384 MiB/iu,
   );
   expect(projected.remediation.action).toContain("update_analysis_settings");
-  expect(projected.remediation.action).toMatch(/restart REA/iu);
+  expect(projected.remediation.action).toContain("launch seed");
   expect(projected.remediation.action).toContain(
     "capture.decoder_old_generation_mib",
   );
-  expect(projected.remediation.action).toMatch(/without a REA restart/iu);
   expect(projected.remediation.action).toMatch(
     /removes inherited NODE_OPTIONS/iu,
   );
@@ -125,7 +124,8 @@ it("uses a valid REA_CAPTURE_DECODER_OLD_GENERATION_MIB override", async () => {
         unit: "MiB",
         applies_at: "next_worker",
         agent_editable: true,
-        rea_restart_required: false,
+        update_mode: "runtime",
+        pending_restart: false,
       },
       reported_limits: {
         old_generation_heap_mib: 256,

src/config/captureDecoderHeap.ts

@@ -71,7 +71,7 @@ export const resolveCaptureDecoderHeap = (
 export const captureDecoderHeapRecovery = (heap: CaptureDecoderHeap): string =>
   [
     `The effective ${CAPTURE_DECODER_OLD_GENERATION_SETTING} value is ${String(heap.oldGenerationMiB)} MiB (source ${heap.source}; default ${String(DEFAULT_CAPTURE_DECODER_OLD_GENERATION_MIB)} MiB, floor ${String(CAPTURE_DECODER_SEMI_SPACE_MIB)} MiB, maximum ${String(MAXIMUM_CAPTURE_DECODER_OLD_GENERATION_MIB)} MiB).`,
-    `Set ${CAPTURE_DECODER_OLD_GENERATION_SETTING} in the environment that starts REA to a decimal integer from ${String(CAPTURE_DECODER_SEMI_SPACE_MIB)} to ${String(MAXIMUM_CAPTURE_DECODER_OLD_GENERATION_MIB)}, then restart REA. In a running MCP server, use update_analysis_settings with setting capture.decoder_old_generation_mib and action set instead; the new value applies to new work without a REA restart.`,
+    `capture.decoder_old_generation_mib can be changed at runtime with update_analysis_settings. The launch seed is ${CAPTURE_DECODER_OLD_GENERATION_SETTING}, a decimal integer from ${String(CAPTURE_DECODER_SEMI_SPACE_MIB)} to ${String(MAXIMUM_CAPTURE_DECODER_OLD_GENERATION_MIB)}.`,
     `On Node.js the decoder passes --max-old-space-size=${String(heap.oldGenerationMiB)} and removes inherited NODE_OPTIONS, so additional host memory and NODE_OPTIONS do not raise this heap. The ${String(CAPTURE_DECODER_SEMI_SPACE_MIB)} MiB semi-space stays fixed.`,
     "On Bun the V8 flags are ignored. Windows limits the decoder's committed memory to this value plus a 512 MiB reserve. Other Bun hosts enforce no decoder heap limit.",
   ].join(" ");

src/config/debuggerSessionLimits.test.ts

@@ -171,7 +171,8 @@ it("names the session setting on a Rizin capacity refusal", async () => {
         minimum: MINIMUM_DEBUGGER_MAX_SESSIONS,
         maximum: MAXIMUM_DEBUGGER_MAX_SESSIONS,
         unit: "sessions",
-        restart_required: true,
+        update_mode: "restart",
+        pending_restart: false,
         max_sessions: 1,
         active_or_reserved_sessions: 1,
       },

src/config/environment.ts

@@ -6,7 +6,10 @@ import { z } from "zod";
 import { mcpInputBudgetSchema } from "./mcpInputBudget.js";
 import { isAbsolute } from "node:path";
 
-import { ConfigurationError } from "../domain/configurationErrors.js";
+import {
+  ConfigurationError,
+  type ConfigurationSettingIssue,
+} from "../domain/configurationErrors.js";
 import { err, ok, type Result } from "../domain/result.js";
 import { analysisProviderSelectorSchema } from "../domain/providerSelection.js";
 import {
@@ -165,6 +168,31 @@ const environmentSchema = z.object({
   REA_REFERENCE_SECRET_PATTERNS_JSON: z.string().default("[]"),
 });
 
+/** Copy producer-owned update facts from a Zod issue without inventing them. */
+const configurationSettingFromIssue = (issue: {
+  readonly path: readonly PropertyKey[];
+  readonly message: string;
+  readonly code: string;
+  readonly params?: Readonly<Record<string, unknown>> | undefined;
+}): ConfigurationSettingIssue => {
+  const setting = String(issue.path[0] ?? "environment");
+  const constraint = issue.message;
+  if (issue.code !== "custom" || issue.params === undefined)
+    return { setting, constraint };
+  const updateMode = issue.params.update_mode;
+  const pendingRestart = issue.params.pending_restart;
+  return {
+    setting,
+    constraint,
+    ...(updateMode === "runtime" || updateMode === "restart"
+      ? { update_mode: updateMode }
+      : {}),
+    ...(typeof pendingRestart === "boolean"
+      ? { pending_restart: pendingRestart }
+      : {}),
+  };
+};
+
 /** Parse the JEB-only timeout at its isolated provider-composition boundary. */
 export const parseJebMcpTimeoutMs = (
   value: string | undefined,
@@ -229,10 +257,10 @@ export const parseEnvironmentValues = (
   const parsed = environmentSchema.safeParse(literals);
   if (!parsed.success) {
     // Zod messages name the constraint, never the rejected value.
-    const settings = parsed.error.issues.map((issue) => ({
-      setting: String(issue.path[0] ?? "environment"),
-      constraint: issue.message,
-    }));
+    // A producer may attach update_mode and pending_restart on a custom issue.
+    const settings = parsed.error.issues.map((issue) =>
+      configurationSettingFromIssue(issue),
+    );
     return err(
       new ConfigurationError("Invalid REA environment configuration", {
         cause: parsed.error,

src/config/regexMatchingDeadline.test.ts

@@ -49,7 +49,8 @@ it("keeps the 5000 ms deadline when the setting is unset", () => {
         applies_at: "next_operation",
         agent_editable: true,
         scope: "rea-runtime",
-        rea_restart_required: false,
+        update_mode: "runtime",
+        pending_restart: false,
       },
       reported_limits: {
         matching_deadline_ms: DEFAULT_REGEX_MATCHING_DEADLINE.deadlineMs,
@@ -65,8 +66,7 @@ it("keeps the 5000 ms deadline when the setting is unset", () => {
   expect(projected.remediation.action).toContain("source default");
   expect(projected.remediation.action).toContain("update_analysis_settings");
   expect(projected.remediation.action).toContain("regex.matching_deadline_ms");
-  expect(projected.remediation.action).toMatch(/restart REA/i);
-  expect(projected.remediation.action).toMatch(/without a REA restart/i);
+  expect(projected.remediation.action).toContain("launch seed");
   expect(projected.remediation.action).toMatch(
     /heap does not raise this deadline/i,
   );

src/config/regexMatchingDeadline.ts

@@ -70,6 +70,6 @@ export const regexMatchingDeadlineRecovery = (
 ): string =>
   [
     `The effective ${REGEX_MATCHING_DEADLINE_SETTING} value is ${String(deadline.deadlineMs)} ms (source ${deadline.source}; default ${String(DEFAULT_REGEX_MATCHING_DEADLINE_MS)} ms, floor 1 ms, maximum ${String(MAXIMUM_REGEX_MATCHING_DEADLINE_MS)} ms).`,
-    `Set ${REGEX_MATCHING_DEADLINE_SETTING} in the environment that starts REA to a positive decimal integer from 1 to ${String(MAXIMUM_REGEX_MATCHING_DEADLINE_MS)}, then restart REA. In a running MCP server, use update_analysis_settings with setting regex.matching_deadline_ms and action set instead; the new value applies to new work without a REA restart.`,
+    `regex.matching_deadline_ms can be changed at runtime with update_analysis_settings. The launch seed is ${REGEX_MATCHING_DEADLINE_SETTING}, a positive decimal integer from 1 to ${String(MAXIMUM_REGEX_MATCHING_DEADLINE_MS)}.`,
     "The worker old-generation heap does not raise this deadline.",
   ].join(" ");

src/config/regexWorkerHeapRequest.test.ts

@@ -56,7 +56,8 @@ it("keeps the 128 MiB request when the setting is unset", () => {
         applies_at: "next_worker",
         agent_editable: true,
         scope: "rea-runtime",
-        rea_restart_required: false,
+        update_mode: "runtime",
+        pending_restart: false,
       },
       reported_limits: {
         requested_worker_heap_mib: DEFAULT_REGEX_WORKER_HEAP_REQUEST.heapMiB,
@@ -76,8 +77,7 @@ it("keeps the 128 MiB request when the setting is unset", () => {
   );
   expect(projected.remediation.action).toContain("--max-old-space-size");
   expect(projected.remediation.action).toContain("NODE_OPTIONS");
-  expect(projected.remediation.action).toMatch(/restart REA/i);
-  expect(projected.remediation.action).toMatch(/without a REA restart/i);
+  expect(projected.remediation.action).toContain("launch seed");
   expect(projected.remediation.action).toMatch(/overrides this request/i);
 });
 

src/config/regexWorkerHeapRequest.ts

@@ -72,6 +72,6 @@ export const regexWorkerHeapRequestRecovery = (
 ): string =>
   [
     `The effective ${REGEX_WORKER_HEAP_REQUEST_SETTING} value is ${String(request.heapMiB)} MiB (source ${request.source}; default ${String(DEFAULT_REGEX_WORKER_HEAP_REQUEST_MIB)} MiB, floor ${String(MINIMUM_REGEX_WORKER_HEAP_REQUEST_MIB)} MiB, maximum ${String(MAXIMUM_REGEX_WORKER_HEAP_REQUEST_MIB)} MiB).`,
-    `Set ${REGEX_WORKER_HEAP_REQUEST_SETTING} in the environment that starts REA to a decimal integer from ${String(MINIMUM_REGEX_WORKER_HEAP_REQUEST_MIB)} to ${String(MAXIMUM_REGEX_WORKER_HEAP_REQUEST_MIB)}, then restart REA. In a running MCP server, use update_analysis_settings with setting regex.worker_heap_request_mib and action set instead; the new value applies to new work without a REA restart.`,
+    `regex.worker_heap_request_mib can be changed at runtime with update_analysis_settings. The launch seed is ${REGEX_WORKER_HEAP_REQUEST_SETTING}, a decimal integer from ${String(MINIMUM_REGEX_WORKER_HEAP_REQUEST_MIB)} to ${String(MAXIMUM_REGEX_WORKER_HEAP_REQUEST_MIB)}.`,
     "A positive Node --max-old-space-size, including NODE_OPTIONS, overrides this request and sets the effective heap. That flag is not limited by this setting's floor or maximum. Raising host memory alone does not raise the request.",
   ].join(" ");

src/config/windowsChangeTimeSettle.test.ts

@@ -27,7 +27,7 @@ const requiredFacts = (
   expect(message).toContain(String(fact.defaultValue));
   expect(message).toContain(String(fact.floor));
   expect(message).toContain(String(fact.maximum));
-  expect(message).toContain("restart REA");
+  expect(message).toContain("positive decimal integer");
 };
 
 it("keeps today's settle defaults when the settings are unset", () => {
@@ -98,17 +98,21 @@ it.each([
   const message = resolved.error.settings[0]?.constraint;
   if (message === undefined) throw new Error("Missing configuration guidance");
   requiredFacts(message, { ...bound, currentValue: value, source: "setting" });
-  expect(resolved.error.settings).toEqual([
-    { setting: bound.setting, constraint: message },
-  ]);
+  const facts = {
+    setting: bound.setting,
+    constraint: message,
+    update_mode: "restart" as const,
+    pending_restart: true as const,
+  };
+  expect(resolved.error.settings).toEqual([facts]);
   expect(projectAnalysisError(resolved.error)).toMatchObject({
     code: "configuration_invalid",
-    details: { settings: [{ setting: bound.setting, constraint: message }] },
+    details: { settings: [facts] },
   });
   const startup = parseEnvironment({ [bound.setting]: value });
   if (startup.ok) throw new Error(`Startup accepted ${bound.setting}=${value}`);
-  expect(startup.error.settings).toContainEqual({
-    setting: bound.setting,
-    constraint: message,
+  expect(startup.error.settings).toContainEqual(facts);
+  expect(projectAnalysisError(startup.error).details).toMatchObject({
+    settings: [facts],
   });
 });

src/config/windowsChangeTimeSettle.ts

@@ -55,7 +55,9 @@ export interface WindowsChangeTimeSettingFailure {
   readonly floor: number;
   readonly maximum: number;
   readonly unit: "ms" | "polls";
-  readonly restartRequired: true;
+  readonly update_mode: "restart";
+  /** The rejected environment value is not the value this process is using. */
+  readonly pending_restart: true;
 }
 
 /** Effective settle budget and whether each part came from the default or a setting. */
@@ -79,10 +81,11 @@ const windowsChangeTimeSettingFailure = (
   floor: bound.floor,
   maximum: bound.maximum,
   unit: bound.unit,
-  restartRequired: true,
+  update_mode: "restart",
+  pending_restart: true,
 });
 
-/** Name the rejected value, its source, the allowed range, and the required restart. */
+/** Name the rejected value, its source, and the allowed range. */
 const windowsChangeTimeSettingMessage = (
   failure: WindowsChangeTimeSettingFailure,
 ): string =>
@@ -100,11 +103,14 @@ export const windowsChangeTimeSettingSchema = (bound: WindowsChangeTimeBound) =>
     .string()
     .superRefine((value, context) => {
       if (inRange(bound, value)) return;
+      const failure = windowsChangeTimeSettingFailure(bound, value);
       context.addIssue({
         code: "custom",
-        message: windowsChangeTimeSettingMessage(
-          windowsChangeTimeSettingFailure(bound, value),
-        ),
+        message: windowsChangeTimeSettingMessage(failure),
+        params: {
+          update_mode: failure.update_mode,
+          pending_restart: failure.pending_restart,
+        },
       });
     })
     .transform(Number);
@@ -160,6 +166,8 @@ export const resolveWindowsChangeTimeSettle = (
           settings: failures.map((failure) => ({
             setting: failure.setting,
             constraint: windowsChangeTimeSettingMessage(failure),
+            update_mode: failure.update_mode,
+            pending_restart: failure.pending_restart,
           })),
         },
       ),

src/contracts/analysisSettingsContracts.ts

@@ -48,6 +48,8 @@ const descriptor = z.strictObject({
   minimum: z.number(),
   maximum: z.number(),
   applies_at: z.enum(["next_operation", "next_worker", "next_startup"]),
+  update_mode: z.enum(["runtime", "restart"]),
+  pending_restart: z.boolean(),
   provider: z.string(),
   operations: z.array(z.string()),
   environment: z.string().optional(),
@@ -83,7 +85,7 @@ export const updateAnalysisSettingsResultSchema =
       }),
     ),
     active_work_changed: z.literal(false),
-    rea_restart_required: z.literal(false),
+    pending_restart: z.literal(false),
   });
 export const ANALYSIS_SETTINGS_TOOL_CONTRACTS = [
   {

src/domain/analysisErrorProjection.test.ts

@@ -1512,7 +1512,8 @@ describe("analysis error projection: caller lifetime", () => {
       minimum: limit.minimum,
       maximum: limit.maximum,
       agent_editable: true,
-      rea_restart_required: false,
+      update_mode: "runtime",
+      pending_restart: false,
       applies_at: "next_invocation",
       scope: "rea-invocation",
       environment: "REA_REGEX_MATCHING_DEADLINE_MS",
@@ -1529,11 +1530,25 @@ describe("analysis error projection: caller lifetime", () => {
       expect.objectContaining({
         setting: "wasm.command_timeout_ms",
         source: "request",
+        update_mode: "parameter",
         parameter: "command_timeout_ms",
         applies_at: "next_invocation",
         scope: "rea-invocation",
       }),
     );
+    expect(
+      analysisSettingLimit("wasm.command_timeout_ms", 100, "runtime"),
+    ).toMatchObject({ source: "runtime", update_mode: "runtime" });
+    expect(
+      analysisSettingLimit("javascript.max_heap_mb", 128, "request"),
+    ).toMatchObject({
+      source: "request",
+      update_mode: "parameter",
+      parameter: "max_heap_mb",
+    });
+    expect(
+      analysisSettingLimit("javascript.analysis_timeout_ms", 1, "default"),
+    ).toMatchObject({ source: "default", update_mode: "runtime" });
     expect(
       projectAnalysisError(error, "invocation").details?.limit,
     ).not.toHaveProperty("environment");

src/domain/analysisErrorProjection.ts

@@ -1044,9 +1044,15 @@ const lifecycleErrorDetails = (
   }
   if (error instanceof ConfigurationError && error.settings.length > 0)
     return {
-      settings: error.settings.map(({ setting, constraint }) => ({
-        setting,
-        constraint,
+      settings: error.settings.map((issue) => ({
+        setting: issue.setting,
+        constraint: issue.constraint,
+        ...(issue.update_mode === undefined
+          ? {}
+          : { update_mode: issue.update_mode }),
+        ...(issue.pending_restart === undefined
+          ? {}
+          : { pending_restart: issue.pending_restart }),
       })),
     };
   if (error instanceof BinaryTargetError)

src/domain/analysisSettingDefinitions.ts

@@ -46,6 +46,7 @@ interface SettingDefinition {
   readonly maximum: number;
   readonly unit: "bytes" | "ms" | "seconds" | "MiB" | "count";
   readonly appliesAt: "next_operation" | "next_worker" | "next_startup";
+  readonly updateMode: "runtime" | "restart";
   readonly provider: string;
   readonly relatedProviders?: readonly string[];
   readonly operations: readonly string[];
@@ -60,6 +61,7 @@ const timer = {
 /** Metadata aggregates existing owner definitions; no environment mutation or engine discovery. */
 export const ANALYSIS_SETTINGS = {
   "ghidra.decompile_max_payload_mib": {
+    updateMode: "runtime",
     environment: "REA_GHIDRA_DECOMPILE_MAX_PAYLOAD_MIB",
     defaultValue: null,
     minimum: 1,
@@ -77,6 +79,7 @@ export const ANALYSIS_SETTINGS = {
     ],
   },
   "asar.max_listing_bytes": {
+    updateMode: "runtime",
     environment: "REA_ASAR_MAX_LISTING_BYTES",
     defaultValue: DEFAULT_ASAR_MAX_LISTING_BYTES,
     minimum: 1,
@@ -100,6 +103,7 @@ export const ANALYSIS_SETTINGS = {
     ],
   },
   "ghidra.value_flow_max_operations": {
+    updateMode: "runtime",
     environment: "REA_GHIDRA_VALUE_FLOW_MAX_OPERATIONS",
     defaultValue: DEFAULT_GHIDRA_VALUE_FLOW_MAX_OPERATIONS,
     minimum: 1,
@@ -114,6 +118,7 @@ export const ANALYSIS_SETTINGS = {
     ],
   },
   "ghidra.value_flow_max_inputs_per_operation": {
+    updateMode: "runtime",
     environment: "REA_GHIDRA_VALUE_FLOW_MAX_INPUTS_PER_OPERATION",
     defaultValue: DEFAULT_GHIDRA_VALUE_FLOW_MAX_INPUTS_PER_OPERATION,
     minimum: 1,
@@ -128,6 +133,7 @@ export const ANALYSIS_SETTINGS = {
     ],
   },
   "ghidra.value_flow_max_edges": {
+    updateMode: "runtime",
     environment: "REA_GHIDRA_VALUE_FLOW_MAX_EDGES",
     defaultValue: DEFAULT_GHIDRA_VALUE_FLOW_MAX_EDGES,
     minimum: 1,
@@ -142,6 +148,7 @@ export const ANALYSIS_SETTINGS = {
     ],
   },
   "asar.max_header_bytes": {
+    updateMode: "runtime",
     environment: "REA_ASAR_MAX_HEADER_BYTES",
     defaultValue: DEFAULT_ASAR_MAX_HEADER_BYTES,
     minimum: 1,
@@ -165,6 +172,7 @@ export const ANALYSIS_SETTINGS = {
     ],
   },
   "ghidra.startup_timeout_ms": {
+    updateMode: "runtime",
     ...timer,
     environment: "REA_GHIDRA_STARTUP_TIMEOUT_MS",
     defaultValue: DEFAULT_GHIDRA_STARTUP_TIMEOUT_MS,
@@ -185,6 +193,7 @@ export const ANALYSIS_SETTINGS = {
     ],
   },
   "hopper.startup_timeout_ms": {
+    updateMode: "runtime",
     ...timer,
     environment: "REA_HOPPER_STARTUP_TIMEOUT_MS",
     defaultValue: DEFAULT_HOPPER_STARTUP_TIMEOUT_MS,
@@ -205,6 +214,7 @@ export const ANALYSIS_SETTINGS = {
     ],
   },
   "ghidra.decompile_timeout_seconds": {
+    updateMode: "runtime",
     environment: "REA_GHIDRA_DECOMPILE_TIMEOUT_SECONDS",
     defaultValue: null,
     minimum: 1,
@@ -220,6 +230,7 @@ export const ANALYSIS_SETTINGS = {
     ],
   },
   "jadx.operation_timeout_ms": {
+    updateMode: "runtime",
     ...timer,
     environment: "REA_JADX_OPERATION_TIMEOUT_MS",
     defaultValue: DEFAULT_JADX_OPERATION_TIMEOUT_MS,
@@ -233,6 +244,7 @@ export const ANALYSIS_SETTINGS = {
     ],
   },
   "regex.matching_deadline_ms": {
+    updateMode: "runtime",
     ...timer,
     environment: "REA_REGEX_MATCHING_DEADLINE_MS",
     defaultValue: DEFAULT_REGEX_MATCHING_DEADLINE_MS,
@@ -242,6 +254,7 @@ export const ANALYSIS_SETTINGS = {
     operations: ["search_strings", "search_procedures"],
   },
   "regex.worker_heap_request_mib": {
+    updateMode: "runtime",
     environment: "REA_REGEX_WORKER_HEAP_REQUEST_MIB",
     defaultValue: DEFAULT_REGEX_WORKER_HEAP_REQUEST_MIB,
     minimum: MINIMUM_REGEX_WORKER_HEAP_REQUEST_MIB,
@@ -253,6 +266,7 @@ export const ANALYSIS_SETTINGS = {
     operations: ["search_strings", "search_procedures"],
   },
   "capture.decoder_old_generation_mib": {
+    updateMode: "runtime",
     environment: "REA_CAPTURE_DECODER_OLD_GENERATION_MIB",
     defaultValue: DEFAULT_CAPTURE_DECODER_OLD_GENERATION_MIB,
     minimum: CAPTURE_DECODER_SEMI_SPACE_MIB,
@@ -263,6 +277,7 @@ export const ANALYSIS_SETTINGS = {
     operations: ["inspect_web_network_capture"],
   },
   "javascript.analysis_timeout_ms": {
+    updateMode: "runtime",
     ...timer,
     defaultValue: JAVASCRIPT_ANALYSIS_RESOURCE_DEFAULTS.timeoutMs,
     provider: "rea-javascript-application",
@@ -270,6 +285,7 @@ export const ANALYSIS_SETTINGS = {
     parameter: "analysis_timeout_ms",
   },
   "javascript.max_heap_mb": {
+    updateMode: "runtime",
     defaultValue: JAVASCRIPT_ANALYSIS_RESOURCE_DEFAULTS.heapMb,
     minimum: JAVASCRIPT_WORKER_HEAP_MIB_MINIMUM,
     maximum: JAVASCRIPT_WORKER_HEAP_MIB_MAXIMUM,
@@ -280,6 +296,7 @@ export const ANALYSIS_SETTINGS = {
     parameter: "max_heap_mb",
   },
   "wasm.command_timeout_ms": {
+    updateMode: "runtime",
     ...timer,
     defaultValue: DEFAULT_WASM_COMMAND_TIMEOUT_MS,
     maximum: MAX_WASM_COMMAND_TIMEOUT_MS,
@@ -288,6 +305,7 @@ export const ANALYSIS_SETTINGS = {
     parameter: "command_timeout_ms",
   },
   "native.matching_representation_bytes": {
+    updateMode: "runtime",
     environment: "REA_NATIVE_MATCHING_REPRESENTATION_BYTES",
     defaultValue: DEFAULT_NATIVE_MATCHING_REPRESENTATION_BYTES,
     minimum: 1,
@@ -317,6 +335,23 @@ export const analysisSettingValueSchema = (id: AnalysisSettingId) => {
     ? number.nullable()
     : number;
 };
+/**
+ * Caller argument, then the accepted runtime override, then the built-in
+ * default. The third value is that default, not a launch-environment seed.
+ */
+export const selectedAnalysisSetting = (
+  request: number | undefined,
+  override: number | undefined,
+  fallback: number,
+): {
+  readonly value: number;
+  readonly source: "default" | "runtime" | "request";
+} => {
+  if (request !== undefined) return { value: request, source: "request" };
+  if (override !== undefined) return { value: override, source: "runtime" };
+  return { value: fallback, source: "default" };
+};
+
 /** Project one owner's accepted budget, including the explicit per-call precedence. */
 export const analysisSettingLimit = (
   setting: AnalysisSettingId,
@@ -334,7 +369,8 @@ export const analysisSettingLimit = (
     applies_at: definition.appliesAt,
     agent_editable: true,
     scope: "rea-runtime",
-    rea_restart_required: false,
+    update_mode: source === "request" ? "parameter" : definition.updateMode,
+    pending_restart: false,
     ...(definition.parameter === undefined
       ? {}
       : { parameter: definition.parameter }),

src/domain/analysisSettings.ts

@@ -26,6 +26,14 @@ export interface AnalysisSettingsSnapshot {
 /** Injected by the runtime owner; reads the accepted operation's immutable snapshot. */
 export type AnalysisSettingsReader = () => AnalysisSettingsSnapshot;
 
+/**
+ * Whether `update_analysis_settings` can change the active value.
+ * `runtime` applies inside this process. `restart` stays at the value this
+ * process already started with. A limit fact uses `parameter` when this call
+ * supplied the value; the setting definition itself stays `runtime`.
+ */
+export type AnalysisSettingUpdateMode = "runtime" | "restart";
+
 /** Limit facts supplied by the failing owner, never inferred from error prose. */
 export interface AnalysisSettingLimit {
   readonly setting: AnalysisSettingId;
@@ -37,7 +45,12 @@ export interface AnalysisSettingLimit {
   readonly applies_at: "next_operation" | "next_worker" | "next_startup";
   readonly agent_editable: true;
   readonly scope: "rea-runtime";
-  readonly rea_restart_required: false;
+  readonly update_mode: AnalysisSettingUpdateMode | "parameter";
+  /**
+   * True only when a configured value is not the one this process is using.
+   * The reported effective value is active, so limit errors leave this false.
+   */
+  readonly pending_restart: false;
   readonly required_value?: number;
   readonly parameter?: string;
 }

src/domain/analysisSnapshot.ts

@@ -86,6 +86,7 @@ const formatSchema = z.enum([
   "dfuse",
   "esp-flash-image",
   "fit",
+  "ntfs-mft",
   "plist",
   "javascript",
   "source-map",

src/domain/artifactFormatMetadata.ts

@@ -1,4 +1,5 @@
 import { z } from "zod";
+import { MFT_INLINE_HEADER_LIMIT } from "./windows/ntfsMft.js";
 
 import { jsonRecordSchema } from "./jsonRecord.js";
 import { jsonValueSchema } from "./jsonValue.js";
@@ -299,6 +300,56 @@ export const hexRecordRegionMetadataSchema = z.strictObject({
   last_line: offset,
 });
 
+const ntfsFileReferenceSchema = z.strictObject({
+  recordNumber: offset,
+  sequenceNumber: offset,
+});
+const ntfsRecordHeaderSchema = z.strictObject({
+  index: offset,
+  fileOffset: offset,
+  sequenceNumber: offset,
+  hardLinkCount: offset,
+  flags: z.strictObject({
+    raw: offset,
+    names: z.array(z.string()),
+    unknownMask: offset,
+  }),
+  inUse: z.boolean(),
+  directory: z.boolean(),
+  usedBytes: offset,
+  totalBytes: offset,
+  extentValid: z.boolean(),
+  baseRecordReference: ntfsFileReferenceSchema.nullable(),
+  declaredRecordNumber: offset.nullable(),
+  recordNumberMatchesSlot: z.boolean().nullable(),
+  logFileSequenceNumber: z.string().regex(/^(?:0|[1-9][0-9]*)$/u),
+  firstAttributeOffset: offset,
+  nextAttributeId: offset,
+  fixups: z.strictObject({
+    sequenceNumber: offset,
+    entries: offset,
+    verifiedSectors: offset,
+    coveredSectors: offset,
+    mismatchSector: offset.nullable(),
+    malformed: z.boolean(),
+  }),
+});
+
+/** NTFS $MFT record-header facts; attributes, names and the USN journal are later steps. */
+export const ntfsMftMetadataSchema = z.strictObject({
+  kind: z.literal("ntfs-mft"),
+  record_bytes: offset,
+  record_count: offset,
+  in_use_record_count: offset,
+  directory_record_count: offset,
+  extension_record_count: offset,
+  fixup_mismatch_count: offset,
+  fixup_unverified_record_count: offset,
+  truncated_tail_bytes: offset,
+  record_headers: z.array(ntfsRecordHeaderSchema).max(MFT_INLINE_HEADER_LIMIT),
+  omitted_record_headers: offset,
+});
+
 /**
  * U-Boot FIT root facts. Configuration references are declared names; a
  * reference that names no image is reported as a limitation, not removed.
@@ -886,6 +937,7 @@ export const artifactFormatMetadataSchema = z.discriminatedUnion("kind", [
   espImageSignatureBlockMetadataSchema,
   uf2MetadataSchema,
   uf2RegionMetadataSchema,
+  ntfsMftMetadataSchema,
   dfuSeMetadataSchema,
   dfuSeElementMetadataSchema,
   espFlashImageMetadataSchema,
@@ -1036,6 +1088,10 @@ const declarationsByKind: {
     notCompared:
       "Its familyID and addresses are in its path, family_name is REA's label for the familyID, and its block counts and positions are layout; content comparison covers its bytes.",
   },
+  "ntfs-mft": {
+    notCompared:
+      "Its fields come from the record headers in the file's own FILE records, which content comparison covers.",
+  },
   dfuse: {
     notCompared:
       "Its fields come from the file's own prefix, target headers and suffix, which content comparison covers.",

src/domain/artifactGraph.ts

@@ -103,6 +103,7 @@ export const ARTIFACT_FORMAT_IDS = [
   "dfuse",
   "esp-flash-image",
   "fit",
+  "ntfs-mft",
   "unknown",
 ] as const;
 
@@ -124,6 +125,7 @@ const artifactContentFormatSchema = artifactFormatSchema.extract([
   "uf2",
   "dfuse",
   "fit",
+  "ntfs-mft",
   "mach-o-universal",
   "mach-o",
   "elf",

src/domain/artifactReaderLimits.ts

@@ -191,7 +191,8 @@ export interface ArtifactReaderLimitFact {
   readonly applies_at: "next_operation";
   readonly agent_editable: true;
   readonly can_increase: boolean;
-  readonly rea_restart_required: false;
+  readonly update_mode: "parameter";
+  readonly pending_restart: false;
 }
 
 /** Runtime representation ceilings differ from aggregate plaintext-work budgets. */

src/domain/binaryTargetTypes.ts

@@ -302,7 +302,8 @@ export type BinaryTarget =
           | "esp-image"
           | "uf2"
           | "dfuse"
-          | "fit";
+          | "fit"
+          | "ntfs-mft";
       })
   | (BinaryTargetIdentity &
       NonExecutableMetadata & {

src/domain/configurationErrors.ts

@@ -7,6 +7,10 @@ import {
 export interface ConfigurationSettingIssue {
   readonly setting: string;
   readonly constraint: string;
+  /** Whether `update_analysis_settings` can change this setting, when the producer owns that fact. */
+  readonly update_mode?: "runtime" | "restart";
+  /** True when the rejected value is configured but not active in this process. */
+  readonly pending_restart?: boolean;
 }
 
 /** Rejected settings retained for diagnostics; their values are not. */
@@ -24,9 +28,16 @@ export class ConfigurationError extends AnalysisError {
   constructor(message: string, options?: ConfigurationErrorOptions) {
     super(message, options);
     this.remediationAction = options?.remediationAction;
-    this.settings = (options?.settings ?? []).map(
-      ({ setting, constraint }) => ({ setting, constraint }),
-    );
+    this.settings = (options?.settings ?? []).map((issue) => ({
+      setting: issue.setting,
+      constraint: issue.constraint,
+      ...(issue.update_mode === undefined
+        ? {}
+        : { update_mode: issue.update_mode }),
+      ...(issue.pending_restart === undefined
+        ? {}
+        : { pending_restart: issue.pending_restart }),
+    }));
   }
 }
 

src/domain/evidenceSubject.ts

@@ -52,6 +52,7 @@ export const evidenceSubjectFormatSchema = z.enum([
   "dfuse",
   "esp-flash-image",
   "fit",
+  "ntfs-mft",
 ]);
 
 /** Format-owned architecture declarations shared by Evidence and persisted executions. */

src/domain/javascript/childProcessOptionsRegression.test.ts

@@ -137,7 +137,8 @@ describe("child-process options detection", () => {
       ["negative-zero", `spawn("tool", [], { env: -0 })`, false],
       ["bigint-zero", `spawn("tool", [], { env: 0n })`, false],
       ["zero-float", `spawn("tool", [], { env: 0.0 })`, false],
-      ["nan", `spawn("tool", [], { env: NaN })`, true],
+      ["nan", `spawn("tool", [], { env: NaN })`, false],
+      ["shadowed-nan", `((NaN) => spawn("tool", [], { env: NaN }))({})`, true],
       ["empty-string", `spawn("tool", [], { env: "" })`, false],
       ["false", `spawn("tool", [], { env: false })`, false],
       ["empty-object", `spawn("tool", [], { env: {} })`, true],

src/domain/javascript/javascriptSemanticChildProcesses.ts

@@ -81,7 +81,7 @@ export const collectJavaScriptSemanticChildProcesses = (
       !rejectsOptionsShape(candidate, context, state),
   );
   const childProcessSpawns = candidates.map((candidate) =>
-    immutableSpawn(candidate, context),
+    immutableSpawn(candidate, context, state),
   );
   return {
     childProcessSpawns,
@@ -146,6 +146,7 @@ const collectSpawns = (
 const immutableSpawn = (
   candidate: SpawnCandidate,
   context: ChildSpawnContext,
+  state: JavaScriptSemanticAnalysisState,
 ): JavaScriptSemanticChildProcessSpawn => {
   const { options, argvCount } = spawnArguments(candidate, context);
   const command = dataEffectLiteralString(candidate.node.arguments[0]);
@@ -161,7 +162,7 @@ const immutableSpawn = (
     projectedResultBindings: candidate.projectedResultBindings,
     command,
     argvCount,
-    environmentSupplied: environmentSupplied(options, context),
+    environmentSupplied: environmentSupplied(options, context, state),
     stdioMode: stdioMode(options, candidate.method, context),
     resolution: command === null ? "partial" : "complete",
   };
@@ -443,11 +444,17 @@ const spawnArguments = (
 const environmentSupplied = (
   options: t.Node | undefined,
   context: ChildSpawnContext,
+  state: JavaScriptSemanticAnalysisState,
 ): boolean => {
   const lookup = staticObjectProperty(options, "env", context.isNullish);
   if (lookup.status !== "explicit") return false;
   const value = unwrapJavaScriptExpression(lookup.value).node;
   if (context.isNullish(value)) return false;
+  if (
+    t.isIdentifier(value, { name: "NaN" }) &&
+    isUnshadowedGlobal(value, state, "NaN")
+  )
+    return false;
   return staticJavaScriptTruthiness(value) !== false;
 };
 

src/domain/javascript/javascriptSourceAnalysis.ts

@@ -41,7 +41,21 @@ export interface JavaScriptSourceAnalysisFailure {
 export interface JavaScriptSourceAnalysisLimits {
   readonly semanticPrecision?: JavaScriptSemanticPrecisionInput;
   readonly heapMb: number;
+  /** Where `heapMb` was selected. Absent when the caller did not record it. */
+  readonly heapSource?:
+    | "default"
+    | "setting"
+    | "runtime"
+    | "request"
+    | undefined;
   readonly timeoutMs: number;
+  /** Where `timeoutMs` was selected. Absent when the caller did not record it. */
+  readonly timeoutSource?:
+    | "default"
+    | "setting"
+    | "runtime"
+    | "request"
+    | undefined;
 }
 
 /** Sequential source-analysis authority with independently verified cleanup. */

src/domain/liveBinaryTargetSummary.ts

@@ -83,6 +83,7 @@ export const liveBinaryTargetSummaryVariants = [
       "uf2",
       "dfuse",
       "fit",
+      "ntfs-mft",
     ]),
     architecture: z.null(),
   }),

src/domain/numericLimitRegistry.test.ts

@@ -0,0 +1,37 @@
+import { readFileSync } from "node:fs";
+import { join } from "node:path";
+import { expect, it } from "vitest";
+import { ANALYSIS_SETTINGS } from "./analysisSettingDefinitions.js";
+import { NUMERIC_LIMIT_REGISTRY } from "./numericLimitRegistry.js";
+
+const inventory = readFileSync(
+  join(import.meta.dirname, "../../docs/design/rea-limit-inventory.md"),
+  "utf8",
+);
+
+it("gives every numeric exception a reason and a matching inventory row", () => {
+  const names = NUMERIC_LIMIT_REGISTRY.map((entry) => entry.name);
+  expect(new Set(names).size).toBe(names.length);
+  const runtimeEnvironments = new Set<string>(
+    Object.values(ANALYSIS_SETTINGS).flatMap((definition) =>
+      "environment" in definition && definition.environment !== undefined
+        ? [definition.environment]
+        : [],
+    ),
+  );
+  for (const entry of NUMERIC_LIMIT_REGISTRY) {
+    expect(entry.update_mode).toBe("restart");
+    expect(entry.reason.trim().length).toBeGreaterThanOrEqual(40);
+    expect(entry.read).toBe(
+      entry.policy === "per-call" ? "per-call" : "startup",
+    );
+    expect(runtimeEnvironments.has(entry.name)).toBe(false);
+    expect(inventory).toContain(`\`${entry.name}\``);
+  }
+  expect(
+    NUMERIC_LIMIT_REGISTRY.filter((entry) => entry.policy === "restart-only"),
+  ).toHaveLength(15);
+  expect(
+    NUMERIC_LIMIT_REGISTRY.filter((entry) => entry.policy === "per-call"),
+  ).toHaveLength(11);
+});

src/domain/numericLimitRegistry.ts

@@ -0,0 +1,230 @@
+/**
+ * Numeric `REA_` limits that are not runtime analysis settings.
+ *
+ * The static check in `scripts/verify/rea-numeric-limits.mjs` fails when a
+ * numeric `REA_` limit in `src` is absent from both this list and
+ * `ANALYSIS_SETTINGS`. `restart-only` is the durable exception. `per-call`
+ * names a limit still read from the launch environment on each use; the
+ * runtime migration deletes that entry instead of leaving it here.
+ *
+ * Narrative bounds, owners, and the non-numeric settings live in
+ * `docs/design/rea-limit-inventory.md`.
+ */
+export interface NumericLimitRegistryEntry {
+  readonly name: string;
+  readonly update_mode: "restart";
+  readonly policy: "restart-only" | "per-call";
+  readonly read: "startup" | "per-call";
+  readonly reason: string;
+}
+
+export const NUMERIC_LIMIT_REGISTRY = [
+  {
+    name: "REA_BRIDGE_MAX_REQUEST_BYTES",
+    update_mode: "restart",
+    policy: "per-call",
+    read: "per-call",
+    reason:
+      "The Hopper, Ghidra, and Cutter clients measure each request against this ceiling. The value is still taken from the launch environment, so a running process does not see a new value until the runtime migration removes this entry.",
+  },
+  {
+    name: "REA_DEBUGGER_COMMAND_DEADLINE_MS",
+    update_mode: "restart",
+    policy: "per-call",
+    read: "per-call",
+    reason:
+      "GDB and Rizin apply this deadline to each debugger command. The session manager currently copies it at construction from the launch environment; the runtime migration reads the accepted operation snapshot instead.",
+  },
+  {
+    name: "REA_DEBUGGER_MAX_COMMAND_OUTPUT_BYTES",
+    update_mode: "restart",
+    policy: "per-call",
+    read: "per-call",
+    reason:
+      "Each debugger command retains output up to this ceiling. The session manager currently copies it at construction from the launch environment; the runtime migration reads the accepted operation snapshot instead.",
+  },
+  {
+    name: "REA_DEBUGGER_MAX_PENDING_COMMANDS",
+    update_mode: "restart",
+    policy: "restart-only",
+    read: "startup",
+    reason:
+      "Caps the pending-command queue created with the debugger session manager. It is pool capacity for that manager, not a budget checked independently on each command.",
+  },
+  {
+    name: "REA_DEBUGGER_MAX_SESSIONS",
+    update_mode: "restart",
+    policy: "restart-only",
+    read: "startup",
+    reason:
+      "Caps how many debugger sessions the session manager created at startup will admit. Changing the pool size of a manager that already exists requires a new REA process.",
+  },
+  {
+    name: "REA_HOPPER_CALLER_SCAN_BUDGET_MS",
+    update_mode: "restart",
+    policy: "per-call",
+    read: "per-call",
+    reason:
+      "Forward caller reconciliation reads this budget for each scan. It is still resolved from the launch environment; the runtime migration reads the accepted operation snapshot.",
+  },
+  {
+    name: "REA_HOPPER_DEMO_DIALOG_TIMEOUT_MS",
+    update_mode: "restart",
+    policy: "restart-only",
+    read: "startup",
+    reason:
+      "The Linux private-display adapter reads this watch from the process environment when that helper starts. A running adapter does not observe a value stored only in the REA process.",
+  },
+  {
+    name: "REA_JADX_ACTIVE_PROCESSOR_COUNT",
+    update_mode: "restart",
+    policy: "restart-only",
+    read: "startup",
+    reason:
+      "Passed to the JVM as -XX:ActiveProcessorCount when the JADX worker process starts. A running JVM does not adopt a new processor count.",
+  },
+  {
+    name: "REA_JADX_HEAP_MIB",
+    update_mode: "restart",
+    policy: "restart-only",
+    read: "startup",
+    reason:
+      "Passed to the JVM as -Xmx when the JADX worker process starts. A running JVM does not adopt a new heap size.",
+  },
+  {
+    name: "REA_JEB_MCP_TIMEOUT_MS",
+    update_mode: "restart",
+    policy: "per-call",
+    read: "per-call",
+    reason:
+      "Bounds each JEB tools/call and catalog discovery. The connection currently captures the launch value when it is created; the runtime migration reads the accepted snapshot for each request.",
+  },
+  {
+    name: "REA_KUNA_CACHE_MAX_BYTES",
+    update_mode: "restart",
+    policy: "restart-only",
+    read: "startup",
+    reason:
+      "Sizes the Kuna project cache when that cache is constructed for the provider. An existing cache does not resize from a later value.",
+  },
+  {
+    name: "REA_KUNA_JOBS",
+    update_mode: "restart",
+    policy: "restart-only",
+    read: "startup",
+    reason:
+      "Selects the Kuna worker-pool width when the provider builds its run slots. The pool is not resized for a process that is already running.",
+  },
+  {
+    name: "REA_KUNA_PROJECT_TIMEOUT_MS",
+    update_mode: "restart",
+    policy: "per-call",
+    read: "per-call",
+    reason:
+      "Bounds a Kuna project export and the whole-program scans that use that deadline. The configuration object still comes from the launch environment; the runtime migration reads the accepted snapshot at each use.",
+  },
+  {
+    name: "REA_KUNA_TIMEOUT_MS",
+    update_mode: "restart",
+    policy: "per-call",
+    read: "per-call",
+    reason:
+      "Bounds one Kuna function decompile. The configuration object still comes from the launch environment; the runtime migration reads the accepted snapshot at each use.",
+  },
+  {
+    name: "REA_MCP_DOCTOR_DEADLINE_MS",
+    update_mode: "restart",
+    policy: "restart-only",
+    read: "startup",
+    reason:
+      "Bounds the MCP doctor phase while the server is starting. It shapes server startup, not a later analysis call.",
+  },
+  {
+    name: "REA_MCP_FIRST_CATALOG_BUDGET_MS",
+    update_mode: "restart",
+    policy: "restart-only",
+    read: "startup",
+    reason:
+      "Bounds the first MCP catalog phase while the server is starting. It shapes server startup, not a later analysis call.",
+  },
+  {
+    name: "REA_MCP_INITIALIZE_BUDGET_MS",
+    update_mode: "restart",
+    policy: "restart-only",
+    read: "startup",
+    reason:
+      "Bounds the MCP initialize phase while the server is starting. It shapes server startup, not a later analysis call.",
+  },
+  {
+    name: "REA_MCP_MAX_INPUT_ELEMENTS",
+    update_mode: "restart",
+    policy: "restart-only",
+    read: "startup",
+    reason:
+      "Sizes argument admission for the MCP server created at process start. A running server does not rebuild that parser for a new allowance.",
+  },
+  {
+    name: "REA_MCP_MAX_RESPONSE_BYTES",
+    update_mode: "restart",
+    policy: "restart-only",
+    read: "startup",
+    reason:
+      "Sizes the MCP stdio receive buffer when the server process starts. A running transport and the paired client buffer do not adopt a new ceiling.",
+  },
+  {
+    name: "REA_MCP_TOOLS_LIST_BUDGET_MS",
+    update_mode: "restart",
+    policy: "restart-only",
+    read: "startup",
+    reason:
+      "Bounds the MCP tools/list phase while the server is starting. It shapes server startup, not a later analysis call.",
+  },
+  {
+    name: "REA_PWNTOOLS_MAX_INPUT_BYTES",
+    update_mode: "restart",
+    policy: "per-call",
+    read: "per-call",
+    reason:
+      "Default artifact-size admission for a pwntools decode. A per-call producer_limits.input_bytes override already exists; the launch default is still read from the environment and the runtime migration moves that default into the snapshot.",
+  },
+  {
+    name: "REA_RECONNECT_TIMEOUT_SECONDS",
+    update_mode: "restart",
+    policy: "restart-only",
+    read: "startup",
+    reason:
+      "Fixed 30-second value written into the Hopper bridge helper from HOPPER_SHUTDOWN_TIMEOUT_MS. It is not a caller-selected launch setting.",
+  },
+  {
+    name: "REA_SWIFTC_COMPILE_TIMEOUT_MS",
+    update_mode: "restart",
+    policy: "restart-only",
+    read: "startup",
+    reason:
+      "Captured when the Darwin ownership-helper reader is constructed at process start. That helper compile uses the captured deadline, so a later value needs a new REA process.",
+  },
+  {
+    name: "REA_WINDOWS_CHANGE_TIME_DRIFT_MS",
+    update_mode: "restart",
+    policy: "per-call",
+    read: "per-call",
+    reason:
+      "Each Windows ChangeTime observation compares the file against this drift window. The observation still parses the launch environment; the runtime migration reads the accepted operation snapshot.",
+  },
+  {
+    name: "REA_WINDOWS_CHANGE_TIME_SETTLE_MAX_POLLS",
+    update_mode: "restart",
+    policy: "per-call",
+    read: "per-call",
+    reason:
+      "Each Windows ChangeTime settle loop stops after this many polls. The observation still parses the launch environment; the runtime migration reads the accepted operation snapshot.",
+  },
+  {
+    name: "REA_WINDOWS_CHANGE_TIME_SETTLE_POLL_MS",
+    update_mode: "restart",
+    policy: "per-call",
+    read: "per-call",
+    reason:
+      "Each Windows ChangeTime settle loop waits this long between reads. The observation still parses the launch environment; the runtime migration reads the accepted operation snapshot.",
+  },
+] as const satisfies readonly NumericLimitRegistryEntry[];

src/domain/webNetworkCapture.ts

@@ -79,7 +79,8 @@ export const webNetworkCaptureBudgetFacts = (
         ? MAX_WEB_NETWORK_CAPTURE_NESTING_DEPTH
         : Number.MAX_SAFE_INTEGER,
     applies_at: "next_operation",
-    rea_restart_required: false,
+    update_mode: "parameter" as const,
+    pending_restart: false as const,
   };
 };
 

src/domain/windows/ntfsMft.test.ts

@@ -0,0 +1,478 @@
+import { describe, expect, it } from "vitest";
+
+import {
+  MFT_DEFAULT_RECORD_BYTES,
+  MFT_MAX_FILE_BYTES,
+  createMftDecoder,
+  parseMftImage,
+} from "./ntfsMft.js";
+
+interface RecordOptions {
+  readonly sequenceNumber?: number;
+  readonly hardLinkCount?: number;
+  readonly flags?: number;
+  readonly baseRecordReference?: bigint;
+  readonly recordBytes?: number;
+  /** Write a wrong trailer value into this sector, leaving the others correct. */
+  readonly corruptSector?: number;
+  /** Truncate the update sequence array to this many entries; 0 omits it. */
+  readonly usaEntries?: number;
+  readonly usaOffset?: number;
+}
+
+/** One FILE record with a zeroed body and fixups applied the way NTFS writes them. */
+const recordOf = (options: RecordOptions = {}): Uint8Array => {
+  const recordBytes = options.recordBytes ?? MFT_DEFAULT_RECORD_BYTES;
+  const record = new Uint8Array(recordBytes);
+  const view = new DataView(record.buffer);
+  record.set(Buffer.from("FILE", "latin1"), 0);
+  const sequenceNumber = options.sequenceNumber ?? 1;
+  const sectors = recordBytes / 512;
+
+  const usaEntries = options.usaEntries ?? sectors + 1;
+  if (usaEntries > 0) {
+    const usaOffset = options.usaOffset ?? 48;
+    view.setUint16(4, usaOffset, true);
+    view.setUint16(6, usaEntries, true);
+    view.setUint16(usaOffset, sequenceNumber, true);
+    for (let sector = 0; sector < sectors; sector += 1) {
+      const trailer = sector * 512 + 510;
+      // A saved trailer is what the sector held before the fixup replaced it.
+      view.setUint16(usaOffset + 2 * (sector + 1), 0xa5a5, true);
+      view.setUint16(
+        trailer,
+        sector >= usaEntries - 1
+          ? 0xa5a5
+          : sector === options.corruptSector
+            ? 0x1234
+            : sequenceNumber,
+        true,
+      );
+    }
+  } else {
+    view.setUint16(4, 0, true);
+    view.setUint16(6, 0, true);
+  }
+
+  view.setUint16(16, sequenceNumber, true);
+  view.setUint16(18, options.hardLinkCount ?? 1, true);
+  const firstAttributeOffset =
+    Math.ceil((48 + Math.max(0, usaEntries) * 2) / 8) * 8;
+  view.setUint16(20, firstAttributeOffset, true);
+  view.setUint16(22, options.flags ?? 0x0001, true);
+  view.setUint32(24, firstAttributeOffset + 8, true);
+  view.setUint32(firstAttributeOffset, 0xffff_ffff, true);
+  view.setUint32(28, recordBytes, true);
+  view.setBigUint64(32, options.baseRecordReference ?? 0n, true);
+  view.setUint16(40, 0, true);
+  return record;
+};
+
+const streamOf = (...records: readonly Uint8Array[]): Uint8Array => {
+  const total = records.reduce((sum, record) => sum + record.length, 0);
+  const stream = new Uint8Array(total);
+  let offset = 0;
+  for (const record of records) {
+    stream.set(record, offset);
+    const view = new DataView(stream.buffer, offset, record.length);
+    if (view.getUint16(4, true) >= 48)
+      view.setUint32(44, offset / record.length, true);
+    offset += record.length;
+  }
+  return stream;
+};
+
+describe("MFT header facts", () => {
+  it("inventories a clean record's header and verified fixups", () => {
+    const parsed = parseMftImage(streamOf(recordOf({ sequenceNumber: 5 })));
+    expect(parsed.ok).toBe(true);
+    if (!parsed.ok) return;
+    expect(parsed.value.metadata).toMatchObject({
+      kind: "ntfs-mft",
+      record_bytes: MFT_DEFAULT_RECORD_BYTES,
+      record_count: 1,
+      in_use_record_count: 1,
+      directory_record_count: 0,
+      extension_record_count: 0,
+      fixup_mismatch_count: 0,
+      fixup_unverified_record_count: 0,
+      truncated_tail_bytes: 0,
+    });
+    expect(parsed.value.records[0]).toMatchObject({
+      index: 0,
+      fileOffset: 0,
+      sequenceNumber: 5,
+      inUse: true,
+      directory: false,
+      baseRecordReference: null,
+      firstAttributeOffset: 56,
+      fixups: {
+        sequenceNumber: 5,
+        verifiedSectors: 2,
+        coveredSectors: 2,
+        mismatchSector: null,
+        malformed: false,
+      },
+    });
+    expect(parsed.value.limitations).toEqual([]);
+  });
+
+  it("identifies unrecognized flags without losing their source bits", () => {
+    const parsed = parseMftImage(recordOf({ flags: 0x8001 }));
+    if (!parsed.ok) throw new Error(parsed.error.message);
+    expect(parsed.value.metadata.record_headers[0]?.flags).toEqual({
+      raw: 0x8001,
+      names: ["in-use"],
+      unknownMask: 0x8000,
+    });
+  });
+
+  it.each([0, 49, 64, 72])(
+    "reports an invalid first attribute offset %i",
+    (offset) => {
+      const record = recordOf();
+      new DataView(record.buffer).setUint16(20, offset, true);
+      const parsed = parseMftImage(record);
+      if (!parsed.ok) throw new Error(parsed.error.message);
+      expect(parsed.value.records[0]?.extentValid).toBe(false);
+      expect(parsed.value.limitations.join(" ")).toContain("inconsistent");
+    },
+  );
+
+  it("reports a declared record number that contradicts its stream position", () => {
+    const bytes = streamOf(recordOf(), recordOf());
+    new DataView(bytes.buffer).setUint32(1024 + 44, 0, true);
+    const parsed = parseMftImage(bytes);
+    if (!parsed.ok) throw new Error(parsed.error.message);
+    expect(parsed.value.records[1]).toMatchObject({
+      index: 1,
+      declaredRecordNumber: 0,
+      recordNumberMatchesSlot: false,
+    });
+    expect(parsed.value.limitations.join(" ")).toContain(
+      "disagrees with their stream slot",
+    );
+  });
+
+  it("keeps the inline header window separate from the parsed record count", () => {
+    const parsed = parseMftImage(
+      streamOf(...Array.from({ length: 33 }, () => recordOf())),
+    );
+    if (!parsed.ok) throw new Error(parsed.error.message);
+    expect(parsed.value.metadata).toMatchObject({
+      record_count: 33,
+      omitted_record_headers: 1,
+    });
+    expect(parsed.value.metadata.record_headers).toHaveLength(32);
+    expect(parsed.value.records).toHaveLength(33);
+    expect(parsed.value.limitations).toEqual([]);
+  });
+
+  it("preserves a log sequence number above the safe integer range", () => {
+    const record = recordOf();
+    new DataView(record.buffer).setBigUint64(8, 9007199254740993n, true);
+    const parsed = parseMftImage(record);
+    if (!parsed.ok) throw new Error(parsed.error.message);
+    expect(parsed.value.records[0]?.logFileSequenceNumber).toBe(
+      "9007199254740993",
+    );
+  });
+
+  it("verifies the older FILE header without reading its fixup words as a record number", () => {
+    const oldRecord = recordOf({ usaOffset: 42 });
+    const newRecord = recordOf();
+    new DataView(newRecord.buffer).setUint32(44, 1, true);
+
+    const parsed = parseMftImage(streamOf(oldRecord, newRecord));
+    if (!parsed.ok) throw new Error(parsed.error.message);
+    expect(parsed.value.records).toMatchObject([
+      {
+        declaredRecordNumber: null,
+        fixups: { malformed: false, verifiedSectors: 2 },
+      },
+      {
+        declaredRecordNumber: 1,
+        fixups: { malformed: false, verifiedSectors: 2 },
+      },
+    ]);
+    expect(parsed.value.limitations).toEqual([]);
+  });
+});
+
+describe("MFT fixup coverage", () => {
+  it.each([0, 0xffff])(
+    "leaves reserved USA sequence %i unverified",
+    (sequenceNumber) => {
+      const parsed = parseMftImage(recordOf({ sequenceNumber }));
+      if (!parsed.ok) throw new Error(parsed.error.message);
+      expect(parsed.value.records[0]?.fixups).toMatchObject({
+        sequenceNumber,
+        verifiedSectors: 0,
+        malformed: true,
+      });
+      expect(parsed.value.metadata.fixup_unverified_record_count).toBe(1);
+      expect(parsed.value.limitations.join(" ")).toContain(
+        "invalid or incomplete",
+      );
+    },
+  );
+
+  it("keeps a record whose sector trailer is not the sequence number, and reports it", () => {
+    const parsed = parseMftImage(
+      streamOf(recordOf(), recordOf({ sequenceNumber: 9, corruptSector: 1 })),
+    );
+    expect(parsed.ok).toBe(true);
+    if (!parsed.ok) return;
+    expect(parsed.value.records[1]?.fixups).toMatchObject({
+      sequenceNumber: 9,
+      verifiedSectors: 1,
+      mismatchSector: 1,
+    });
+    expect(parsed.value.metadata.fixup_mismatch_count).toBe(1);
+    expect(parsed.value.limitations.join(" ")).toMatch(
+      /1 record\(s\) hold a sector trailer/u,
+    );
+  });
+
+  it("treats a missing update sequence array as unverified rather than clean", () => {
+    const parsed = parseMftImage(streamOf(recordOf({ usaEntries: 0 })));
+    expect(parsed.ok).toBe(true);
+    if (!parsed.ok) return;
+    expect(parsed.value.records[0]?.fixups).toMatchObject({
+      malformed: true,
+      mismatchSector: null,
+      coveredSectors: 0,
+    });
+    expect(parsed.value.metadata.fixup_mismatch_count).toBe(0);
+    expect(parsed.value.metadata.fixup_unverified_record_count).toBe(1);
+    expect(parsed.value.limitations.join(" ")).toMatch(
+      /invalid or incomplete update sequence array/u,
+    );
+  });
+
+  it.each([
+    [48, 4],
+    [49, 3],
+    [508, 3],
+  ])(
+    "rejects a malformed update sequence array at %i with %i entries",
+    (offset, count) => {
+      const bytes = recordOf();
+      const view = new DataView(bytes.buffer);
+      view.setUint16(4, offset, true);
+      view.setUint16(6, count, true);
+
+      const parsed = parseMftImage(bytes);
+      if (!parsed.ok) throw new Error(parsed.error.message);
+      expect(parsed.value.records[0]?.fixups).toMatchObject({
+        malformed: true,
+        verifiedSectors: 0,
+      });
+      expect(parsed.value.metadata.fixup_unverified_record_count).toBe(1);
+    },
+  );
+
+  it("counts uncovered sectors independently of a mismatched trailer", () => {
+    const parsed = parseMftImage(
+      recordOf({ recordBytes: 4096, usaEntries: 3, corruptSector: 0 }),
+    );
+    if (!parsed.ok) throw new Error(parsed.error.message);
+    expect(parsed.value.metadata).toMatchObject({
+      fixup_mismatch_count: 1,
+      fixup_unverified_record_count: 1,
+    });
+    expect(parsed.value.limitations.join(" ")).toContain(
+      "invalid or incomplete",
+    );
+  });
+
+  it("reports an array that covers fewer sectors than the record has", () => {
+    // A 4 KiB record holds eight sectors; a three-entry array leaves five unverified.
+    const parsed = parseMftImage(
+      streamOf(recordOf({ recordBytes: 4096, usaEntries: 3 })),
+    );
+    expect(parsed.ok).toBe(true);
+    if (!parsed.ok) return;
+    expect(parsed.value.records[0]?.fixups).toMatchObject({
+      coveredSectors: 2,
+      verifiedSectors: 2,
+      mismatchSector: null,
+      malformed: false,
+    });
+    expect(parsed.value.metadata.fixup_unverified_record_count).toBe(1);
+  });
+});
+
+describe("MFT record slots", () => {
+  it("distinguishes extension records from base records", () => {
+    const parsed = parseMftImage(
+      streamOf(
+        recordOf(),
+        recordOf({
+          flags: 0x0005,
+          baseRecordReference: 0xabcd_fedc_ba98_7654n,
+        }),
+      ),
+    );
+    expect(parsed.ok).toBe(true);
+    if (!parsed.ok) return;
+    expect(parsed.value.records[1]).toMatchObject({
+      baseRecordReference: {
+        recordNumber: 0xfedc_ba98_7654,
+        sequenceNumber: 0xabcd,
+      },
+      flags: { raw: 0x0005, names: ["in-use", "system"] },
+    });
+    expect(parsed.value.metadata.extension_record_count).toBe(1);
+  });
+
+  it("names directory records without losing the in-use flag", () => {
+    const parsed = parseMftImage(streamOf(recordOf({ flags: 0x0003 })));
+    expect(parsed.ok).toBe(true);
+    if (!parsed.ok) return;
+    expect(parsed.value.records[0]).toMatchObject({
+      directory: true,
+      inUse: true,
+      flags: { names: ["in-use", "directory"] },
+    });
+    expect(parsed.value.metadata.directory_record_count).toBe(1);
+  });
+
+  it("skips malformed and sparse slots without renumbering valid records", () => {
+    const stream = streamOf(
+      recordOf(),
+      recordOf(),
+      new Uint8Array(MFT_DEFAULT_RECORD_BYTES),
+      recordOf(),
+    );
+    stream.set(Buffer.from("BAAD", "latin1"), MFT_DEFAULT_RECORD_BYTES);
+    const parsed = parseMftImage(stream);
+    expect(parsed.ok).toBe(true);
+    if (!parsed.ok) return;
+    expect(parsed.value.metadata.record_count).toBe(2);
+    expect(
+      parsed.value.records.map(({ index, fileOffset }) => [index, fileOffset]),
+    ).toEqual([
+      [0, 0],
+      [3, 3 * MFT_DEFAULT_RECORD_BYTES],
+    ]);
+    expect(parsed.value.limitations).toEqual([
+      "record at 1024 does not begin with the FILE signature; 2 slot(s) without a FILE signature were found",
+    ]);
+  });
+
+  it("keeps a trailing partial record out of the inventory", () => {
+    const stream = new Uint8Array(1024 + 300);
+    stream.set(recordOf(), 0);
+    const parsed = parseMftImage(stream);
+    expect(parsed.ok).toBe(true);
+    if (!parsed.ok) return;
+    expect(parsed.value.metadata.record_count).toBe(1);
+    expect(parsed.value.metadata.truncated_tail_bytes).toBe(300);
+    expect(parsed.value.limitations.join(" ")).toMatch(
+      /300 trailing byte\(s\)/u,
+    );
+  });
+
+  it("falls back to the default record size when record 0 declares none", () => {
+    const record = recordOf();
+    new DataView(record.buffer).setUint32(28, 0, true);
+    const parsed = parseMftImage(record);
+    expect(parsed.ok).toBe(true);
+    if (!parsed.ok) return;
+    expect(parsed.value.recordBytes).toBe(MFT_DEFAULT_RECORD_BYTES);
+    expect(parsed.value.limitations.join(" ")).toMatch(
+      /not a supported power of two/u,
+    );
+  });
+
+  it("reads a 4 KiB record at its declared size", () => {
+    const parsed = parseMftImage(streamOf(recordOf({ recordBytes: 4096 })));
+    expect(parsed.ok).toBe(true);
+    if (!parsed.ok) return;
+    expect(parsed.value.recordBytes).toBe(4096);
+    expect(parsed.value.metadata.record_count).toBe(1);
+  });
+});
+
+describe("parseMftImage refusals and limits", () => {
+  it("refuses a file shorter than one record header", () => {
+    const parsed = parseMftImage(new Uint8Array(32));
+    expect(parsed.ok).toBe(false);
+    if (parsed.ok) return;
+    expect(parsed.error.message).toMatch(
+      /shorter than one 42-byte record header/u,
+    );
+  });
+
+  it("refuses bytes that do not start with the FILE signature", () => {
+    const parsed = parseMftImage(Buffer.alloc(2048, 0x41));
+    expect(parsed.ok).toBe(false);
+    if (parsed.ok) return;
+    expect(parsed.error.message).toMatch(
+      /does not begin with the FILE signature/u,
+    );
+  });
+
+  it("refuses a file above the fixed ceiling", () => {
+    const decoder = createMftDecoder(MFT_MAX_FILE_BYTES + 1);
+    expect(decoder.ok).toBe(false);
+    if (decoder.ok) return;
+    expect(decoder.error).toMatchObject({
+      kind: "limit",
+      boundary: "file-size",
+    });
+  });
+
+  it("refuses streams that differ from the admitted byte count", () => {
+    const stream = recordOf();
+    const short = createMftDecoder(stream.length + 1);
+    if (!short.ok) throw new Error(short.error.message);
+    expect(short.value.write(stream).ok).toBe(true);
+    expect(short.value.finish()).toMatchObject({
+      ok: false,
+      error: {
+        kind: "format",
+        message: "stream ended before the admitted file length",
+      },
+    });
+
+    const long = createMftDecoder(stream.length - 1);
+    if (!long.ok) throw new Error(long.error.message);
+    expect(long.value.write(stream)).toMatchObject({
+      ok: false,
+      error: {
+        kind: "format",
+        message: "received more bytes than the admitted file length",
+      },
+    });
+  });
+
+  it("returns the same inventory without retaining all the decoded records", () => {
+    const bytes = streamOf(...Array.from({ length: 33 }, () => recordOf()));
+    const decoder = createMftDecoder(bytes.length);
+    const parsed = parseMftImage(bytes);
+    if (!decoder.ok || !parsed.ok) throw new Error("invalid fixture");
+    expect(decoder.value.write(bytes).ok).toBe(true);
+    const inventory = decoder.value.finish();
+    if (!inventory.ok) throw new Error(inventory.error.message);
+    expect(inventory.value).not.toHaveProperty("records");
+    expect(inventory.value.metadata).toEqual(parsed.value.metadata);
+    expect(inventory.value.limitations).toEqual(parsed.value.limitations);
+  });
+
+  it("decodes the same stream through the chunked decoder", () => {
+    const stream = streamOf(recordOf(), recordOf({ flags: 0x0003 }));
+    const decoder = createMftDecoder(stream.length);
+    expect(decoder.ok).toBe(true);
+    if (!decoder.ok) return;
+    // Feed one byte past a record boundary so a chunk never aligns with a record.
+    for (const chunk of [stream.subarray(0, 1025), stream.subarray(1025)])
+      expect(decoder.value.write(chunk).ok).toBe(true);
+    const image = decoder.value.finish();
+    expect(image.ok).toBe(true);
+    if (!image.ok) return;
+    expect(image.value.metadata.record_count).toBe(2);
+    expect(image.value.metadata.directory_record_count).toBe(1);
+  });
+});

src/domain/windows/ntfsMft.ts

@@ -0,0 +1,512 @@
+import { err, ok, type Result } from "../result.js";
+import type {
+  RegionDecodeFailure,
+  RegionDecoder,
+} from "../firmware/regionDecoder.js";
+
+/** An NTFS FILE record begins with this signature. */
+const MFT_FILE_RECORD_MAGIC = "FILE";
+const MFT_FILE_RECORD_MAGIC_BYTES = 4;
+/** The shortest header worth reading: a FILE record header ends at `nextAttributeId`. */
+const NTFS_MFT_MIN_RECORD_BYTES = 42;
+
+const hasMftFileRecordSignature = (prefix: Uint8Array): boolean =>
+  prefix.length >= MFT_FILE_RECORD_MAGIC_BYTES &&
+  text(prefix.subarray(0, MFT_FILE_RECORD_MAGIC_BYTES)) ===
+    MFT_FILE_RECORD_MAGIC;
+
+/** A record's declared size is a power of two from 512 upward; 1024 is NTFS's default. */
+export const MFT_DEFAULT_RECORD_BYTES = 1024;
+const MFT_MIN_RECORD_BYTES = 512;
+const MFT_MAX_RECORD_BYTES = 65536;
+/** An update sequence array holds at least the sequence number and one fixup word. */
+const USA_MIN_ENTRIES = 2;
+/** NTFS fixes up one 512-byte sector at a time. */
+const FIXUP_SECTOR_BYTES = 512;
+/**
+ * Fixed whole-file ceiling. Records are inventoried from their headers only;
+ * attributes and their payloads are not decoded, so this is deliberately
+ * smaller than a volume's MFT can be.
+ */
+export const MFT_MAX_FILE_BYTES = 512 * 1024 * 1024;
+/** Limit the inline header sample independently of full stream coverage. */
+export const MFT_INLINE_HEADER_LIMIT = 32;
+
+/** FILE record header fields; offsets are from the record start. */
+const RECORD_HEADER = {
+  updateSequenceOffset: 4,
+  updateSequenceCount: 6,
+  logFileSequenceNumber: 8,
+  sequenceNumber: 16,
+  hardLinkCount: 18,
+  firstAttributeOffset: 20,
+  flags: 22,
+  usedBytes: 24,
+  totalBytes: 28,
+  baseRecordReference: 32,
+  nextAttributeId: 40,
+  recordNumber: 44,
+} as const;
+const RECORD_HEADER_WITH_NUMBER_BYTES = 48;
+
+const FILE_RECORD_IN_USE = 0x0001;
+const FILE_RECORD_IS_DIRECTORY = 0x0002;
+const FILE_RECORD_NAMES = new Map<number, string>([
+  [FILE_RECORD_IN_USE, "in-use"],
+  [FILE_RECORD_IS_DIRECTORY, "directory"],
+  [0x0004, "system"],
+  [0x0008, "view-index"],
+]);
+
+export interface MftRecordFixups {
+  /** The update sequence number the array's first entry holds, or 0 when absent. */
+  readonly sequenceNumber: number;
+  readonly entries: number;
+  /** Sectors whose trailer held the sequence number before a mismatch. */
+  readonly verifiedSectors: number;
+  /** Sectors the array covers; fewer than the record's sectors leaves a tail unverified. */
+  readonly coveredSectors: number;
+  /** The first sector whose trailer did not hold the sequence number, if any. */
+  readonly mismatchSector: number | null;
+  /** The array itself is missing, misplaced or too short to cover one sector. */
+  readonly malformed: boolean;
+}
+
+export interface MftFileReference {
+  readonly recordNumber: number;
+  readonly sequenceNumber: number;
+}
+
+const decodeMftFileReference = (reference: bigint): MftFileReference => ({
+  recordNumber: Number(reference & 0xffff_ffff_ffffn),
+  sequenceNumber: Number(reference >> 48n),
+});
+
+export interface MftRecord {
+  /** Zero-based ordinal of this record in the stream. */
+  readonly index: number;
+  /** File offset where the record starts. */
+  readonly fileOffset: number;
+  readonly sequenceNumber: number;
+  /** Hard link count as recorded; a zero count marks a record with no names. */
+  readonly hardLinkCount: number;
+  readonly flags: {
+    readonly raw: number;
+    readonly names: readonly string[];
+    readonly unknownMask: number;
+  };
+  readonly inUse: boolean;
+  readonly directory: boolean;
+  /** Used and allocated byte counts as declared by the FILE record. */
+  readonly usedBytes: number;
+  readonly totalBytes: number;
+  readonly extentValid: boolean;
+  /** The base record this is an extension of, or `null` for a base record. */
+  readonly baseRecordReference: MftFileReference | null;
+  /** The record number the header declares; `null` when the header predates the field. */
+  readonly declaredRecordNumber: number | null;
+  readonly recordNumberMatchesSlot: boolean | null;
+  /** `$LogFile` sequence number as recorded; no log is read. */
+  readonly logFileSequenceNumber: string;
+  /** Declared offset of the first attribute; no attribute decoding follows yet. */
+  readonly firstAttributeOffset: number;
+  /** Declared id of the next attribute; no attribute decoding follows yet. */
+  readonly nextAttributeId: number;
+  readonly fixups: MftRecordFixups;
+}
+
+export interface MftInventory {
+  readonly metadata: MftMetadata;
+  /** The record size the stream declared, or the fixed default when it did not. */
+  readonly recordBytes: number;
+  readonly limitations: readonly string[];
+}
+
+export interface MftImage extends MftInventory {
+  readonly records: readonly MftRecord[];
+}
+
+export interface MftMetadata {
+  readonly kind: "ntfs-mft";
+  readonly record_bytes: number;
+  readonly record_count: number;
+  readonly in_use_record_count: number;
+  readonly directory_record_count: number;
+  readonly extension_record_count: number;
+  readonly fixup_mismatch_count: number;
+  readonly fixup_unverified_record_count: number;
+  readonly truncated_tail_bytes: number;
+  readonly record_headers: readonly MftRecord[];
+  readonly omitted_record_headers: number;
+}
+
+const text = (ascii: Uint8Array): string =>
+  Buffer.from(ascii).toString("latin1");
+
+const failure = (
+  message: string,
+  boundary?: RegionDecodeFailure["boundary"],
+): RegionDecodeFailure => ({
+  kind: "format",
+  message,
+  ...(boundary === undefined ? {} : { boundary }),
+});
+
+const limit = (
+  message: string,
+  boundary?: RegionDecodeFailure["boundary"],
+): RegionDecodeFailure => ({
+  kind: "limit",
+  message,
+  ...(boundary === undefined ? {} : { boundary }),
+});
+
+const parseRecordFlags = (raw: number): MftRecord["flags"] => {
+  const names: string[] = [];
+  let unknownMask = raw;
+  for (const [bit, name] of FILE_RECORD_NAMES) {
+    if ((raw & bit) !== 0) names.push(name);
+    unknownMask &= ~bit;
+  }
+  return { raw, names, unknownMask };
+};
+
+const isRecordSize = (bytes: number): boolean =>
+  bytes >= MFT_MIN_RECORD_BYTES &&
+  bytes <= MFT_MAX_RECORD_BYTES &&
+  (bytes & (bytes - 1)) === 0;
+
+const updateSequenceArrayFits = (
+  offset: number,
+  entries: number,
+  sectors: number,
+): boolean =>
+  entries >= USA_MIN_ENTRIES &&
+  entries <= sectors + 1 &&
+  offset >= NTFS_MFT_MIN_RECORD_BYTES &&
+  offset % 2 === 0 &&
+  offset + entries * 2 <= FIXUP_SECTOR_BYTES - 2;
+
+/** Recognize a FILE header without requiring its trailers or attributes to verify. */
+export const hasMftFileRecordHeader = (prefix: Uint8Array): boolean => {
+  if (
+    prefix.length < NTFS_MFT_MIN_RECORD_BYTES ||
+    !hasMftFileRecordSignature(prefix)
+  )
+    return false;
+  const view = new DataView(
+    prefix.buffer,
+    prefix.byteOffset,
+    prefix.byteLength,
+  );
+  return (
+    isRecordSize(view.getUint32(RECORD_HEADER.totalBytes, true)) ||
+    updateSequenceArrayFits(
+      view.getUint16(RECORD_HEADER.updateSequenceOffset, true),
+      view.getUint16(RECORD_HEADER.updateSequenceCount, true),
+      MFT_DEFAULT_RECORD_BYTES / FIXUP_SECTOR_BYTES,
+    )
+  );
+};
+
+/**
+ * Parse FILE record headers and verify their update-sequence fixups.
+ *
+ * Only the record structure is established: attributes, their names, the
+ * filenames and the USN journal are left to later steps, so no path, timestamp
+ * or file identity beyond the record itself is claimed. A record whose fixup
+ * does not match keeps its header facts, because the stream still declares it,
+ * and the mismatch is reported on the record and counted in the metadata.
+ */
+export const parseMftImage = (
+  bytes: Uint8Array,
+): Result<MftImage, RegionDecodeFailure> => {
+  const records: MftRecord[] = [];
+  const inventory = scanMft(bytes, records);
+  return inventory.ok ? ok({ ...inventory.value, records }) : inventory;
+};
+
+const scanMft = (
+  bytes: Uint8Array,
+  retainedRecords?: MftRecord[],
+): Result<MftInventory, RegionDecodeFailure> => {
+  if (bytes.length < NTFS_MFT_MIN_RECORD_BYTES)
+    return err(
+      failure(
+        `file is ${String(bytes.length)} bytes, shorter than one ${String(NTFS_MFT_MIN_RECORD_BYTES)}-byte record header`,
+      ),
+    );
+  if (!hasMftFileRecordSignature(bytes))
+    return err(failure("record 0 does not begin with the FILE signature"));
+
+  const limitations: string[] = [];
+  const header = new DataView(bytes.buffer, bytes.byteOffset, bytes.byteLength);
+  const declaredRecordBytes = header.getUint32(RECORD_HEADER.totalBytes, true);
+  const recordBytes = isRecordSize(declaredRecordBytes)
+    ? declaredRecordBytes
+    : MFT_DEFAULT_RECORD_BYTES;
+  if (!isRecordSize(declaredRecordBytes))
+    limitations.push(
+      `record 0 declares a record size of ${String(declaredRecordBytes)} bytes, which is not a supported power of two from ${String(MFT_MIN_RECORD_BYTES)} to ${String(MFT_MAX_RECORD_BYTES)}; records were read at the ${String(MFT_DEFAULT_RECORD_BYTES)}-byte default`,
+    );
+
+  const recordHeaders: MftRecord[] = [];
+  let recordCount = 0;
+  let inUseCount = 0;
+  let directoryCount = 0;
+  let extensionCount = 0;
+  let mismatchCount = 0;
+  let unverifiedCount = 0;
+  let invalidSlots = 0;
+  let invalidExtents = 0;
+  let inconsistentRecordNumbers = 0;
+  let firstInvalidExtentOffset: number | undefined;
+  let firstInvalidOffset: number | undefined;
+  for (
+    let offset = 0;
+    offset + recordBytes <= bytes.length;
+    offset += recordBytes
+  ) {
+    if (
+      !hasMftFileRecordSignature(
+        bytes.subarray(offset, offset + MFT_FILE_RECORD_MAGIC_BYTES),
+      )
+    ) {
+      invalidSlots += 1;
+      firstInvalidOffset ??= offset;
+      if (!isRecordSize(declaredRecordBytes)) break;
+      continue;
+    }
+    const record = readRecord(bytes, offset, offset / recordBytes, recordBytes);
+    retainedRecords?.push(record);
+    if (recordHeaders.length < MFT_INLINE_HEADER_LIMIT)
+      recordHeaders.push(record);
+    recordCount += 1;
+    if (record.inUse) inUseCount += 1;
+    if (record.directory) directoryCount += 1;
+    if (record.baseRecordReference !== null) extensionCount += 1;
+    if (record.fixups.mismatchSector !== null) mismatchCount += 1;
+    if (
+      record.fixups.malformed ||
+      record.fixups.coveredSectors < recordBytes / FIXUP_SECTOR_BYTES
+    )
+      unverifiedCount += 1;
+    if (record.recordNumberMatchesSlot === false)
+      inconsistentRecordNumbers += 1;
+    if (!record.extentValid) {
+      invalidExtents += 1;
+      firstInvalidExtentOffset ??= offset;
+    }
+  }
+  if (firstInvalidOffset !== undefined)
+    limitations.push(
+      `record at ${String(firstInvalidOffset)} does not begin with the FILE signature; ${String(invalidSlots)} slot(s) without a FILE signature were found`,
+    );
+
+  if (firstInvalidExtentOffset !== undefined)
+    limitations.push(
+      `${String(invalidExtents)} FILE record(s) declare inconsistent record extents, beginning at offset ${String(firstInvalidExtentOffset)}`,
+    );
+
+  if (inconsistentRecordNumbers !== 0)
+    limitations.push(
+      `${String(inconsistentRecordNumbers)} FILE record(s) declare a record number that disagrees with their stream slot`,
+    );
+
+  const truncatedTailBytes = bytes.length % recordBytes;
+  if (truncatedTailBytes !== 0)
+    limitations.push(
+      `${String(truncatedTailBytes)} trailing byte(s) do not fill a ${String(recordBytes)}-byte record and were not parsed`,
+    );
+
+  if (mismatchCount !== 0)
+    limitations.push(
+      `${String(mismatchCount)} record(s) hold a sector trailer that is not the update sequence number; their attributes were not read and no later step should trust them`,
+    );
+  if (unverifiedCount !== 0)
+    limitations.push(
+      `${String(unverifiedCount)} record(s) carry an invalid or incomplete update sequence array; their sector trailers were not fully verified`,
+    );
+
+  const metadata: MftMetadata = {
+    kind: "ntfs-mft",
+    record_bytes: recordBytes,
+    record_count: recordCount,
+    in_use_record_count: inUseCount,
+    directory_record_count: directoryCount,
+    extension_record_count: extensionCount,
+    fixup_mismatch_count: mismatchCount,
+    fixup_unverified_record_count: unverifiedCount,
+    truncated_tail_bytes: truncatedTailBytes,
+    record_headers: recordHeaders,
+    omitted_record_headers: recordCount - recordHeaders.length,
+  };
+
+  return ok({ metadata, recordBytes, limitations });
+};
+
+const readRecord = (
+  bytes: Uint8Array,
+  fileOffset: number,
+  index: number,
+  recordBytes: number,
+): MftRecord => {
+  const record = bytes.subarray(fileOffset, fileOffset + recordBytes);
+  const view = new DataView(
+    record.buffer,
+    record.byteOffset,
+    record.byteLength,
+  );
+
+  const flags = view.getUint16(RECORD_HEADER.flags, true);
+  const baseRecordReference = view.getBigUint64(
+    RECORD_HEADER.baseRecordReference,
+    true,
+  );
+
+  const usaOffset = view.getUint16(RECORD_HEADER.updateSequenceOffset, true);
+  const fixups = verifyFixups(record, view, usaOffset);
+  const usedBytes = view.getUint32(RECORD_HEADER.usedBytes, true);
+  const totalBytes = view.getUint32(RECORD_HEADER.totalBytes, true);
+  const firstAttributeOffset = view.getUint16(
+    RECORD_HEADER.firstAttributeOffset,
+    true,
+  );
+
+  const declaredRecordNumber =
+    !fixups.malformed && usaOffset >= RECORD_HEADER_WITH_NUMBER_BYTES
+      ? view.getUint32(RECORD_HEADER.recordNumber, true)
+      : null;
+
+  return {
+    index,
+    fileOffset,
+    sequenceNumber: view.getUint16(RECORD_HEADER.sequenceNumber, true),
+    hardLinkCount: view.getUint16(RECORD_HEADER.hardLinkCount, true),
+    flags: parseRecordFlags(flags),
+    inUse: (flags & FILE_RECORD_IN_USE) !== 0,
+    directory: (flags & FILE_RECORD_IS_DIRECTORY) !== 0,
+    usedBytes,
+    totalBytes,
+    extentValid:
+      totalBytes === recordBytes &&
+      usedBytes >= NTFS_MFT_MIN_RECORD_BYTES &&
+      usedBytes % 8 === 0 &&
+      usedBytes <= totalBytes &&
+      firstAttributeOffset % 8 === 0 &&
+      firstAttributeOffset >= NTFS_MFT_MIN_RECORD_BYTES &&
+      firstAttributeOffset + 4 <= usedBytes &&
+      (fixups.malformed ||
+        firstAttributeOffset >= usaOffset + fixups.entries * 2),
+    baseRecordReference:
+      baseRecordReference === 0n
+        ? null
+        : decodeMftFileReference(baseRecordReference),
+    declaredRecordNumber,
+    recordNumberMatchesSlot:
+      declaredRecordNumber === null ? null : declaredRecordNumber === index,
+    logFileSequenceNumber: view
+      .getBigUint64(RECORD_HEADER.logFileSequenceNumber, true)
+      .toString(),
+    firstAttributeOffset,
+    nextAttributeId: view.getUint16(RECORD_HEADER.nextAttributeId, true),
+    fixups,
+  };
+};
+
+/**
+ * Each sector's last two bytes are replaced on disk by the update sequence
+ * number; the array's first entry is that number and the rest are the saved
+ * trailer words. A sector whose trailer is not the sequence number was not
+ * cleanly written, so the record's attributes cannot be trusted.
+ */
+const verifyFixups = (
+  record: Uint8Array,
+  view: DataView,
+  usaOffset: number,
+): MftRecordFixups => {
+  const usaCount = view.getUint16(RECORD_HEADER.updateSequenceCount, true);
+  const sectors = Math.floor(record.length / FIXUP_SECTOR_BYTES);
+  // Entry 0 holds the sequence number, so sector n's saved trailer is entry n+1.
+  const coveredSectors = Math.max(0, usaCount - 1);
+
+  if (!updateSequenceArrayFits(usaOffset, usaCount, sectors))
+    return {
+      sequenceNumber: 0,
+      entries: usaCount,
+      verifiedSectors: 0,
+      coveredSectors: 0,
+      mismatchSector: null,
+      malformed: true,
+    };
+
+  const sequenceNumber = view.getUint16(usaOffset, true);
+  if (sequenceNumber === 0 || sequenceNumber === 0xffff)
+    return {
+      sequenceNumber,
+      entries: usaCount,
+      verifiedSectors: 0,
+      coveredSectors,
+      mismatchSector: null,
+      malformed: true,
+    };
+  for (let sector = 0; sector < coveredSectors; sector += 1) {
+    const trailerOffset = sector * FIXUP_SECTOR_BYTES + FIXUP_SECTOR_BYTES - 2;
+    if (view.getUint16(trailerOffset, true) !== sequenceNumber)
+      return {
+        sequenceNumber,
+        entries: usaCount,
+        verifiedSectors: sector,
+        coveredSectors,
+        mismatchSector: sector,
+        malformed: false,
+      };
+  }
+
+  return {
+    sequenceNumber,
+    entries: usaCount,
+    verifiedSectors: coveredSectors,
+    coveredSectors,
+    mismatchSector: null,
+    malformed: false,
+  };
+};
+
+/** Buffer the admitted input once while accepting arbitrary chunk boundaries. */
+class StreamingMftDecoder implements RegionDecoder<MftInventory> {
+  readonly #bytes: Uint8Array;
+  #written = 0;
+
+  constructor(fileBytes: number) {
+    this.#bytes = new Uint8Array(fileBytes);
+  }
+
+  write(chunk: Uint8Array): Result<void, RegionDecodeFailure> {
+    if (chunk.length > this.#bytes.length - this.#written)
+      return err(failure("received more bytes than the admitted file length"));
+    this.#bytes.set(chunk, this.#written);
+    this.#written += chunk.length;
+    return ok(undefined);
+  }
+
+  finish(): Result<MftInventory, RegionDecodeFailure> {
+    if (this.#written !== this.#bytes.length)
+      return err(failure("stream ended before the admitted file length"));
+    return scanMft(this.#bytes);
+  }
+}
+
+export const createMftDecoder = (
+  fileBytes: number,
+): Result<RegionDecoder<MftInventory>, RegionDecodeFailure> =>
+  Number.isSafeInteger(fileBytes) &&
+  fileBytes >= 0 &&
+  fileBytes <= MFT_MAX_FILE_BYTES
+    ? ok(new StreamingMftDecoder(fileBytes))
+    : err(
+        limit(
+          `file is ${String(fileBytes)} bytes, exceeding the fixed ${String(MFT_MAX_FILE_BYTES)}-byte (512 MiB) ceiling`,
+          "file-size",
+        ),
+      );

src/ghidra/GhidraFunctionValues.test.ts

@@ -1120,23 +1120,23 @@ describe("Ghidra native instruction decoder context for ARM TMode and MIPS ISA_M
   });
 });
 
-describe("Ghidra bridge script processor context contract", () => {
-  it("binds TMode to ARM and ISA_MODE to MIPS in ReaGhidraBridge.java", async () => {
-    const bridgePath = new URL(
-      "../../bridge/ghidra/ReaGhidraBridge.java",
-      import.meta.url,
-    );
-    const content = await readFile(bridgePath, "utf8");
-    expect(content).toContain(
-      'currentProgram.getLanguageID().getIdAsString().startsWith("ARM:")',
-    );
-    expect(content).toContain(
-      'currentProgram.getLanguageID().getIdAsString().startsWith("MIPS:")',
-    );
-    expect(content).toContain('"TMode"');
-    expect(content).toContain('"ISA_MODE"');
-    expect(content).toContain('"ghidra-program-context"');
-  });
+// Kept on the file rather than in its own suite. The full source run collected
+// that suite with no tests and failed the file before this assertion could run.
+it("binds TMode to ARM and ISA_MODE to MIPS in the Ghidra bridge script", async () => {
+  const bridgePath = new URL(
+    "../../bridge/ghidra/ReaGhidraBridge.java",
+    import.meta.url,
+  );
+  const content = await readFile(bridgePath, "utf8");
+  expect(content).toContain(
+    'currentProgram.getLanguageID().getIdAsString().startsWith("ARM:")',
+  );
+  expect(content).toContain(
+    'currentProgram.getLanguageID().getIdAsString().startsWith("MIPS:")',
+  );
+  expect(content).toContain('"TMode"');
+  expect(content).toContain('"ISA_MODE"');
+  expect(content).toContain('"ghidra-program-context"');
 });
 
 describe("Ghidra dossier nested boundary composition", () => {

src/ghidra/GhidraProvider.behavior.test.ts

@@ -1868,7 +1868,8 @@ describe("Ghidra decompiler failure projection", () => {
             applies_at: "next_operation",
             agent_editable: true,
             scope: "rea-runtime",
-            rea_restart_required: false,
+            update_mode: "runtime",
+            pending_restart: false,
           },
         },
       });
@@ -1983,7 +1984,8 @@ describe("Ghidra decompile deadline for a finished invocation", () => {
         minimum: 1,
         maximum: 2_147_483,
         agent_editable: true,
-        rea_restart_required: false,
+        update_mode: "runtime",
+        pending_restart: false,
       };
       expect(invocation.details?.limit).toEqual({
         ...enforced,
@@ -4365,7 +4367,8 @@ describe("Ghidra startup deadline for a finished invocation", () => {
     setting: "ghidra.startup_timeout_ms",
     effective_value: 250,
     agent_editable: true,
-    rea_restart_required: false,
+    update_mode: "runtime",
+    pending_restart: false,
   };
   const invocationLimit = {
     ...startupLimit,

src/ida/IdaSessionClient.test.ts

@@ -187,7 +187,8 @@ it.each(["initialize", "transport"])(
         applies_at: "next_operation",
         agent_editable: true,
         scope: "rea-runtime",
-        rea_restart_required: false,
+        update_mode: "runtime",
+        pending_restart: false,
       },
       diagnostics: { producer_fact: "retained" },
     });

src/javascript/analysis/JavaScriptAnalysisWorker.ts

@@ -948,7 +948,7 @@ export class JavaScriptAnalysisWorker {
         limit: analysisSettingLimit(
           "javascript.max_heap_mb",
           this.limits.heapMb,
-          undefined,
+          this.limits.heapSource,
         ),
         remediationAction: javaScriptWorkerHeapRemediation(this.limits.heapMb),
       },
@@ -1147,7 +1147,7 @@ export class JavaScriptAnalysisWorker {
           limit: analysisSettingLimit(
             "javascript.analysis_timeout_ms",
             this.limits.timeoutMs,
-            undefined,
+            this.limits.timeoutSource,
           ),
           capturedOutput: this.#captured(),
           userMessage: `JavaScript source analysis reached its ${String(this.limits.timeoutMs)}ms per-source server/provider deadline (analysis_timeout_ms), not the MCP client deadline.`,
@@ -1182,7 +1182,7 @@ export class JavaScriptAnalysisWorker {
                 limit: analysisSettingLimit(
                   "javascript.max_heap_mb",
                   this.limits.heapMb,
-                  undefined,
+                  this.limits.heapSource,
                 ),
               }
             : {}),

src/process/ownedDebuggerSession.ts

@@ -22,7 +22,8 @@ export const debuggerAdmissionDiagnostics = (
   readonly minimum: number;
   readonly maximum: number;
   readonly unit: DebuggerAdmissionLimit["unit"];
-  readonly restart_required: true;
+  readonly update_mode: "restart";
+  readonly pending_restart: false;
 } => ({
   setting: admission.setting,
   effective_value: admission.value,
@@ -31,7 +32,8 @@ export const debuggerAdmissionDiagnostics = (
   minimum: admission.minimum,
   maximum: admission.maximum,
   unit: admission.unit,
-  restart_required: true,
+  update_mode: "restart",
+  pending_restart: false,
 });
 
 /** Name the setting, the effective value, its source, the range, and a restart. */

src/server/environment/StateResponseCleanup.ts

@@ -0,0 +1,53 @@
+import {
+  ArtifactResourceScope,
+  type ArtifactResourceOwner,
+} from "../../artifacts/ArtifactResourceScope.js";
+
+/** Retain a cancelled response branch until its underlying cancellation settles. */
+export class StateResponseCleanup {
+  readonly #scope = new ArtifactResourceScope();
+
+  constructor(private readonly changed: () => void = () => {}) {}
+
+  get cleanupResources(): readonly string[] {
+    return this.#scope.cleanupResources;
+  }
+
+  cancel(cancel: () => Promise<void>): void {
+    let state:
+      | { status: "pending" }
+      | { status: "done" }
+      | { status: "failed"; cause: unknown } = { status: "pending" };
+    const owner: ArtifactResourceOwner = {
+      kind: "lease",
+      resource: "state response body",
+      lease: {
+        release: async () => {
+          if (state.status === "pending")
+            throw new Error("State response cancellation is still pending");
+          if (state.status === "failed") throw state.cause;
+        },
+      },
+    };
+    this.#scope.retain(owner);
+    this.changed();
+    void Promise.resolve()
+      .then(cancel)
+      .then(
+        async () => {
+          state = { status: "done" };
+          await this.#scope.release(owner);
+          this.changed();
+        },
+        (cause: unknown) => {
+          state = { status: "failed", cause };
+          this.changed();
+        },
+      )
+      .catch(() => undefined);
+  }
+
+  close(): Promise<void> {
+    return this.#scope.close();
+  }
+}

src/server/environment/environmentDataPlane.ts

@@ -1,3 +1,4 @@
+import { StateResponseCleanup } from "./StateResponseCleanup.js";
 import type { EnvironmentStateStore } from "../../application/environment/environmentStateStore.js";
 import { ArtifactResourceScope } from "../../artifacts/ArtifactResourceScope.js";
 import type { AnalysisInputError } from "../../domain/analysisErrorCore.js";
@@ -46,17 +47,24 @@ interface CallContext {
   readonly intake: StateIntake;
   readonly cancel: AbortSignal;
   readonly files: ArtifactResourceScope;
+  readonly responses: StateResponseCleanup;
 }
 
 /** `data/add`: exactly one part, whose document replaces the whole state only when it is valid and the call was not cancelled. */
 const addState = async (
-  { store, intake, cancel, files }: CallContext,
+  { store, intake, cancel, files, responses }: CallContext,
   params: unknown,
 ): Promise<RpcOutcome> => {
   const parts = isRecord(params) ? params.parts : undefined;
   if (!Array.isArray(parts) || parts.length !== 1)
     return { error: invalidParams("data/add takes exactly one part") };
-  const document = await partDocument(parts[0], intake, cancel, files);
+  const document = await partDocument(
+    parts[0],
+    intake,
+    cancel,
+    files,
+    responses,
+  );
   if (!document.ok) return { error: document.error };
   // A file read is not interruptible, so a cancellation during it is honoured here, before the state changes.
   if (cancel.aborted) return { error: CANCELLED };
@@ -101,6 +109,7 @@ export const stateDataPlane = (
   intake: StateIntake = DEFAULT_STATE_INTAKE,
   stopped: AbortSignal = new AbortController().signal,
   files: ArtifactResourceScope = new ArtifactResourceScope(),
+  responses: StateResponseCleanup = new StateResponseCleanup(),
 ): RpcMethods => {
   let previous: Promise<unknown> = Promise.resolve();
   return (method, params, signal) => {
@@ -111,7 +120,7 @@ export const stateDataPlane = (
     const next = previous.then(() =>
       cancel.aborted
         ? { error: CANCELLED }
-        : run({ store, intake, cancel, files }, params),
+        : run({ store, intake, cancel, files, responses }, params),
     );
     previous = next.catch(() => undefined);
     return next;

src/server/environment/serveEnvironment.ts

@@ -1,3 +1,4 @@
+import { StateResponseCleanup } from "./StateResponseCleanup.js";
 import type { HttpHostname } from "../../domain/httpHostname.js";
 import { STATE_INTAKE_LIMIT_BYTES } from "../../application/environment/environmentStateDocument.js";
 import {
@@ -477,6 +478,8 @@ export const serveEnvironment = async (
   };
   // A state file whose close failed stays owned here: later file reads retry it, and close reports it if it is still open.
   const stateFiles = new ArtifactResourceScope();
+  let responseStateChanged = (): void => undefined;
+  const stateResponses = new StateResponseCleanup(() => responseStateChanged());
   const served = await serveStreamableHttp({
     host,
     port,
@@ -499,6 +502,7 @@ export const serveEnvironment = async (
             },
             stopping.signal,
             stateFiles,
+            stateResponses,
           ),
     ),
     ...(logger === undefined ? {} : { logger }),
@@ -512,11 +516,14 @@ export const serveEnvironment = async (
   });
   let httpClosePending = false;
   let stateClosePending = false;
+  let responseClosePending = false;
+  let settledSuccessful = false;
   let closing: Promise<void> | undefined;
   const shutdownFailures = new Set<AnalysisError>();
   const owner: EnvironmentShutdownOwner = {
     get cleanupResources() {
       const stateResources = stateFiles.cleanupResources;
+      const responseResources = stateResponses.cleanupResources;
       return [
         ...(httpClosePending ? [served.url] : []),
         ...(stateClosePending
@@ -524,6 +531,11 @@ export const serveEnvironment = async (
             ? stateResources
             : ["environment state-file scope"]
           : []),
+        ...(responseClosePending
+          ? responseResources.length > 0
+            ? responseResources
+            : ["state response cleanup"]
+          : []),
       ];
     },
     close: () => {
@@ -531,7 +543,7 @@ export const serveEnvironment = async (
       const attempt = (async () => {
         stop();
         const failures: {
-          readonly stage: "http-server" | "state-files";
+          readonly stage: "http-server" | "state-files" | "state-responses";
           readonly cause: unknown;
         }[] = [];
         try {
@@ -548,6 +560,15 @@ export const serveEnvironment = async (
           stateClosePending = true;
           failures.push({ stage: "state-files", cause });
         }
+        try {
+          await stateResponses.close();
+          if (stateResponses.cleanupResources.length > 0)
+            throw new Error("State response cleanup remains pending");
+          responseClosePending = false;
+        } catch (cause: unknown) {
+          responseClosePending = true;
+          failures.push({ stage: "state-responses", cause });
+        }
         if (failures.length > 0) {
           retainedEnvironmentShutdowns.add(owner);
           const failure = new AnalysisOutputError(
@@ -583,6 +604,7 @@ export const serveEnvironment = async (
         retainedEnvironmentShutdowns.delete(owner);
         for (const failure of shutdownFailures) failure.resolveCleanup();
         shutdownFailures.clear();
+        settledSuccessful = true;
       })();
       closing = attempt;
       void attempt.catch(() => {
@@ -591,6 +613,26 @@ export const serveEnvironment = async (
       return attempt;
     },
   };
+  // A fetcher may hand back a response after an earlier stop completed. Keep
+  // that newly acquired body owned and make a later close observe its outcome.
+  responseStateChanged = () => {
+    if (!stopping.signal.aborted) return;
+    if (stateResponses.cleanupResources.length > 0) {
+      responseClosePending = true;
+      retainedEnvironmentShutdowns.add(owner);
+      if (settledSuccessful) {
+        settledSuccessful = false;
+        closing = undefined;
+      }
+    } else if (
+      shutdownFailures.size === 0 &&
+      !httpClosePending &&
+      !stateClosePending
+    ) {
+      responseClosePending = false;
+      retainedEnvironmentShutdowns.delete(owner);
+    }
+  };
   return ok({
     ...served,
     card,

src/server/environment/serveEnvironmentState.test.ts

@@ -1037,7 +1037,122 @@ describe("serveEnvironment with a state", () => {
       expect(missing.isError).toBe(true);
     });
   });
+});
 
+describe("state response cleanup", () => {
+  it("reports pending teed response cleanup and retries after its sibling closes", async () => {
+    let sourceCancelled = false;
+    const stream = new ReadableStream<Uint8Array>({
+      start(controller) {
+        controller.enqueue(new Uint8Array([1]));
+      },
+      cancel() {
+        sourceCancelled = true;
+      },
+    });
+    const [body, sibling] = stream.tee();
+    const environment = await serveState(
+      { source, state },
+      {
+        fetchState: async () =>
+          new Response(body, {
+            headers: { "content-length": String(STATE_INTAKE_LIMIT_BYTES + 1) },
+          }),
+      },
+    );
+    let timer: ReturnType<typeof setTimeout> | undefined;
+    try {
+      const response = await fetch(
+        new URL(ENVIRONMENT_RPC_PATH, environment.url),
+        {
+          method: "POST",
+          headers: { "content-type": "application/json" },
+          body: JSON.stringify(addByUri(1)),
+          signal: AbortSignal.timeout(10_000),
+        },
+      );
+      await expect(response.json()).resolves.toMatchObject({
+        error: { code: -32602 },
+      });
+      const closing = Promise.race([
+        environment.close(),
+        new Promise<never>((_, reject) => {
+          timer = setTimeout(
+            () => reject(new Error("server close waited for tee sibling")),
+            10_000,
+          );
+        }),
+      ]);
+      await expect(closing).rejects.toMatchObject({
+        _tag: "AnalysisOutputError",
+        cleanupIncomplete: true,
+        diagnostics: { cleanup_stages: ["state-responses"] },
+      });
+      expect(sourceCancelled).toBe(false);
+      await sibling.cancel();
+      expect(sourceCancelled).toBe(true);
+      await expect(environment.close()).resolves.toBeUndefined();
+    } finally {
+      if (timer !== undefined) clearTimeout(timer);
+      await Promise.allSettled([body.cancel(), sibling.cancel()]);
+      await environment.close();
+    }
+  });
+
+  it("retains a late response cleanup after an earlier successful stop", async () => {
+    let giveResponse!: (response: Response) => void;
+    const gate = new Promise<Response>((resolve) => {
+      giveResponse = resolve;
+    });
+    const held = heldFetcher(() => gate);
+    let release!: () => void;
+    let acknowledge!: () => void;
+    const cleanup = new Promise<void>((resolve) => {
+      release = resolve;
+    });
+    const entered = new Promise<void>((resolve) => {
+      acknowledge = resolve;
+    });
+    const environment = await serveState(
+      { source, state },
+      { fetchState: held.fetchState },
+    );
+    const answered = fetch(new URL(ENVIRONMENT_RPC_PATH, environment.url), {
+      method: "POST",
+      headers: { "content-type": "application/json" },
+      body: JSON.stringify(addByUri(1)),
+      signal: AbortSignal.timeout(10_000),
+    }).catch(() => undefined);
+    try {
+      await held.fetching;
+      await environment.close();
+      giveResponse(
+        new Response(
+          new ReadableStream<Uint8Array>({
+            cancel() {
+              acknowledge();
+              return cleanup;
+            },
+          }),
+        ),
+      );
+      await entered;
+      await expect(environment.close()).rejects.toMatchObject({
+        _tag: "AnalysisOutputError",
+        cleanupIncomplete: true,
+        diagnostics: { cleanup_stages: ["state-responses"] },
+      });
+      release();
+      await expect(environment.close()).resolves.toBeUndefined();
+    } finally {
+      release();
+      await answered;
+      await environment.close();
+    }
+  });
+});
+
+describe("serveEnvironment state file cleanup", () => {
   it("stops without waiting for a stalled download", async () => {
     const held = heldFetcher(() => new Promise<Response>(() => undefined));
     const environment = await serveState(

src/server/environment/stateIntake.test.ts

@@ -1,3 +1,4 @@
+import { StateResponseCleanup } from "./StateResponseCleanup.js";
 import { expect, it, vi } from "vitest";
 
 import { ArtifactResourceScope } from "../../artifacts/ArtifactResourceScope.js";
@@ -11,14 +12,17 @@ import {
 
 const inlineDocument = async (bytes: string) => {
   const files = new ArtifactResourceScope();
+  const responses = new StateResponseCleanup();
   try {
     return await partDocument(
       { kind: "file", file: { bytes } },
       DEFAULT_STATE_INTAKE,
       new AbortController().signal,
       files,
+      responses,
     );
   } finally {
+    await responses.close();
     await files.close();
   }
 };
@@ -79,3 +83,109 @@ it("downloads with redirects refused and the 30 s timeout by default", async ()
     fetched.mockRestore();
   }
 });
+
+it.each(["http-status", "declared-overflow", "stream-overflow"] as const)(
+  "returns %s while a teed response's other branch remains open",
+  async (scenario) => {
+    const source = new ReadableStream<Uint8Array>({
+      start(controller) {
+        controller.enqueue(
+          new Uint8Array(
+            scenario === "stream-overflow" ? STATE_INTAKE_LIMIT_BYTES + 1 : 1,
+          ),
+        );
+      },
+    });
+    const [body, sibling] = source.tee();
+    const response = new Response(body, {
+      status: scenario === "http-status" ? 503 : 200,
+      ...(scenario === "declared-overflow"
+        ? {
+            headers: { "content-length": String(STATE_INTAKE_LIMIT_BYTES + 1) },
+          }
+        : {}),
+    });
+    const files = new ArtifactResourceScope();
+    const responses = new StateResponseCleanup();
+    const operation = partDocument(
+      { kind: "file", file: { uri: "https://state.example.test/state.json" } },
+      { fetchState: async () => response, timeoutMs: 1_000 },
+      new AbortController().signal,
+      files,
+      responses,
+    );
+    let timer: ReturnType<typeof setTimeout> | undefined;
+    try {
+      const outcome = await Promise.race([
+        operation.then((result) => ({ kind: "result" as const, result })),
+        new Promise<{ kind: "deadline" }>((resolve) => {
+          timer = setTimeout(() => resolve({ kind: "deadline" }), 3_000);
+        }),
+      ]);
+      expect(outcome.kind).toBe("result");
+      expect(responses.cleanupResources).toEqual(["state response body"]);
+      await expect(responses.close()).rejects.toMatchObject({
+        reason: "unavailable",
+      });
+      if (outcome.kind === "result") {
+        expect(outcome.result).toMatchObject(
+          scenario === "http-status"
+            ? { ok: false, error: { data: { reason: "http-status" } } }
+            : {
+                ok: false,
+                error: {
+                  message: expect.stringContaining(
+                    String(STATE_INTAKE_LIMIT_BYTES),
+                  ),
+                },
+              },
+        );
+      }
+    } finally {
+      if (timer !== undefined) clearTimeout(timer);
+      await sibling.cancel();
+      await operation;
+      await responses.close();
+      await files.close();
+    }
+  },
+);
+
+it("retains the exact failure when response cancellation rejects", async () => {
+  const failure = new Error("controlled cancellation failure");
+  const body = new ReadableStream<Uint8Array>({
+    cancel() {
+      throw failure;
+    },
+  });
+  const files = new ArtifactResourceScope();
+  const responses = new StateResponseCleanup();
+  const result = await partDocument(
+    { kind: "file", file: { uri: "https://state.example.test/state.json" } },
+    {
+      fetchState: async () => new Response(body, { status: 503 }),
+      timeoutMs: 1_000,
+    },
+    new AbortController().signal,
+    files,
+    responses,
+  );
+  expect(result).toMatchObject({
+    ok: false,
+    error: { data: { reason: "http-status" } },
+  });
+  await new Promise<void>((resolve) => setImmediate(resolve));
+  const rejected: unknown = await responses
+    .close()
+    .catch((cause: unknown) => cause);
+  expect(rejected).toMatchObject({
+    reason: "unavailable",
+    cleanup: { resources: ["state response body"] },
+  });
+  expect((rejected as Error).cause).toBeInstanceOf(AggregateError);
+  expect(((rejected as Error).cause as AggregateError).errors).toContain(
+    failure,
+  );
+  expect(responses.cleanupResources).toEqual(["state response body"]);
+  await files.close();
+});

src/server/environment/stateIntake.ts

@@ -1,3 +1,4 @@
+import { StateResponseCleanup } from "./StateResponseCleanup.js";
 import { fileURLToPath } from "node:url";
 
 import { mediaTypeEssence } from "../../domain/analysisView/recordedMediaType.js";
@@ -228,6 +229,7 @@ const downloadBounds = (
 const readCapped = async (
   body: ReadableStream<Uint8Array> | null,
   bounds: DownloadBounds,
+  responses: StateResponseCleanup,
 ): Promise<Result<Buffer, RpcError>> => {
   if (body === null) return ok(Buffer.alloc(0));
   const reader = body.getReader();
@@ -240,13 +242,15 @@ const readCapped = async (
       if (done) return ok(Buffer.concat(chunks, total));
       total += value.byteLength;
       if (total > STATE_INTAKE_LIMIT_BYTES) {
-        await reader.cancel().catch(() => undefined);
+        // A tee's sibling can be caller-owned; cancelling this branch must not
+        // delay the refusal until that sibling finishes.
+        responses.cancel(() => reader.cancel());
         return err(TOO_LARGE);
       }
       chunks.push(value);
     }
   } catch {
-    void reader.cancel().catch(() => undefined);
+    responses.cancel(() => reader.cancel());
     return err(fetchFailed(bounds.failure()));
   }
 };
@@ -256,6 +260,7 @@ const downloadState = async (
   url: URL,
   intake: StateIntake,
   cancel: AbortSignal,
+  responses: StateResponseCleanup,
 ): Promise<Result<Buffer, RpcError>> => {
   const bounds = downloadBounds(intake.timeoutMs, cancel);
   const fetching = Promise.resolve().then(() =>
@@ -268,7 +273,9 @@ const downloadState = async (
     // A response that arrives after the download ended is released unread.
     void fetching
       .then(
-        (late) => late.body?.cancel(),
+        (late) => {
+          if (late.body !== null) responses.cancel(() => late.body!.cancel());
+        },
         () => undefined,
       )
       .catch(() => undefined);
@@ -276,7 +283,8 @@ const downloadState = async (
   }
   const declared = Number(response.headers.get("content-length") ?? "0");
   if (!response.ok || declared > STATE_INTAKE_LIMIT_BYTES) {
-    await response.body?.cancel().catch(() => undefined);
+    // Cancel this branch without waiting for a potentially independent tee sibling.
+    if (response.body !== null) responses.cancel(() => response.body!.cancel());
     return err(
       response.ok
         ? TOO_LARGE
@@ -286,7 +294,7 @@ const downloadState = async (
           },
     );
   }
-  return readCapped(response.body, bounds);
+  return readCapped(response.body, bounds, responses);
 };
 
 const NETWORK_SHARE = invalidParams(
@@ -315,6 +323,7 @@ const fileBytes = async (
   intake: StateIntake,
   cancel: AbortSignal,
   files: ArtifactResourceScope,
+  responses: StateResponseCleanup,
 ): Promise<Result<Uint8Array, RpcError>> => {
   if (typeof file.bytes === "string" && file.uri === undefined) {
     if (file.bytes.length > MAX_INLINE_BASE64_CHARACTERS) return err(TOO_LARGE);
@@ -333,7 +342,8 @@ const fileBytes = async (
       invalidParams("a file part carries exactly one of bytes or a uri"),
     );
   const url = new URL(file.uri);
-  if (url.protocol === "https:") return downloadState(url, intake, cancel);
+  if (url.protocol === "https:")
+    return downloadState(url, intake, cancel, responses);
   if (url.protocol !== "file:")
     return err(invalidParams("a file uri is a file:// or https:// URL"));
   const path = localFilePath(url);
@@ -346,6 +356,7 @@ export const partDocument = async (
   intake: StateIntake,
   cancel: AbortSignal,
   files: ArtifactResourceScope,
+  responses: StateResponseCleanup,
 ): Promise<Result<unknown, RpcError>> => {
   if (isRecord(part) && part.kind === "data" && isRecord(part.data))
     return ok(part.data);
@@ -362,6 +373,6 @@ export const partDocument = async (
       mediaTypeEssence(mimeType) !== "application/json")
   )
     return err(invalidParams("a file part's mimeType is application/json"));
-  const bytes = await fileBytes(part.file, intake, cancel, files);
+  const bytes = await fileBytes(part.file, intake, cancel, files, responses);
   return bytes.ok ? parseDocumentBytes(bytes.value) : bytes;
 };

src/wasm/WabtArtifactProvider.test.ts

@@ -1066,7 +1066,15 @@ it("enforces a selected command deadline and completes a corrected retry after c
   if (timed.ok) throw new Error("Selected 20 ms deadline was not enforced");
   expect(projectAnalysisError(timed.error)).toMatchObject({
     code: "provider_timeout",
-    details: { timeout_ms: 20 },
+    details: {
+      timeout_ms: 20,
+      limit: {
+        setting: "wasm.command_timeout_ms",
+        effective_value: 20,
+        source: "request",
+        update_mode: "parameter",
+      },
+    },
     remediation: { action: expect.stringContaining("command_timeout_ms") },
   });
   expect(timed.error.cleanupIncomplete).toBe(false);

src/wasm/WabtArtifactProvider.ts

@@ -229,6 +229,10 @@ export class WabtArtifactProvider implements WasmArtifactPort {
       input.command_timeout_ms ??
       options?.commandTimeoutMs ??
       DEFAULT_WASM_COMMAND_TIMEOUT_MS;
+    const commandTimeoutSource =
+      input.command_timeout_ms !== undefined
+        ? "request"
+        : options?.commandTimeoutSource;
     try {
       if (options?.signal?.aborted)
         throw new AnalysisCancelledError(OPERATION, {
@@ -695,6 +699,9 @@ export class WabtArtifactProvider implements WasmArtifactPort {
         phase === "artifact" ? artifactInput.path : input.path,
         {
           commandTimeoutMs,
+          ...(commandTimeoutSource === undefined
+            ? {}
+            : { commandTimeoutSource }),
           limits,
           inputField: artifactInput.field,
           observedOutput: capturedOutput,

src/wasm/WabtFailures.test.ts

@@ -52,7 +52,8 @@ it.each(["configuration", "producer"] as const)(
               limit: {
                 setting: "wasm.command_timeout_ms",
                 effective_value: 1234,
-                rea_restart_required: false,
+                update_mode: "runtime",
+                pending_restart: false,
               },
             },
             recovery: { kind: "inspect_timeout_state", timeout_ms: 1234 },
@@ -115,11 +116,30 @@ it.each([1234, 2_147_483_647])(
           parameter: "command_timeout_ms",
           applies_at: "next_operation",
           agent_editable: true,
-          rea_restart_required: false,
+          update_mode: "runtime",
+          pending_restart: false,
         },
       },
       recovery: { kind: "inspect_timeout_state", timeout_ms: commandTimeoutMs },
     });
+    expect(
+      projectAnalysisError(
+        wabtFailure(
+          new OwnedCommandFailure("timeout", "deadline"),
+          "producer",
+          "/module.wasm",
+          { commandTimeoutMs, commandTimeoutSource: "request" },
+        ),
+      ),
+    ).toMatchObject({
+      details: {
+        limit: {
+          source: "request",
+          update_mode: "parameter",
+          effective_value: commandTimeoutMs,
+        },
+      },
+    });
   },
 );
 

src/wasm/WabtFailures.ts

@@ -63,6 +63,12 @@ export type WabtVersionProbe = {
 
 type WabtFailureContext = {
   readonly commandTimeoutMs: number;
+  readonly commandTimeoutSource?:
+    | "default"
+    | "setting"
+    | "runtime"
+    | "request"
+    | undefined;
   readonly limits?: WasmResourceLimits;
   readonly versionProbe?: WabtVersionProbe;
   readonly inputField?: readonly (string | number)[];
@@ -182,13 +188,17 @@ const classifyWabtFailure = (
     // Classify once before cleanup takes precedence, so the primary remains actionable.
     const timeoutMs =
       execution?.commandTimeoutMs ?? DEFAULT_WASM_COMMAND_TIMEOUT_MS;
+    const timeoutSource =
+      execution?.commandTimeoutMs === undefined
+        ? "default"
+        : execution.commandTimeoutSource;
     const timeout =
       cause.reason === "timeout"
         ? new AnalysisTimeoutError("inspect_wasm_artifact", timeoutMs, {
             limit: analysisSettingLimit(
               "wasm.command_timeout_ms",
               timeoutMs,
-              undefined,
+              timeoutSource,
             ),
             ...options,
             // The cleanup wrapper retains output once; its nested primary needs only recovery facts.

tests/acceptance/analysisSettingsMcp.test.ts

@@ -148,7 +148,7 @@ it("advertises exact contracts and recovers a full large-index archive over one
     expect(updated.structuredContent).toMatchObject({
       revision: 1,
       active_work_changed: false,
-      rea_restart_required: false,
+      pending_restart: false,
     });
     const recovered = await invoke("inspect_artifact", {});
     expect(recovered.isError).not.toBe(true);

tests/acceptance/applications/javascriptChildProcessStdioCli.test.ts

@@ -15,6 +15,13 @@ const cases = [
     0,
   ],
   ['spawn("control", [], {stdio:"inherit"})', "control", "inherit", 0],
+  ['spawn("global-nan-env", [], {env:NaN})', "global-nan-env", "default", 0],
+  [
+    '((NaN) => spawn("shadowed-nan-env", [], {env:NaN}))({})',
+    "shadowed-nan-env",
+    "default",
+    0,
+  ],
   [
     'spawn("spread-after", [], {stdio:"pipe",...input})',
     "spread-after",
@@ -233,6 +240,12 @@ const exportAssignmentSource = `
   export = argv;
 `;
 
+const environmentCommands = [
+  "named-argv",
+  "execfile-dynamic",
+  "shadowed-nan-env",
+];
+
 // Callback/options ambiguity cannot change exec's effective stdio.
 const opaqueExecCallbackCommands = [
   "eval-callback",
@@ -342,7 +355,7 @@ cliTest.for(["CLI", "stdio MCP"] as const)(
       expect(child.properties.stdio_mode, String(command)).toBe(mode);
       expect(child.properties.argv_count, String(command)).toBe(count);
       expect(child.properties.environment_supplied, String(command)).toBe(
-        ["named-argv", "execfile-dynamic"].includes(String(command)),
+        environmentCommands.includes(String(command)),
       );
       expect(child.identity.source_range).not.toBeNull();
       const stdio = graph.relations.filter(
@@ -377,9 +390,7 @@ cliTest.for(["CLI", "stdio MCP"] as const)(
             relation.relation === "supplies-env" &&
             relation.target_node_id === child.node_id,
         ),
-      ).toHaveLength(
-        ["named-argv", "execfile-dynamic"].includes(String(command)) ? 1 : 0,
-      );
+      ).toHaveLength(environmentCommands.includes(String(command)) ? 1 : 0);
       expect(
         graph.relations.filter(
           (relation) =>

tests/acceptance/artifactReaderLimitsCli.test.ts

@@ -186,7 +186,8 @@ for (const reader of readers)
           effective_value: reader.value,
           applies_at: "next_operation",
           agent_editable: true,
-          rea_restart_required: false,
+          update_mode: "parameter",
+          pending_restart: false,
         },
       },
     });

tests/acceptance/tools/historicalCaptureResources.test.ts

@@ -87,7 +87,8 @@ it.skipIf(process.versions.bun !== undefined && jobLimit === null)(
             setting: "capture.decoder_old_generation_mib",
             effective_value: testHeap.oldGenerationMiB,
             source: testHeap.source,
-            rea_restart_required: false,
+            update_mode: "runtime",
+            pending_restart: false,
           },
           reported_limits:
             jobLimit === null
@@ -147,7 +148,8 @@ it.skipIf(process.versions.bun !== undefined && jobLimit === null)(
           setting: "capture.decoder_old_generation_mib",
           effective_value: testHeap.oldGenerationMiB,
           source: testHeap.source,
-          rea_restart_required: false,
+          update_mode: "runtime",
+          pending_restart: false,
         },
       },
       remediation: {

tests/boundary/cli/cutterVerifierPlatform.test.ts

@@ -1,81 +1,72 @@
-import { execFile } from "node:child_process";
-import { readFile, readdir, writeFile } from "node:fs/promises";
+import { readFile, readdir } from "node:fs/promises";
 import { join } from "node:path";
-import { promisify } from "node:util";
-import { expect, it } from "vitest";
+import { expect } from "vitest";
 import { parse } from "yaml";
+import { resolveSwiftcCompileTimeoutMs } from "../../../src/process/DarwinProcessRunTokenReader.js";
 import { createTestTempDirectory } from "../../fixtures/temporaryDirectory.js";
+import { processTest } from "../../support/process/processFixture.js";
 
-const execute = promisify(execFile);
+const hostLabels: Partial<Record<NodeJS.Platform, string>> = {
+  linux: "linux",
+  darwin: "macOS",
+  win32: "Windows",
+};
+const hostLabel = hostLabels[process.platform];
+const preparationBudget =
+  process.platform === "darwin" ? resolveSwiftcCompileTimeoutMs() : 0;
 
-it("enforces workflow platform labels before contacting Cutter", async () => {
-  const workflow = parse(
-    await readFile(".github/workflows/real-cutter.yml", "utf8"),
-  );
-  const labels = workflow.jobs["cutter-upstream"].strategy.matrix.include.map(
-    (entry: { platform: string }) => entry.platform,
-  );
-  expect(labels).toEqual(["linux", "macOS", "Windows"]);
-  const root = await createTestTempDirectory("rea-cutter-platform-");
-  const preload = join(root, "preload.mjs");
-  await writeFile(
-    preload,
-    `
-import childProcess from 'node:child_process';
-import { syncBuiltinESMExports } from 'node:module';
-Object.defineProperty(process, 'platform', { value: process.env.REA_PROBE_PLATFORM });
-childProcess.execFile = () => { throw new Error('fixture discovery reached'); };
-syncBuiltinESMExports();
-`,
-  );
-  const check = async (selected: string | undefined, actual: string) => {
-    const env: NodeJS.ProcessEnv = {
-      ...process.env,
-      TMPDIR: root,
-      TMP: root,
-      TEMP: root,
-      REA_PROBE_PLATFORM: actual,
+processTest.skipIf(hostLabel === undefined)(
+  "enforces workflow platform labels before contacting Cutter",
+  { timeout: Math.min(2_147_483_647, 2 * preparationBudget + 30_000) },
+  async ({ processes }) => {
+    const workflow = parse(
+      await readFile(".github/workflows/real-cutter.yml", "utf8"),
+    );
+    const labels = workflow.jobs["cutter-upstream"].strategy.matrix.include.map(
+      (entry: { platform: string }) => entry.platform,
+    );
+    expect(labels).toEqual(["linux", "macOS", "Windows"]);
+    const root = await createTestTempDirectory("rea-cutter-platform-");
+    // A private absent directory proves actual discovery without contacting a user's bridge.
+    const bridgeDirectory = join(root, "absent-bridge");
+    const check = async (selected: string | undefined) => {
+      const env: NodeJS.ProcessEnv = {
+        ...process.env,
+        TMPDIR: root,
+        TMP: root,
+        TEMP: root,
+        REA_CUTTER_BRIDGE_DIR: bridgeDirectory,
+      };
+      if (selected === undefined) delete env.EXPECTED_PLATFORM;
+      else env.EXPECTED_PLATFORM = selected;
+      const result = await processes.run(
+        process.execPath,
+        ["scripts/verify-real-cutter.mjs"],
+        {
+          env,
+          timeoutMs: Math.min(2_147_483_647, preparationBudget + 15_000),
+        },
+      );
+      expect(result.exitCode, result.stderr).toBe(1);
+      return result.stderr;
     };
-    if (selected === undefined) delete env.EXPECTED_PLATFORM;
-    else env.EXPECTED_PLATFORM = selected;
-    const result: unknown = await execute(
-      process.execPath,
-      ["--import", preload, "scripts/verify-real-cutter.mjs"],
-      { env, timeout: 10_000 },
-    ).then(
-      () => {
-        throw new Error("Fixture discovery should stop the verifier");
-      },
-      (error: unknown) => error,
+    for (const selected of [hostLabel, undefined]) {
+      const stderr = await check(selected);
+      expect(stderr).toContain("Open upstream Cutter with the REA plugin");
+      expect(stderr).toContain("no_live_bridge_found");
+    }
+    const other = labels.find((label: string) => label !== hostLabel);
+    expect(other).toBeDefined();
+    expect(await check(other)).toContain(
+      "Cutter runner did not match its declared platform",
     );
-    if (
-      !(result instanceof Error) ||
-      !("stderr" in result) ||
-      typeof result.stderr !== "string"
-    )
-      throw new Error("Missing verifier diagnostics");
-    return result.stderr;
-  };
-  for (const [selected, actual] of [
-    ["linux", "linux"],
-    ["macOS", "darwin"],
-    ["Windows", "win32"],
-  ] as const)
-    expect(await check(selected, actual)).toContain(
-      "fixture discovery reached",
+    expect(await check("unknown")).toContain(
+      "Unrecognized Cutter runner platform",
     );
-  expect(await check("linux", "darwin")).toContain(
-    "Cutter runner did not match its declared platform",
-  );
-  expect(await check("unknown", "linux")).toContain(
-    "Unrecognized Cutter runner platform",
-  );
-  expect(await check(undefined, "linux")).toContain(
-    "fixture discovery reached",
-  );
-  expect(
-    (await readdir(root)).filter((name) =>
-      name.startsWith("rea-cutter-smoke-"),
-    ),
-  ).toEqual([]);
-});
+    expect(
+      (await readdir(root)).filter((name) =>
+        name.startsWith("rea-cutter-smoke-"),
+      ),
+    ).toEqual([]);
+  },
+);

tests/boundary/filesystem/asarHeaderAdmission.test.ts

@@ -216,7 +216,8 @@ it("names the default header admission budget and its setting when source select
         applies_at: "next_operation",
         agent_editable: true,
         scope: "rea-runtime",
-        rea_restart_required: false,
+        update_mode: "runtime",
+        pending_restart: false,
       },
     },
   });

tests/boundary/filesystem/ciChangeScope.test.ts

@@ -783,7 +783,7 @@ it.each([
     { code: false, ghidra_bridge: true, runtime: false, package: false },
   ],
   [
-    "tests/conformance/ghidra/ReaArmContextProbe.java",
+    "tests/conformance/ghidra/ReaInstructionContextProbe.java",
     { ghidra_bridge: true, runtime: false, package: false },
   ],
   [

tests/boundary/filesystem/dotnetBundleArtifacts.test.ts

@@ -66,7 +66,8 @@ it.each([
           required_value: needed,
           applies_at: "next_operation",
           agent_editable: true,
-          rea_restart_required: false,
+          update_mode: "parameter",
+          pending_restart: false,
         },
       });
       const recovered = await readDotnetBundle(

tests/boundary/mcp/contractPresentation.test.ts

@@ -36,7 +36,9 @@ it("preserves full advertised schemas and handler meaning when presentation orde
       REA_MCP_INPUT_SCHEMA_PROFILE: "full",
       REA_MCP_OUTPUT_SCHEMA_PROFILE: "full",
     },
-    timeout: 20_000,
+    // The child advertises the full catalog and calls every browser, Electron,
+    // and JavaScript example. A busy runner exceeded 20s and was killed.
+    timeout: 60_000,
     maxBuffer: 16 * 1024 * 1024,
   });
   const result = resultSchema.parse(JSON.parse(stdout));

tests/boundary/mcp/ghidraDecompileDeadlineLifetime.test.ts

@@ -72,7 +72,8 @@ const enforcedLimit = {
   minimum: 1,
   maximum: 2_147_483,
   agent_editable: true,
-  rea_restart_required: false,
+  update_mode: "runtime",
+  pending_restart: false,
 };
 /** A finished invocation can act only through a new invocation's launch value. */
 const invocationLimit = {