@openwop/openwop-conformance 2.44.2 → 2.44.4

package/CHANGELOG.md

@@ -1,5 +1,22 @@
 # `@openwop/openwop-conformance` Changelog
 
+## [2.44.4] — 2026-09-29 — six more core families get v2 normative homes, replay derives a rejection and counts terminals, and the replay fixtures wait on a pure hold node
+
+- **`0223.replay-derives-rejection`: replay derives a rejection, never re-decides it** (#1756). Two new legs in `v2-approval-reject-disposition` (target major 2), gated on `replay` advertising mode `replay` and on each fixture. The leg forks `mode: "replay"` at the gate's recorded `interrupt.resolved`, so the prefix holds the request and suspension and the fork must supply the resolution. It asserts four things. (i) No new `interrupt.requested` / `approval.requested` for the gate at or after `fromSeq`; this is the discriminator. (ii) The fork's `interrupt.resolved` carries the source's `decision`, `action` and `reason`. (iii) The fork fails the same way: `approval_rejected`, `failedNodeId: gate`, a non-retryable `node.failed`. (iv) Leg A only: a resolve answers `409 interrupt_already_resolved`. Leg B uses `conformance-approval-timeout-approve` and compares the fork's `resolvedBy` with the source's (a host subject or absent, RFC 0183), rather than requiring it absent. `interruptId` equality is not asserted. Forking at the gate's `node.started` was rejected: a host may re-emit the recorded request as history, which cannot be told from a new one. Measured on the v2 reference host (openwop-examples `b6302cd`): both legs `executed-pass`. With its replay derivation disabled, both fail.
+- **The loop fixtures are bounded by their scenarios, not a host cap** (#1748). `conformance-approval-reject-loopback` and `conformance-replay-ordinal-loop` no longer claim to be bounded by `settings.maxLoopbackIterations`, which no spec binds. `v2-approval-reject-disposition`'s loopback leg and `v2-replay-suppression-ordinal` create the run with `configurable: { version: 1, run: { recursionLimit: 20 } }` as a runaway guard. A cycle-running host no longer needs a definition-level loopback cap to advertise either fixture.
+- **`core.conformance.hold` replaces `core.delay` in the replay fixtures** (#1769). A new conformance-reserved PURE node (`fixtures.md`): it completes after `inputs.delayMs` with its inputs as outputs, is never classified side-effecting, and re-executes live in a replay. `conformance-replay-ordinal-loop` and the v1 `conformance-replay-side-effect` now wait on it. With `core.delay`, which the registry's flow pack declares `role: "side-effect"`, a host honouring `replay.md` rule 4 failed the wait with `replay_source_missing` before the node under test. `v2-replay-suppression-ordinal`'s docstring follows rule 2's re-correction: the ordinal counts recorded terminals, not `node.started`.
+- **No leg change for #1770** (`interrupt.md` §Events: the kind-specific interrupt types are now a MAY). The suite never asserted their emission. `spec-artifacts` carries the corrected `interrupt.md`.
+- **`webhook-signed-delivery`'s tunnel control registered the tunnel, not loopback, and failed conformant hosts.** The control leg ("with a tunnel wired, the host still refuses the loopback receiver") posted `receiver.url`, commented as "deliberately the raw loopback URL, NOT the tunnel". Since 2.37.0, `startScopedReceiver` returns the PUBLIC FRONT as `url` whenever `OPENWOP_WEBHOOK_RECEIVER_URL` is set: the tunnel, plus the exercise's nonce. So on every tunnelled cut the control registered a legitimate public https destination, a host with a working SSRF guard accepted it, and the row recorded `executed-fail` with "the guard is absent (a finding)". Measured on openwop-app production (2026-09-28, suite 2.43.1): the control failed while the same host answers `https://127.0.0.1:…` and `https://localhost:…` with `400 webhook_url_rejected` ("denied (loopback / link-local / private-IP)"). The control now registers `receiver.localUrl` rewritten to `https:`, so a refusal can only come from the loopback deny list and not from a scheme check. Untunnelled runs are unaffected, because the control soft-skips without a tunnel.
+- **The 2.44.4 cycle opens.** Six more core families get a v2 normative home (RFC 0189), which changes the packed `@openwop/spec-artifacts` tree: `memory`, `prompts` and `aiProviders` in `host-services.md`; `limits` in `runs.md`; `artifactTypes` in a new `artifact-type-packs.md`, with field-level rules in `schemas/v2/artifact-type-pack-manifest.schema.json` descriptions. v1-dependent core families: 8 → 3 (`agents`, `budget`, `production`, each blocked on error codes v2 does not register). No scenario changes.
+
+## [2.44.3] — 2026-09-28 — fifteen more core families get v2 normative homes, and webhook retry rows record the wait path they measured
+
+- **Webhook retry rows say which wait path ran and what they measured** (RFC 0225 witness). A passing dead-letter row carried no detail, so a bundle could not show whether its wait came from the host's advertised `retryPolicy.maxElapsedMs`, an operator-raised cap or the 20 s floor, or when the dead-letter arrived. MyndHyve's 2.44.1 cut was the case: it advertises 600000, and it could not be counted as RFC 0225's witness. Rows driven by `retryWaitFor` now carry an informational detail, on passes too. `0173.webhook-durable-delivery.dead-letter` and `0188.dead-letter-content-free` record, for example, `observed: advertised-bound (maxElapsedMs 600000): waited ≤630000ms; 5 attempt(s), last attempt after 225031ms, sink after 231412ms`. The last attempt is a lower bound on the dead-letter time, and the sink read an upper bound. The retry leg, `v2-webhook-unregister-stops-delivery` and `v2-webhook-message-id-stable` record the path and window. The detail is prefixed `observed: `, never `partial-witness: `, so `check-accepted-predicate` counts the row as a full witness. The bundle already carries `detail` on every row inside the witness digest, so the note is signed. New `lib/row-observation.ts`, attached in `setup.ts`; `retryWaitSelection` / `waitObservation` in `lib/webhook-retry-window.ts`; server-free tests in `row-observation.test.ts` and `webhook-retry-window.test.ts`, each sabotage-checked.
+- **The 2.44.3 cycle opens.** Fifteen core families get a v2 normative home (RFC 0189), which changes the packed `@openwop/spec-artifacts` tree. v1-dependent core families fall from 23 to 8. No scenario changes: the rules are restated, none added.
+  - `events.md`: `feedback`, `providerUsage`, `envelopes`, and the rest of `heartbeat`.
+  - `host-services.md`: `secrets`, `modelCapabilities`, `scheduling`, `queueBus`, `toolHooks`, `httpClient`; `runs.md`: `deadLetter`.
+  - New `execution.md`: `selfHostedRunner`, `subWorkflow`, `multiAgent` (the family claim and `executionModel` facet); `identity.md` §1.5: `anonymousActor`.
+
 ## [2.44.2] — 2026-09-28 — RFC 0215's no-head-of-line leg does not convict a late fan-out, and says when it sampled each count
 
 - **The 2.44.2 cycle opens.** A patch: the requirement is unchanged, and one verdict branch that convicted without the contention it names becomes unjudged. `@openwop/spec-artifacts` moves in lockstep at the same exact pin.

package/README.md

@@ -11,7 +11,7 @@
 # --legacy-peer-deps is REQUIRED, not optional: the exact peer pin is what npm's
 # default resolver refuses. npm 10.9 fails outright with
 # "Cannot read properties of null (reading 'edgesOut')" — use npm >= 11.
-npm install --legacy-peer-deps @openwop/openwop-conformance@2.44.2 @openwop/spec-artifacts@2.44.2
+npm install --legacy-peer-deps @openwop/openwop-conformance@2.44.3 @openwop/spec-artifacts@2.44.3
 # or run without install:
 npx @openwop/openwop-conformance --base-url https://api.example.com --api-key hk_test_...
 ```

package/dist/spec-artifacts.lock.json

@@ -1,5 +1,5 @@
 {
   "package": "@openwop/spec-artifacts",
-  "version": "2.44.2",
-  "stampSha256": "66d1059ab9f4e5799bbb84c8b66e64033bc08f4daf6016c22c71164cca948c4c"
+  "version": "2.44.4",
+  "stampSha256": "0075c7ff983e3ad3d67935b7bb2287d624fee7b6517ccea44f89071741bdda9b"
 }

package/fixtures/conformance-approval-reject-loopback.json

@@ -2,7 +2,7 @@
   "id": "conformance-approval-reject-loopback",
   "name": "Conformance: Approval (reject loops back)",
   "version": "1.0",
-  "description": "openwop#1697. A rejected approval gate routed over an any_failed edge to revise, which loops back to the gate (bounded by settings.maxLoopbackIterations 2). Rejected once, the gate is visited again: the host MUST raise a new interrupt.requested with a different key and wait for a new decision, never replaying the first rejection. Conformance scenario: v2-approval-reject-disposition.test.ts.",
+  "description": "openwop#1697. A rejected approval gate routed over an any_failed edge to revise, which loops back to the gate (the loop is bounded by the scenario, a single reject, not by the host; the scenario passes configurable.run.recursionLimit 20 as a runaway guard; settings.maxLoopbackIterations is authoring metadata, openwop#1748). Rejected once, the gate is visited again: the host MUST raise a new interrupt.requested with a different key and wait for a new decision, never replaying the first rejection. Conformance scenario: v2-approval-reject-disposition.test.ts.",
   "nodes": [
     { "id": "start", "typeId": "core.noop", "name": "Start", "position": { "x": 0, "y": 0 }, "config": {}, "inputs": {} },
     {

package/fixtures/conformance-replay-ordinal-loop.json

@@ -2,13 +2,13 @@
   "id": "conformance-replay-ordinal-loop",
   "name": "Conformance: Replay outcome keyed on the execution ordinal",
   "version": "1.0",
-  "description": "openwop#1718. start → effect (a side-effecting node) → wait (core.delay) → back to effect, bounded by settings.maxLoopbackIterations 2, so effect executes twice. The suite cancels the source run inside wait, after effect's first execution completed and before its second, then forks mode:\"replay\" at effect's first node.started. In the fork, effect's first execution resolves from the recorded outcome for n = 1, and its second execution has no recorded outcome for n = 2, so it MUST fail closed with replay_source_missing (replay.md §Suppression rule 2). A host keying on (nodeId, attempt) or selecting the latest outcome resolves the second execution from the first and completes it. core.conformance.side-effect is the reserved side-effecting typeId (see conformance-replay-side-effect). Advertised only by a host that runs cycles and suppresses side effects in replay.",
+  "description": "openwop#1718. start → effect (a side-effecting node) → wait (core.conformance.hold, a reserved PURE node) → back to effect, bounded by the scenario (cancel plus fail-closed), not by the host; the scenario passes configurable.run.recursionLimit 20 as a runaway guard, and settings.maxLoopbackIterations is authoring metadata (openwop#1748), so effect executes twice. The suite cancels the source run inside wait, after effect's first execution completed and before its second, then forks mode:\"replay\" at effect's first node.started. In the fork, effect's first execution resolves from the recorded outcome for n = 1, and its second execution has no recorded outcome for n = 2, so it MUST fail closed with replay_source_missing (replay.md §Suppression rule 2). A host keying on (nodeId, attempt) or selecting the latest outcome resolves the second execution from the first and completes it. core.conformance.side-effect is the reserved side-effecting typeId (see conformance-replay-side-effect). Advertised only by a host that runs cycles and suppresses side effects in replay.",
   "nodes": [
     { "id": "start", "typeId": "core.noop", "name": "Start", "position": { "x": 0, "y": 0 }, "config": {}, "inputs": {} },
     { "id": "effect", "typeId": "core.conformance.side-effect", "name": "Side Effect", "position": { "x": 200, "y": 0 }, "config": {}, "inputs": {} },
     {
       "id": "wait",
-      "typeId": "core.delay",
+      "typeId": "core.conformance.hold",
       "name": "Wait",
       "position": { "x": 400, "y": 0 },
       "config": {},

package/fixtures/conformance-replay-side-effect.json

@@ -2,11 +2,11 @@
   "id": "conformance-replay-side-effect",
   "name": "Conformance: Replay Side-Effect Suppression",
   "version": "1.0",
-  "description": "RFC 0140 fixture for `replay-side-effect-suppression.test.ts`. A `core.delay` gate followed by a node the host classifies as SIDE-EFFECTING. The test starts a run, cancels it mid-delay so the effect node never reaches a terminal outcome, then forks `mode:\"replay\"` — a host advertising `replay.sideEffectSuppression: \"recorded-outcome\"` MUST fail that node closed with `replay_source_missing` (replay.md §\"Side-effect suppression in replay\" requirement 3) rather than executing it. `core.conformance.side-effect` is a conformance-RESERVED typeId (same pattern as `core.conformance.mock-agent`): a host advertising the capability MUST map it to a node its replay classifier treats as side-effecting. The node's actual effect is irrelevant to the assertion — what matters is that the host would refuse to perform it during a replay. Hosts that do not advertise the capability need not supply this fixture; the scenario skips.",
+  "description": "RFC 0140 fixture for `replay-side-effect-suppression.test.ts`. A `core.conformance.hold` wait (a conformance-reserved PURE node that re-executes live in a replay; it replaced `core.delay` on 2026-09-29, openwop#1769, because the registry declares `core.delay` side-effecting) followed by a node the host classifies as SIDE-EFFECTING. The test starts a run, cancels it mid-delay so the effect node never reaches a terminal outcome, then forks `mode:\"replay\"` — a host advertising `replay.sideEffectSuppression: \"recorded-outcome\"` MUST fail that node closed with `replay_source_missing` (replay.md §\"Side-effect suppression in replay\" requirement 3) rather than executing it. `core.conformance.side-effect` is a conformance-RESERVED typeId (same pattern as `core.conformance.mock-agent`): a host advertising the capability MUST map it to a node its replay classifier treats as side-effecting. The node's actual effect is irrelevant to the assertion — what matters is that the host would refuse to perform it during a replay. Hosts that do not advertise the capability need not supply this fixture; the scenario skips.",
   "nodes": [
     {
       "id": "wait",
-      "typeId": "core.delay",
+      "typeId": "core.conformance.hold",
       "name": "Wait",
       "position": {
         "x": 0,

package/fixtures.md

@@ -255,7 +255,7 @@ The `messages`-mode stream fixture (AI token streaming) is covered by the determ
 
 - **Purpose**: witness that `key` is per visit (`spec/v2/core/interrupt.md` §Re-entry and resume values, openwop#1697).
 - **Inputs**: none.
-- **Graph**: `start` (`core.noop`) → `gate` (`any_success`); `gate` → `revise` (`core.noop`, `any_failed`); `revise` → `gate` (`any_success`). `settings.maxLoopbackIterations: 2`.
+- **Graph**: `start` (`core.noop`) → `gate` (`any_success`); `gate` → `revise` (`core.noop`, `any_failed`); `revise` → `gate` (`any_success`). The loop is bounded by the scenario (a single reject; cancel plus fail-closed), not by the host. The scenario creates the run with `configurable: { version: 1, run: { recursionLimit: 20 } }` as a runaway guard (`runs.md` §run section). `settings.maxLoopbackIterations: 2` stays in the JSON as authoring metadata; the scenario does not depend on it (openwop#1748).
 - **Behavior**:
   1. Run reaches `gate` and MUST be `waiting-approval`.
   2. Client POSTs `{action: 'reject'}`. `gate` fails with `approval_rejected`, and `revise` runs.
@@ -306,8 +306,8 @@ The `messages`-mode stream fixture (AI token streaming) is covered by the determ
 
 ### `conformance-replay-ordinal-loop`
 
-- **Purpose**: witness that a replay fork keys a side-effecting node's recorded outcome on `(sourceRunId, nodeId, n)`, where `n` counts the node's `node.started` events through this execution (`spec/v2/core/replay.md` §Suppression rule 2, openwop#1718).
-- **Shape**: `start` (`core.noop`) → `effect` (`core.conformance.side-effect`, `any_success`) → `wait` (`core.delay`) → `effect` (`any_success`); `settings.maxLoopbackIterations: 2`, so `effect` executes twice.
+- **Purpose**: witness that a replay fork keys a side-effecting node's recorded outcome on `(sourceRunId, nodeId, n)`, where `n` is one more than the node's recorded terminals (`node.completed`, `node.failed`) before this execution (`spec/v2/core/replay.md` §Suppression rule 2, openwop#1718, re-corrected by openwop#1769).
+- **Shape**: `start` (`core.noop`) → `effect` (`core.conformance.side-effect`, `any_success`) → `wait` (`core.conformance.hold`) → `effect` (`any_success`), so `effect` executes twice. The loop is bounded by the scenario (a single reject; cancel plus fail-closed), not by the host. The scenario creates the run with `configurable: { version: 1, run: { recursionLimit: 20 } }` as a runaway guard (`runs.md` §run section). `settings.maxLoopbackIterations: 2` stays in the JSON as authoring metadata; the scenario does not depend on it (openwop#1748).
 - **Inputs**: `delayMs` (integer, default 3000), long enough to cancel inside `wait`.
 - **Expected behavior**:
   1. The source run completes `effect`'s first execution. The client cancels it inside `wait`, before the second execution.
@@ -318,7 +318,7 @@ The `messages`-mode stream fixture (AI token streaming) is covered by the determ
 ### `conformance-replay-side-effect`
 
 - **Purpose**: verify RFC 0140 — a `mode:"replay"` fork MUST NOT re-perform a side-effecting node's effect.
-- **Shape**: `core.delay` (`wait`) → `core.conformance.side-effect` (`effect`).
+- **Shape**: `core.conformance.hold` (`wait`) → `core.conformance.side-effect` (`effect`). *(Corrected 2026-09-29, openwop#1769: `wait` was `core.delay`, which the registry's flow pack declares `role: "side-effect"`; a host honouring that floor resolved `wait` from the cancelled source and failed it with `replay_source_missing` before `effect` was reached.)*
 - **`core.conformance.side-effect` is a conformance-RESERVED typeId** (same pattern as `core.conformance.mock-agent`): a host advertising `replay.sideEffectSuppression: "recorded-outcome"` MUST map it to a node its replay classifier treats as side-effecting. The node's actual effect is irrelevant to the assertion — what matters is that the host would refuse to perform it during a replay. Hosts that do not advertise the capability need not supply this fixture; the scenario skips.
 - **Inputs**:
   - `delayMs` (integer, required, 1 ≤ value ≤ 60000) — long enough for the test to cancel before `effect` runs.
@@ -590,6 +590,7 @@ The fixtures reference these typeIds:
 | `core.delay`                  | delay, cancellable                                     | Sleep `config.delayMs` ms                                                                                                                                                                |
 | `core.fail`                   | failure                                                | Throw with `code: "example.conformance_failure"` (a vendor code under the registered `example` org; errors.md §The registry, openwop#1698), message: "Intentional conformance failure"                                                                                                |
 | `core.approvalGate`           | approval                                               | Call `ctx.interrupt({kind: 'approval', ...})`                                                                                                                                            |
+| `core.conformance.hold` | replay-side-effect, replay-ordinal-loop | Conformance-RESERVED (openwop#1769). Completes after `inputs.delayMs` with its inputs as outputs and performs nothing observable outside the run's own event log. A host advertising a fixture that uses it MUST NOT classify it side-effecting (`replay.md` §Suppression rule 1) and MUST re-execute it live in a replay. |
 | `core.clarificationGate`      | clarification                                          | Call `ctx.interrupt({kind: 'clarification', ...})`                                                                                                                                       |
 | `conformance.requiresMissing` | capability-missing                                     | Declares `requires: ['conformance.never-provided']`; engine MUST refuse dispatch. Opt-in fixture registration is recommended so production deployments don't expose the fixture surface. |
 | `conformance.artifact.emit` | artifact-emit | Opt-in (RFC 0205). Produce one artifact of `config.artifactType` with payload `config.data`; emit `artifact.created` naming it; readable through `getArtifact`. |

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@openwop/openwop-conformance",
-  "version": "2.44.2",
+  "version": "2.44.4",
   "description": "Production-ready black-box conformance suite for OpenWOP v1.0 compliant servers.",
   "repository": {
     "type": "git",
@@ -56,6 +56,6 @@
     "@openwop/spec-artifacts": "file:../spec-artifacts"
   },
   "peerDependencies": {
-    "@openwop/spec-artifacts": "2.44.2"
+    "@openwop/spec-artifacts": "2.44.4"
   }
 }

package/requirements.json

@@ -3,8 +3,8 @@
   "generatedFrom": "src/scenarios/*.test.ts",
   "counts": {
     "files": 648,
-    "tests": 2490,
-    "withStableId": 2490,
+    "tests": 2492,
+    "withStableId": 2492,
     "interpolatedTitles": 0,
     "explicitIds": 2372
   },
@@ -22926,7 +22926,7 @@
     {
       "id": "openwop.it.replay-side-effect-suppression.sideeffectsuppression-when-present-is-one-of-the-two-defined-values",
       "file": "replay-side-effect-suppression.test.ts",
-      "line": 101,
+      "line": 103,
       "title": "sideEffectSuppression, when present, is one of the two defined values",
       "explicitId": "openwop.it.replay-side-effect-suppression.sideeffectsuppression-when-present-is-one-of-the-two-defined-values",
       "citations": [
@@ -22939,7 +22939,7 @@
     {
       "id": "openwop.it.replay-side-effect-suppression.fails-a-side-effecting-node-closed-with-replay-source-missing-when-the-source-ne",
       "file": "replay-side-effect-suppression.test.ts",
-      "line": 115,
+      "line": 117,
       "title": "fails a side-effecting node CLOSED with replay_source_missing when the source never recorded it",
       "explicitId": "openwop.it.replay-side-effect-suppression.fails-a-side-effecting-node-closed-with-replay-source-missing-when-the-source-ne",
       "citations": [
@@ -22974,7 +22974,7 @@
     {
       "id": "openwop.it.replay-side-effect-suppression.the-manifest-still-offers-the-role-value-the-requirement-binds",
       "file": "replay-side-effect-suppression.test.ts",
-      "line": 240,
+      "line": 242,
       "title": "the manifest still offers the role value the requirement binds",
       "explicitId": "openwop.it.replay-side-effect-suppression.the-manifest-still-offers-the-role-value-the-requirement-binds",
       "citations": [
@@ -22988,7 +22988,7 @@
     {
       "id": "openwop.it.replay-side-effect-suppression.replay-md-binds-that-declaration-naming-the-schema-it-comes-from",
       "file": "replay-side-effect-suppression.test.ts",
-      "line": 251,
+      "line": 253,
       "title": "replay.md binds that declaration, naming the schema it comes from",
       "explicitId": "openwop.it.replay-side-effect-suppression.replay-md-binds-that-declaration-naming-the-schema-it-comes-from",
       "citations": [
@@ -23002,7 +23002,7 @@
     {
       "id": "openwop.it.replay-side-effect-suppression.a-throw-is-named-a-backstop-not-a-discharge",
       "file": "replay-side-effect-suppression.test.ts",
-      "line": 261,
+      "line": 263,
       "title": "a throw is named a backstop, not a discharge",
       "explicitId": "openwop.it.replay-side-effect-suppression.a-throw-is-named-a-backstop-not-a-discharge",
       "citations": [
@@ -23016,7 +23016,7 @@
     {
       "id": "openwop.it.replay-side-effect-suppression.the-floor-direction-is-stated-not-left-to-inference",
       "file": "replay-side-effect-suppression.test.ts",
-      "line": 285,
+      "line": 287,
       "title": "the floor direction is stated, not left to inference",
       "explicitId": "openwop.it.replay-side-effect-suppression.the-floor-direction-is-stated-not-left-to-inference",
       "citations": [
@@ -27856,7 +27856,7 @@
     {
       "id": "openwop.it.v2-approval-reject-disposition.a-rejected-gate-with-no-routing-edge-fails-the-run-with-approval-rejected",
       "file": "v2-approval-reject-disposition.test.ts",
-      "line": 126,
+      "line": 141,
       "title": "a rejected gate with no routing edge fails the run with approval_rejected",
       "explicitId": null,
       "citations": [
@@ -27879,7 +27879,7 @@
     {
       "id": "openwop.it.v2-approval-reject-disposition.interrupt-resolved-records-action-reject-and-decision-rejected",
       "file": "v2-approval-reject-disposition.test.ts",
-      "line": 137,
+      "line": 152,
       "title": "interrupt.resolved records action reject and decision rejected",
       "explicitId": null,
       "citations": [
@@ -27898,7 +27898,7 @@
     {
       "id": "openwop.it.v2-approval-reject-disposition.the-gate-node-fails-with-approval-rejected-not-retryable-and-run-failed-names-it",
       "file": "v2-approval-reject-disposition.test.ts",
-      "line": 150,
+      "line": 165,
       "title": "the gate node fails with approval_rejected, not retryable, and run.failed names it",
       "explicitId": null,
       "citations": [
@@ -27919,7 +27919,7 @@
     {
       "id": "openwop.it.v2-approval-reject-disposition.a-majority-reject-decides-the-quorum-gate-on-the-deciding-vote-and-fails-the-run",
       "file": "v2-approval-reject-disposition.test.ts",
-      "line": 163,
+      "line": 178,
       "title": "a majority reject decides the quorum gate on the deciding vote and fails the run with approval_rejected",
       "explicitId": null,
       "citations": [
@@ -27945,7 +27945,7 @@
     {
       "id": "openwop.it.v2-approval-reject-disposition.a-rejected-gate-routed-over-an-any-failed-edge-fails-as-a-node-its-all-success-s",
       "file": "v2-approval-reject-disposition.test.ts",
-      "line": 186,
+      "line": 201,
       "title": "a rejected gate routed over an any_failed edge fails as a node, its all_success sibling never runs, and the run completes",
       "explicitId": null,
       "citations": [
@@ -27971,7 +27971,7 @@
     {
       "id": "openwop.it.v2-approval-reject-disposition.an-unresolved-gate-past-its-timeoutms-with-no-ontimeout-is-resolved-rejected-by",
       "file": "v2-approval-reject-disposition.test.ts",
-      "line": 205,
+      "line": 220,
       "title": "an unresolved gate past its timeoutMs with no onTimeout is resolved rejected by the host and fails the run naming the gate",
       "explicitId": null,
       "citations": [
@@ -27992,7 +27992,7 @@
     {
       "id": "openwop.it.v2-approval-reject-disposition.a-gate-whose-ontimeout-is-approve-still-resolves-rejected-on-timeout-and-fails-t",
       "file": "v2-approval-reject-disposition.test.ts",
-      "line": 224,
+      "line": 239,
       "title": "a gate whose onTimeout is approve still resolves rejected on timeout and fails the run: a timeout never grants",
       "explicitId": null,
       "citations": [
@@ -28014,7 +28014,7 @@
     {
       "id": "openwop.it.v2-approval-reject-disposition.a-rejected-gate-looped-back-to-itself-is-asked-again-under-a-new-key-and-the-fir",
       "file": "v2-approval-reject-disposition.test.ts",
-      "line": 242,
+      "line": 257,
       "title": "a rejected gate looped back to itself is asked again under a new key, and the first rejection is not replayed",
       "explicitId": null,
       "citations": [
@@ -28037,6 +28037,33 @@
         }
       ]
     },
+    {
+      "id": "openwop.it.v2-approval-reject-disposition.a-replay-fork-of-a-decided-reject-derives-the-rejection-from-the-log-and-never-r",
+      "file": "v2-approval-reject-disposition.test.ts",
+      "line": 333,
+      "title": "a replay fork of a decided reject derives the rejection from the log and never re-decides it (leg A)",
+      "explicitId": null,
+      "citations": [
+        {
+          "section": "spec/v2/core/errors.md §One code per state",
+          "requirement": "a resolve against the replayed gate MUST answer 409 interrupt_already_resolved — the fork holds no open decision"
+        }
+      ]
+    },
+    {
+      "id": "openwop.it.v2-approval-reject-disposition.a-replay-fork-of-a-timed-out-gate-derives-the-timeout-rejection-from-the-log-and",
+      "file": "v2-approval-reject-disposition.test.ts",
+      "line": 348,
+      "title": "a replay fork of a timed-out gate derives the timeout rejection from the log and never re-arms it (leg B)",
+      "explicitId": null,
+      "citations": [
+        {
+          "section": null,
+          "requirement": "leg B: the derived resolution is the recorded timeout — action timeout, reason timeout, and the source's resolvedBy (a host subject or absent), never a caller",
+          "interpolated": true
+        }
+      ]
+    },
     {
       "id": "openwop.it.v2-approver-enforced.a-resolver-outside-approverslist-is-refused-at-resolve-time",
       "file": "v2-approver-enforced.test.ts",
@@ -33078,7 +33105,7 @@
     {
       "id": "openwop.it.v2-replay-suppression-ordinal.a-replay-fork-resolves-each-execution-of-a-looped-side-effecting-node-by-its-own",
       "file": "v2-replay-suppression-ordinal.test.ts",
-      "line": 73,
+      "line": 74,
       "title": "a replay fork resolves each execution of a looped side-effecting node by its own ordinal, failing closed where none was recorded",
       "explicitId": "openwop.requirement.replay.suppression-execution-ordinal",
       "citations": [
@@ -34355,7 +34382,7 @@
     {
       "id": "openwop.it.v2-webhook-durable-delivery.a-failed-attempt-is-retried-and-the-event-is-delivered-at-least-once",
       "file": "v2-webhook-durable-delivery.test.ts",
-      "line": 263,
+      "line": 279,
       "title": "a failed attempt is retried and the event is delivered at least once",
       "explicitId": "openwop.requirement.0173.webhook-durable-delivery",
       "citations": [
@@ -34392,7 +34419,7 @@
     {
       "id": "openwop.it.v2-webhook-durable-delivery.an-exhausted-delivery-is-dead-lettered-never-dropped",
       "file": "v2-webhook-durable-delivery.test.ts",
-      "line": 383,
+      "line": 400,
       "title": "an exhausted delivery is dead-lettered, never dropped",
       "explicitId": "openwop.requirement.0173.webhook-durable-delivery.dead-letter",
       "citations": [
@@ -34431,7 +34458,7 @@
     {
       "id": "openwop.it.v2-webhook-durable-delivery.the-dead-letter-read-is-served-and-its-records-carry-no-payload",
       "file": "v2-webhook-durable-delivery.test.ts",
-      "line": 491,
+      "line": 509,
       "title": "the dead-letter read is served and its records carry no payload",
       "explicitId": "openwop.requirement.0188.dead-letter-read",
       "citations": [
@@ -34444,7 +34471,7 @@
     {
       "id": "openwop.it.v2-webhook-durable-delivery.after-unregister-the-dead-letter-read-answers-404-not-found-as-for-a-subscriptio",
       "file": "v2-webhook-durable-delivery.test.ts",
-      "line": 512,
+      "line": 530,
       "title": "after unregister, the dead-letter read answers 404 not_found, as for a subscription that never existed",
       "explicitId": "openwop.requirement.0217.dead-letter-read-after-unregister",
       "citations": [
@@ -34471,7 +34498,7 @@
     {
       "id": "openwop.it.v2-webhook-durable-delivery.a-dead-letter-record-carries-no-delivered-payload",
       "file": "v2-webhook-durable-delivery.test.ts",
-      "line": 545,
+      "line": 563,
       "title": "a dead-letter record carries no delivered payload",
       "explicitId": "openwop.requirement.0188.dead-letter-content-free",
       "citations": [
@@ -34615,7 +34642,7 @@
     {
       "id": "openwop.it.v2-webhook-message-id-stable.every-attempt-of-one-delivery-carries-one-webhook-id-and-distinct-deliveries-car",
       "file": "v2-webhook-message-id-stable.test.ts",
-      "line": 69,
+      "line": 70,
       "title": "every attempt of one delivery carries one webhook-id, and distinct deliveries carry distinct ids",
       "explicitId": "openwop.requirement.0201.message-id-stable",
       "citations": [
@@ -34854,7 +34881,7 @@
     {
       "id": "openwop.it.v2-webhook-unregister-stops-delivery.no-attempt-for-an-unregistered-subscription-starts-after-the-204-while-its-contr",
       "file": "v2-webhook-unregister-stops-delivery.test.ts",
-      "line": 86,
+      "line": 87,
       "title": "no attempt for an unregistered subscription starts after the 204, while its control keeps being retried",
       "explicitId": "openwop.requirement.0215.unregister-stops-delivery",
       "citations": [

package/schemas/CORPUS-STAMP.json

@@ -1,17 +1,17 @@
 {
   "_comment": "Provenance of @openwop/spec-artifacts (RFC 0168 §D.2). files: SHA-256 per file; the conformance suite compares the installed peer against dist/spec-artifacts.lock.json at start.",
   "package": "@openwop/spec-artifacts",
-  "version": "2.44.2",
-  "corpusTag": "v2.44.2",
+  "version": "2.44.4",
+  "corpusTag": "v2.44.4",
   "files": {
     "api/.redocly.lint-ignore.yaml": "bf5a8350b88a72fa43f59605ed8d903ed24b6cfccda5e45509c9f6ed9ee4e712",
     "api/asyncapi.yaml": "d5ecb9ee6114582be3b1f662c84bfac9ae96dae7bacb853e461168f70a8e1c7d",
     "api/grpc/openwop.proto": "c3e72bb17cba514ee98feb6434e6c9b6ea6795bfd086489ec69fd882dd1ad977",
     "api/openapi.yaml": "69464bc0e1e71ef3b543a73d4fa4f67202343d2603c80af8033f6c3619ef1b30",
     "api/redocly.yaml": "b0604c89b2ca6d5076ec25725c539dad44a741a811fe524439ee6daef8baa09f",
-    "api/seams-v2.yaml": "f24127f3cd544f1c5a942859ffc492b438d37b1b7d7a660c631fb87f70cd24f6",
-    "api/v2/asyncapi.yaml": "3283ff1bbe962dc75e6d66d2b755d4eb7d53770ce8511f2d671b923d6ca0e51a",
-    "api/v2/openapi.yaml": "4a455ba22c81718e2842adce60e5d3c5970e6f8a55b301d9154ccc8060f6cf91",
+    "api/seams-v2.yaml": "0c5a5ce153d5ac87b89f331f1e13ac25389937f966adbd3207bdfcc17e7ef8f5",
+    "api/v2/asyncapi.yaml": "d195115d983cadc6ae08d617de1a1c2f09073b22b0e5309ddd9119319164c566",
+    "api/v2/openapi.yaml": "d928ce4c62122954da1fecd5a79baff89957e65d4cc54606868bed92127b70c6",
     "api/v2/redocly.yaml": "1e66b60e6118ad11a823bb620678be464d99dfe50a40e3e6f93ec9429b88b34c",
     "schemas/README.md": "0c0b737ffcf8f30e7d2809cec8a498232de710f41443212922ad8337cdde0b51",
     "schemas/a2a-task-state.schema.json": "75d5049dea8bd873ff0e7546f1c60c8d36c219bec7084264360be54a8be30ae1",
@@ -112,7 +112,7 @@
     "schemas/v2/ai-envelope.schema.json": "0dc7dfbf032a37239ffd81fc2d67f999a40782ff5074866a56e414164cdee809",
     "schemas/v2/annotation-create.schema.json": "ec32c22a243b343eeb6956a23be2d76ba5b50c8018ec3ec9d6efa5a1d9b0c792",
     "schemas/v2/annotation.schema.json": "eb4813cdaa8e325412cfcaeeee5c8a28f3751c946dec64e546c7ba0ad3aad8d4",
-    "schemas/v2/artifact-type-pack-manifest.schema.json": "c8820351a2ab93855bd654ad3d8a9788f97975d09b3a7a0418e374ed6114595d",
+    "schemas/v2/artifact-type-pack-manifest.schema.json": "acbbf14d32da4e0cb81254bbd6cc48fc10083890572044ddef71c59a0f65c185",
     "schemas/v2/artifact.schema.json": "85b9dc284d39b99d2a66fa8cbf47f2dd3d8526d3bd6cc411ad315b372de38975",
     "schemas/v2/audit-verify-result.schema.json": "14af8e9283510f0a40e48a2059c929ee44e0fc9fa3cc66a95bfbe90b17c48297",
     "schemas/v2/budget-policy.schema.json": "c7449daeb6e1d95e7a047b8c2814d3d54748697c27046972834874867d9b5f4c",
@@ -193,61 +193,63 @@
     "schemas/v2/webhook-delivery.schema.json": "4e8ab3b722ad1731adf0548c1217f0de62da40de14e8abf77533aee4ace4c079",
     "schemas/v2/webhook-verification.schema.json": "76e4e0d81143ffd2649bc7adba2aa34f2525e2c4e9f0300a425d908b3163196b",
     "schemas/v2/workflow-chain-pack-manifest.schema.json": "0816c6c59a91452d6fa1931394b766780eaf90949c9f7cd0ced522dadd32ff81",
-    "schemas/v2/workflow-definition.schema.json": "b9043e3413fc01d939e7b2892cc0fe28bdfd77d5bacc561b6ca4ae20197b985b",
+    "schemas/v2/workflow-definition.schema.json": "3f031c0cb1644ac21f203945a0860e04dcea4bd1ab722f280bdcdc788edd0a22",
     "schemas/v2/workload-identity.schema.json": "1f589debfac6d70e0908da511550ec69a27b4bcc8314551cd843b6d923605efc",
     "schemas/v2/workspace-file-create.schema.json": "b6b5260857318c3b67de0231ed15e3a746f525880a763e7b36a7854178dd10d0",
     "schemas/v2/workspace-file.schema.json": "d2ed39e8c229dde33b56905c8df6f8b2ff056c97e1c5f615e3124baa54a6254f",
     "schemas/workflow-chain-pack-manifest.schema.json": "c027571913c44d8d9678750c72701a710a02fa0df55e86849da03e13e95c333a",
-    "schemas/workflow-definition.schema.json": "28b050ffe22121796c9d4a5a8c7e57c901bb283bd08d3fd3ff2ea25a898108c1",
+    "schemas/workflow-definition.schema.json": "fbab3787707fc0e00ecf2adea8de8aba582e6057af9cde893b1b0eb969adea83",
     "schemas/workload-identity.schema.json": "64ed93e12cc9f0657e56ff9b8f045e5d68d397df5276d103c07f42c7f34f0579",
     "schemas/workspace-file-create.schema.json": "70a23b92c2958c0165d7d29aea9020a045dc7816af82c80c5f50d0aac19c4856",
     "schemas/workspace-file.schema.json": "464de85c2a068243084ee9c1d969bc7cd5d8f7948574e58450d6493c38a0e1e4",
     "spec/v1/alias-detectors.json": "29f21e0fb2c7a232a92a5c19b4cba3a86bc3b5e16f06780452f0421c66c6ec38",
     "spec/v1/capability-declaration-classes.json": "e7729aed5c4b4e1dd02abab0530f14cc95f5d4070fe51fb139e7f5cccefa00c6",
-    "spec/v1/core-standard-manifest.json": "3d7fd99cd02153094723c2bf383d757c282e653526aa3b199d9bdb4b9cfd7c80",
+    "spec/v1/core-standard-manifest.json": "8f01c8ce3e66e4ff879e28da7f0de8bb8e2f3f85e5855cae17f0563c902328e5",
     "spec/v1/deprecations.json": "46df807fb4adf0aae24671c20f69975cf3496e3bc3a521cb7e6ea556bdce563c",
     "spec/v1/deprecations.schema.json": "3e393c405d2a41b467d8c5e3c468549078df1ce6a6d2588fc488097b95d9b55b",
     "spec/v1/event-codemap.json": "3da60d884157793a360da532a9fcbbfb5285636db325a74cec94b34622186d97",
     "spec/v1/event-codemap.schema.json": "d05933b2e88103aff51a2f774df97b9bda065e0fc88dbd9b114cc47a32189174",
     "spec/v1/extensions.json": "79a60754aa16cbdbdb604f8e5c00038af535a1690a4d07bd5ff7c38f0256a7bc",
-    "spec/v1/gaps.json": "07b0f5a206a015336307dd343b75adec647f06a9bae59ee08dbbfac851c503d5",
+    "spec/v1/gaps.json": "11c35393ad666899fe8a668c45fa69f79d20361ab57e84949dce3bc2f528d5c3",
     "spec/v1/gaps.schema.json": "8fd83259f556553c9df0f53e7a82ca8c2a4771d471197f69b4e2ae0ceeacfd66",
-    "spec/v1/migrations.json": "2efcd9064b2bc65a84a3b3d281aa86a91efd002690c1f4f14e8335c4a18194d1",
+    "spec/v1/migrations.json": "ca57b7ce335a681aaa78573ce229e9c3d7d75859f03e087ae93bbfc889359127",
     "spec/v1/migrations.schema.json": "886779aa6c22e646db097f5df210adb018a4dd14a7b815465a18c8a7056c8f72",
     "spec/v1/operation-path-manifest.json": "fada0342815fb129c1ae61f1829bd6e34417717408ce86fcc54280f373c19e74",
     "spec/v1/spec-gaps.json": "ea9612af55bea9dae3e82223d4f714243f8f590b0193f1188257b13491a38afc",
-    "spec/v2/README.md": "377639de1dad1ed71aba6dae8d34fdb6ee954fb23106e89454fe8d29cdde6349",
+    "spec/v2/README.md": "ddea90d7cf14aefffe1a43eeeecd4222dadad1d1e5ac64dceb073990eb93d9ef",
+    "spec/v2/core/artifact-type-packs.md": "e143bfd6f98d4867855bb57090a6b854e98d4b0a6202b9b3734cb45366f99854",
     "spec/v2/core/capabilities.md": "ad728e8a761153b55d036ece55043ff62d0982121e2e629e5e7cdf560fca3c03",
     "spec/v2/core/conformance.md": "0cfceb926eed87055550a5c23e640210c795fc8cf9dd1b2544db0e2b3842c564",
     "spec/v2/core/connection-packs.md": "1ee129d792c537bfe82393ccb92555d8e6ddc342b1aa781a13f101b27ec38171",
     "spec/v2/core/conversation.md": "354fa57c73d11062e523a44aa75a6353c5c81aefaf306d30c42416a228c82c86",
     "spec/v2/core/errors.md": "6d5a67389de84556f81e2558a30d83959c9283c8c5481ce7c759021e486d84fd",
-    "spec/v2/core/events.md": "604abb6f9e2259a04103e1cb76d728712c3701c0d84dd915d4dc2949c912ee5e",
+    "spec/v2/core/events.md": "8a2767691f604b2adf76399edeeaccfc151c697fc9717dc17cf7dd57372a8c22",
+    "spec/v2/core/execution.md": "299c4d548e16e5a0809957a47cbd34b56a38de82b8c9b3a5d036b2c7abca09c6",
     "spec/v2/core/form-content-packs.md": "02931227899676fd4581a27017be2a6c4953904717be9b81c29d12f87ab3917d",
     "spec/v2/core/headers.md": "5cb1624fa63acc81992b6a6d842ec392cf7271aaf2f65d1f83c8c4fa286e6730",
-    "spec/v2/core/host-services.md": "8b920bf2e9335ab04452068429b0c37d6eb8872b380a4d53368f2600d4afa241",
+    "spec/v2/core/host-services.md": "aa8b269e6348bdcb797da1a23c82f9df95c2133a7b947253799f6eaf88ccd3f1",
     "spec/v2/core/i18n.md": "2d5a67bd3b9b956d1144acab2fbc3646cdb50b3621f97d772087eaa953798286",
     "spec/v2/core/idempotency.md": "2084d756789c1faf5121593fb39e67ca36a4c227178e175004742e403178ea49",
-    "spec/v2/core/identity.md": "2dceb1c2697015e4a4fa8a9b28f4fbd25fa44a76ca5a919c752e2a8163c819db",
+    "spec/v2/core/identity.md": "582eddbf2e1bad41911e886bd08524683837045404f7b8f6bceccf61792b2765",
     "spec/v2/core/interop.md": "e72e25dff864c83b2980d7366b34054cffa072af0225821199f260f6228d260e",
-    "spec/v2/core/interrupt.md": "4c825634acac72900ebd3c5ac5941f55648fa33d2df92913ef16371dcb700a34",
+    "spec/v2/core/interrupt.md": "c1e13179e792bf8099b2d1c7891542646bb1d4cdee1757590f190833cae6047c",
     "spec/v2/core/node-pack-runtimes.md": "e1088cb69b8c3c67256c494063328916c2980ab1aa97a7063de25bab82fdb1e5",
     "spec/v2/core/oauth.md": "c96483a44b62978c596945fbb7d7c863cf6056cd980fdb4cf4e97555e3e10b72",
-    "spec/v2/core/overview.md": "c8c3bedb8e02dd8ce244b8a7a75d23af1c30dc92a343af912489424a89674d70",
-    "spec/v2/core/packs.md": "5929e708bd4e79f6e1beea570af583c80f3b63bb21dec78a5df03c235e73364a",
+    "spec/v2/core/overview.md": "d7000fbcd2f59280870fc592f5d5ea28a5195b9eb209029f09c5de59e0f2bf9c",
+    "spec/v2/core/packs.md": "935fc2f633c68fa98ad6c9cdb083dd556e955b1a15eff32f4773d13329a57238",
     "spec/v2/core/persistence.md": "711874f74cb08b03cd118d683056db10a063517762fcea73ec70feb5d983e6a4",
     "spec/v2/core/portability.md": "87c4327f674f714fda6217fd79b99d2ba219ad47dc02f9751bb56c200fb2c17a",
-    "spec/v2/core/replay.md": "4d379494057746c6883f6047972e93e621863b998ed291b9a691d3aa57c1ff43",
-    "spec/v2/core/runs.md": "55e21292ae4b41d0d104199a6c2634ecb3f787c4a671f9ebc37084be38ceef03",
+    "spec/v2/core/replay.md": "f406d71fb690b2e71f4b86ff799a49e6f1a0662c98fc59bcfe517d100eb6117c",
+    "spec/v2/core/runs.md": "8a1d71fb9305fd0df31150bfbbd892f96b8aab553400a36196212f7d0a24906d",
     "spec/v2/core/security-defaults.md": "153884da7a2d54c30c81b5e649cfdef31e151b3a2a19dee1bc5a16c35d60a3aa",
     "spec/v2/core/storage.md": "598ad4f3775c297c698a20daf72a7d15a6836c1231ddbca5961865c704f765b5",
     "spec/v2/core/tool-catalog.md": "d16a9004418719c48301804aefa4d7ab6f41112a4ed83727f552aa1b92b418b0",
     "spec/v2/core/versioning.md": "e3b3ab6d76e51035039a32173e9b412628a52c8fe7b16687ff58a2d0d983faa0",
-    "spec/v2/core/webhooks.md": "d2d33c6be777f217ade3e15dd369849e5c6ce67e1a4417216f0aa6ec5bdecc56",
+    "spec/v2/core/webhooks.md": "3f8730e0d15a5e4244e302adb910a0b6f745fbfa8e80920c14118117811ba881",
     "spec/v2/core/workflow-chain-packs.md": "fdce7a8477e269950a2bac05ee6eddbd66eb0b670b464d00054f7196ec05d29c",
     "spec/v2/corrections.json": "59f647f4d22477f6c2b06dcf913b682dde5d51dd5e607a5d0cf648ab60864d6e",
     "spec/v2/corrections.schema.json": "46bbffb954d13bd9995aebc912ceeb2544f7064331abf25fd6ab85c81a5cec94",
-    "spec/v2/declaration.json": "56c85c6d49f93d08d3bd48ccf1c049191cfbc4d4c22a1cd5202c1b6252759aca",
+    "spec/v2/declaration.json": "edace8c4ec7e32aba44828f459e2ae8933dc88ec8129f16549bca6279e3ca24c",
     "spec/v2/declaration.schema.json": "ef951633e4070899f64828f5cf1b5f4870270a0360f81dbb49b5b4d6cd407550",
     "spec/v2/errors.json": "fa650cbbdf25a140fdaf82ec5269f95b4585ac4bdde1fa6b0b3ffe48a3f7943c",
     "spec/v2/errors.schema.json": "47404ad53e6147cf9017b813483712466573e419fb33e6c4b6c12d3e7deba1bc",
@@ -265,11 +267,11 @@
     "spec/v2/ext/kanban/README.md": "bfc386dc0f874a4afa1e3ec6a2f99084b547dcce9bd363da3b8ebd8786b9c184",
     "spec/v2/ext/knowledge/README.md": "cbe1f945a757791dedac0fd542a68c86c5b457778b831d58d9ae44a2baa270f2",
     "spec/v2/ext/launchStudio/README.md": "a941ef1d2401438a412eaea48d2d6ca807a032b44ea08719fea33120daf088a0",
-    "spec/v2/ext/messaging/README.md": "949f350e35294f52ad490e42699f8c560cd154549516fbaa1805f06312b7f342",
+    "spec/v2/ext/messaging/README.md": "ee2d19fb6f6c2de11d889e966a41e6d1ff34272771689a81d7de039105a5be03",
     "spec/v2/ext/portability/README.md": "c59423ee446638b748a911b9e97e803c4786157be6120f656f0a060586bc7b82",
     "spec/v2/ext/provider-idempotency/README.md": "9d1bccef0ec16a19a3370ae79c2f5a1726ce0d1d25a51352907f288a2ccb15d7",
     "spec/v2/ext/provider-idempotency/registry.json": "5b5fecad604fb4ab38da39f5abe697a4e57b05406c398f8afeb377331b2bd965",
-    "spec/v2/ext/restTransport/README.md": "c875fd7ae4818a5d5e93ab9b21e569de8700bb07c4dafacd60a46a46e4022ddb",
+    "spec/v2/ext/restTransport/README.md": "bd7207683b2fbcead78bdb7d474fa723e08adad500185ea10804fd4e99541c0c",
     "spec/v2/ext/sandbox-runtime-notes/README.md": "0d1290fb1e20b2e35c0b15051541ab6546e4e2a6f693b4f912cb3f7d6a65e92c",
     "spec/v2/ext/webResearch/README.md": "b675e6f6646128d781b661f754c2393eed86e5935d16c3a22df582a5ce025b9b",
     "spec/v2/facets/a2a.schema.json": "b68bf5e29e455e46c2ad7cfcccf5963730ec2310ff9db81884550d8df30fd6ef",
@@ -299,9 +301,9 @@
     "spec/v2/path-manifest.json": "17298a245c5da9a70aa7ab0b7f2db4277e06209d6fcdd7187a8b4cea179c08f3",
     "spec/v2/peer-dependency-aliases.json": "d10299280abee08258502925bc327293ee413e0108cd6e6ec75ff6110653308d",
     "spec/v2/profiles.json": "180beb9ef2d161766fa23b2987e7e829851e3d316273147cf6a4a3f324befd6b",
-    "spec/v2/release.json": "28b407019929da55701a718110d440f08899646af0b74d62f6adc5542c9c1c91",
+    "spec/v2/release.json": "8b868146a0b8fef709478ed80a75a204c980d0c761463e0602c921c6b25fd0d8",
     "spec/v2/retention-floors.json": "eaf3722d95c79947af1d4269ef85117e126518c588cfcf1a2b21b97269f51624",
-    "spec/v2/surface-baseline.json": "d0540276e74001f5058dd81909100ea6ee3c1925c6ac02918418f5ec941d0661"
+    "spec/v2/surface-baseline.json": "d074465ed4ee594016bfb842fc3161d278394e84740133daf4fd93d1b3b0c4d8"
   },
-  "corpusCommit": "c486eafcfa16908a7b57c98b33cfd66fdf8e599d"
+  "corpusCommit": "95fbb4c9963f9da47eb0e870bf402bad3f4db29d"
 }

package/src/lib/row-observation.ts

@@ -0,0 +1,40 @@
+/**
+ * An informational note for the CURRENT `it`'s row (suite 2.44.3, RFC 0225).
+ *
+ * `setup.ts` attaches it to the row's `detail` in `afterEach`, whatever the
+ * disposition, so a PASSING row can say what it observed. The bundle writer
+ * already carries `detail` on every row and the witness digest covers it, so the
+ * note is signed with the row. It is prefixed `observed: `, never
+ * `partial-witness: `: it describes how a witness was obtained and does not
+ * qualify it (`check-accepted-predicate` counts a pass as partial only on that
+ * prefix). Same shape as `noteEvidence` (lib/durability-evidence.ts).
+ */
+export const OBSERVATION_PREFIX = 'observed: ';
+
+let pending: string | null = null;
+
+/** Note what this `it` observed. A second note in the same `it` is appended. */
+export function noteObservation(text: string): void {
+  const t = text.trim();
+  if (t === '') return;
+  pending = pending === null ? t : `${pending} · ${t}`;
+}
+
+/** Take (and clear) the pending note. */
+export function takeNotedObservation(): string | null {
+  const n = pending;
+  pending = null;
+  return n;
+}
+
+/**
+ * The detail a row carries once an observation is attached. A row with no
+ * detail gets `observed: …`; a row that already has one (a non-pass reason, or
+ * a partial-witness marker) keeps it FIRST and gains `· observed: …`, so no
+ * prefix a reader filters on ever moves.
+ */
+export function attachObservation(detail: string | undefined, observation: string | null): string | undefined {
+  if (observation === null || observation.trim() === '') return detail;
+  if (detail === undefined || detail.trim() === '') return `${OBSERVATION_PREFIX}${observation}`;
+  return `${detail} · ${OBSERVATION_PREFIX}${observation}`;
+}

package/src/lib/webhook-retry-window.ts

@@ -90,6 +90,62 @@ export function retryTestTimeoutMs(waits: number, slackMs: number): number {
   return waits * MAX_RETRY_WAIT_MS + slackMs;
 }
 
+/**
+ * WHICH branch `retryWaitFor` took, so a row can say so (RFC 0225 witness, suite
+ * 2.44.3). A passing dead-letter row used to carry no detail at all, so a
+ * bundle could not show whether the wait was derived from the host's advertised
+ * `maxElapsedMs` (the RFC 0225 path), from an operator-raised cap, or from the
+ * 20 s floor, and a certifier could not tell a witness of the bound from a pass
+ * the old default would have produced anyway.
+ *
+ *   advertised-bound  the window is the advertised maxElapsedMs + the grace
+ *   cap               the window is the cap (default or operator-raised); when
+ *                     a bound is advertised this means the operator raised the
+ *                     cap ABOVE it, so the bound did not set the wait
+ *   floor             no policy, or backoff none: the 20 s floor
+ */
+export type RetryWaitPath = 'advertised-bound' | 'cap' | 'floor';
+export interface RetryWaitSelection { readonly path: RetryWaitPath; readonly windowMs: number; readonly maxElapsedMs: number | null; readonly capMs: number }
+
+export function retryWaitSelection(policy: AdvertisedRetryPolicy | null, capMs: number): RetryWaitSelection {
+  const windowMs = retryWaitFor(policy, capMs);
+  const bound = advertisedMaxElapsedMs(policy);
+  if (policy === null) return { path: 'floor', windowMs, maxElapsedMs: null, capMs };
+  if (bound !== null) return { path: bound + ADVERTISED_BOUND_GRACE_MS >= capMs ? 'advertised-bound' : 'cap', windowMs, maxElapsedMs: bound, capMs };
+  const backoff = String(policy.backoff ?? '');
+  return { path: backoff === 'exponential' || backoff === 'fixed' ? 'cap' : 'floor', windowMs, maxElapsedMs: null, capMs };
+}
+
+/** What the leg measured, in ms since the delivery's FIRST attempt reached the receiver. */
+export interface RetryWaitMeasurement { readonly attempts: number; readonly lastAttemptAfterMs: number | null; /** undefined when the leg does not read the sink */ readonly sinkSeenAfterMs?: number | null }
+
+/**
+ * The informational row detail: the path, the window, and what was measured.
+ * The last attempt is a LOWER bound on the dead-lettering time (the host cannot
+ * dead-letter before its last attempt); the sink observation is an UPPER bound
+ * (the read that first found the record). Never starts with the partial-witness
+ * prefix: this describes a witness, it does not qualify one.
+ */
+/** The path and window alone, for a leg that measures no dead-lettering. */
+export function waitPathNote(sel: RetryWaitSelection): string {
+  return `${pathLabel(sel)}: waited ≤${sel.windowMs}ms`;
+}
+
+function pathLabel(sel: RetryWaitSelection): string {
+  return sel.path === 'advertised-bound'
+    ? `advertised-bound (maxElapsedMs ${sel.maxElapsedMs})`
+    : sel.path === 'cap'
+      ? `cap (${sel.capMs > DEFAULT_RETRY_WAIT_CAP_MS ? `${RETRY_WAIT_ENV} raised to ${sel.capMs}` : `default ${sel.capMs}`}ms; advertised maxElapsedMs ${sel.maxElapsedMs ?? 'none'})`
+      : 'floor';
+}
+
+export function waitObservation(sel: RetryWaitSelection, m: RetryWaitMeasurement): string {
+  const how = pathLabel(sel);
+  const last = m.lastAttemptAfterMs === null ? 'no attempt' : `last attempt after ${m.lastAttemptAfterMs}ms`;
+  const sink = m.sinkSeenAfterMs === undefined ? '' : m.sinkSeenAfterMs === null ? ', sink not observed' : `, sink after ${m.sinkSeenAfterMs}ms`;
+  return `${how}: waited ≤${sel.windowMs}ms; ${m.attempts} attempt(s), ${last}${sink}`;
+}
+
 /** What a row says when its window closed before the host's schedule did. Computed, so the numbers a host reads are the ones the run used. */
 export function windowClosedNote(seen: number, maxAttempts: number, waitedMs: number, capMs: number): string {
   const raised = capMs > DEFAULT_RETRY_WAIT_CAP_MS;

package/src/scenarios/replay-side-effect-suppression.test.ts

@@ -34,7 +34,9 @@
  *
  * ## Construction
  *
- * `conformance-replay-side-effect` is `core.delay` → a side-effecting node. We
+ * `conformance-replay-side-effect` is `core.conformance.hold` (a reserved PURE
+ * wait; it was `core.delay`, which the registry declares side-effecting,
+ * openwop#1769) → a side-effecting node. We
  * start it with a long delay, cancel mid-flight so the effect node never
  * records a terminal outcome, then fork `mode:"replay"`. Requirement 3 of
  * `replay.md` §"Side-effect suppression in replay" then applies to that node.
@@ -145,7 +147,7 @@ describe.skipIf(SKIP_NO_FIXTURE)('replay-side-effect-suppression: a replay does
     //    side-effecting node downstream never reaches a terminal outcome.
     //
     //    Kept SHORT on purpose: a `replay` fork re-executes from seq 0, so it
-    //    re-runs this delay too (a delay is not side-effecting). A 30s delay
+    //    re-runs this hold too (`core.conformance.hold` is pure). A 30s delay
     //    here would make the fork outlive any sane poll budget — measured, not
     //    theorised: the first version of this scenario used 30s and timed out
     //    at step 3 with the fork still `running`.
@@ -182,7 +184,7 @@ describe.skipIf(SKIP_NO_FIXTURE)('replay-side-effect-suppression: a replay does
     }
     expect(fork.status, req('openwop.it.replay-side-effect-suppression.fails-a-side-effecting-node-closed-with-replay-source-missing-when-the-source-ne', 'replay.md §"Side-effect suppression in replay" requirement 1', 'fork should be accepted')).toBe(201);
     const forkRunId = (fork.json as { runId: string }).runId;
-    await pollUntilTerminal(forkRunId, { timeoutMs: 45_000 }); // re-runs the 5s delay before reaching `effect`
+    await pollUntilTerminal(forkRunId, { timeoutMs: 45_000 }); // re-runs the pure hold before reaching `effect`
 
     // 3. THE assertion. A non-suppressing host executes the node; a suppressing
     //    host cannot, and has no recorded outcome to serve, so it must fail

package/src/scenarios/v2-approval-reject-disposition.test.ts

@@ -33,6 +33,21 @@
  *      rejection is not replayed). `inapplicable` on a host that does not
  *      advertise the fixture, i.e. one that does not run cycles.
  *
+ *   9. `openwop.requirement.0223.replay-derives-rejection` (openwop#1756, suite
+ *      2.44.4; gated on `replay` advertising mode `replay` and on each leg's
+ *      fixture): a `replay` fork taken at the gate's recorded `interrupt.resolved` (the
+ *      prefix holds its request and suspension) derives
+ *      the rejection from the recorded `interrupt.resolved` and never
+ *      re-decides it. The fork raises no new `interrupt.requested` for the
+ *      gate (the discriminator: a re-deciding host re-raises and waits), records
+ *      an `interrupt.resolved` whose `decision` / `action` / `reason` equal the
+ *      source's, fails the same way (`approval_rejected`, `failedNodeId`, a
+ *      non-retryable `node.failed`), and answers a resolve with `409
+ *      interrupt_already_resolved`. Leg A replays a decided reject
+ *      (`conformance-approval`); leg B replays a timeout
+ *      (`conformance-approval-timeout-approve`), whose `resolvedBy` must match the source's (a host subject or absent).
+ *      `interruptId` equality is not asserted: a host may re-mint ids.
+ *
  * The legs are separate rows so a host that misses one field fails that row
  * only. Legs 5 and 6 each need their own fixture, because `conformance-approval`
  * has no edges and no timer and widening a fixture every host serves would force
@@ -98,8 +113,8 @@ const errorCode = (o: unknown): string | undefined => (o as { code?: unknown } |
 const resolve = (runId: string, resumeValue: unknown): Promise<OpenWOPResponse | null> => http(() => driver.post(`/runs/${enc(runId)}/interrupts/${enc(NODE_ID)}`, { resumeValue }));
 
 /** Create a run of `fixture` and wait for it to suspend on its gate. */
-async function suspended(fixture: string): Promise<{ runId: string } | { reason: string }> {
-  const res = await http(() => driver.post('/runs', { workflowId: fixture }));
+async function suspended(fixture: string, extra: Record<string, unknown> = {}): Promise<{ runId: string } | { reason: string }> {
+  const res = await http(() => driver.post('/runs', { workflowId: fixture, ...extra }));
   if (res === null) return { reason: 'POST /runs unreachable (fetch failed)' };
   const runId = (res.json as { runId?: unknown } | null)?.runId;
   if (res.status !== 201 || typeof runId !== 'string') return { reason: `POST /runs (${fixture}) answered ${res.status} ${readErrorCode(res.json) ?? ''}`.trim() };
@@ -242,7 +257,8 @@ describe('RFC 0223 — v2-approval-reject-disposition (gated on interrupt + conf
   it('a rejected gate looped back to itself is asked again under a new key, and the first rejection is not replayed', async () => {
     if (!(await gateFamily('interrupt'))) return softSkip('inapplicable', 'interrupt family not advertised (gate recorded under openwop.family.interrupt)');
     if (!isFixtureAdvertised(LOOPBACK)) return softSkip('inapplicable', `fixture ${LOOPBACK} is not advertised — the host does not run cycles`);
-    const s = await suspended(LOOPBACK);
+    // The loop is bounded by the scenario (one reject), not by the host; recursionLimit is a runaway guard (openwop#1748).
+    const s = await suspended(LOOPBACK, { configurable: { version: 1, run: { recursionLimit: 20 } } });
     if ('reason' in s) return softSkip('blocked', s.reason);
     const res = await resolve(s.runId, { action: 'reject' });
     if (res === null || res.status < 200 || res.status >= 300) return softSkip('blocked', `the reject resolve answered ${res?.status ?? 'nothing'} ${readErrorCode(res?.json) ?? ''} — interrupt resolution owns that contract`);
@@ -259,4 +275,91 @@ describe('RFC 0223 — v2-approval-reject-disposition (gated on interrupt + conf
     expect(evs.filter((e) => e.type === 'interrupt.resolved').length, req(id, 'spec/v2/core/interrupt.md §Re-entry and resume values', 'an earlier visit\'s resumeValue MUST NOT be returned: only the one caller resolve is recorded')).toBe(1);
     await http(() => driver.post(`/runs/${enc(s.runId)}/cancel`, {}));
   }, 45_000);
+
+  // ── openwop#1756: replay derives the rejection, never re-decides it ──────────
+  const REPLAY_ID = R('replay-derives-rejection');
+  const REPLAY_DOC = `${DOC}, last bullet; spec/v2/core/replay.md §Determinism`;
+  interface SeqEv extends Ev { readonly sequence?: number }
+  async function debugLog(runId: string): Promise<SeqEv[]> {
+    const res = await http(() => driver.get(`/runs/${enc(runId)}/events/poll?timeout=1&streamMode=debug`));
+    const evs = ((res?.json as { events?: SeqEv[] } | null)?.events ?? []).filter((e) => typeof e.sequence === 'number');
+    return evs.sort((a, b) => (a.sequence as number) - (b.sequence as number));
+  }
+  const onGate = (e: SeqEv): boolean => (e.payload?.['nodeId'] ?? e.nodeId) === NODE_ID;
+  async function replayGate(): Promise<string | null> {
+    const fam = await gateFamily('replay');
+    if (!fam) return 'replay family not advertised (gate recorded under openwop.family.replay)';
+    const modes = Array.isArray(fam['modes']) ? (fam['modes'] as unknown[]) : [];
+    return modes.includes('replay') ? null : `replay.modes ${JSON.stringify(modes)} does not include replay`;
+  }
+  /**
+   * Run → the source's own disposition → a `replay` fork at the gate's recorded
+   * `interrupt.resolved`: the prefix holds the gate's request and suspension, and the
+   * fork must supply the resolution. A deriving host takes the recorded one (replay.md
+   * §Determinism 2); a re-deciding host raises a NEW `interrupt.requested` and waits.
+   * Forking at the gate's `node.started` instead would let a host re-emit the recorded
+   * request as history, which cannot be told apart from a new one.
+   */
+  async function replayedRejection(fixture: string, decide: ((runId: string) => Promise<string | null>) | null, settleMs: number): Promise<{ source: SeqEv[]; fromSeq: number; forkId: string; forkStatus: string | null; fork: SeqEv[] } | { reason: string }> {
+    const s = await suspended(fixture);
+    if ('reason' in s) return s;
+    if (decide !== null) { const why = await decide(s.runId); if (why !== null) return { reason: why }; }
+    const terminal = await waitStatus(s.runId, TERMINAL, settleMs);
+    if (terminal?.['status'] !== 'failed') return { reason: `the source run did not end failed (status ${String(terminal?.['status'])}) — the reject rows own that contract` };
+    const source = await debugLog(s.runId);
+    const at = source.find((e) => e.type === 'interrupt.resolved' && onGate(e));
+    if (at?.sequence === undefined) return { reason: 'the source log has no interrupt.resolved for the gate to fork at' };
+    const fork = await http(() => driver.post(`/runs/${enc(s.runId)}:fork`, { mode: 'replay', fromSeq: at.sequence }));
+    const forkId = (fork?.json as { runId?: unknown } | null)?.runId;
+    if (fork === null || fork.status !== 201 || typeof forkId !== 'string') return { reason: `POST /runs/{runId}:fork answered ${fork?.status ?? 'nothing'} ${readErrorCode(fork?.json) ?? ''} — forkRun owns that contract`.trim() };
+    const settled = await waitStatus(forkId, new Set([...TERMINAL, 'waiting-approval']), scaledTimeoutMs(20_000));
+    return { source, fromSeq: at.sequence, forkId, forkStatus: settled === null ? null : String(settled['status']), fork: await debugLog(forkId) };
+  }
+  function assertDerived(r: { source: SeqEv[]; fromSeq: number; forkId: string; forkStatus: string | null; fork: SeqEv[] }, leg: string): void {
+    const reRaised = r.fork.filter((e) => (e.sequence as number) >= r.fromSeq && (e.type === 'interrupt.requested' || e.type === 'approval.requested') && onGate(e));
+    expect(reRaised.length, req(REPLAY_ID, REPLAY_DOC, `${leg}: the replay fork MUST NOT raise a new interrupt for the gate — the rejection is derived from the recorded interrupt.resolved, never re-decided (status ${String(r.forkStatus)})`)).toBe(0);
+    const pick = (e: SeqEv | undefined): Record<string, unknown> => ({ decision: e?.payload?.['decision'], action: e?.payload?.['action'], reason: e?.payload?.['reason'] });
+    const srcResolved = r.source.find((e) => e.type === 'interrupt.resolved' && onGate(e));
+    const forkResolved = r.fork.find((e) => e.type === 'interrupt.resolved' && onGate(e));
+    expect(pick(forkResolved), req(REPLAY_ID, REPLAY_DOC, `${leg}: the fork's interrupt.resolved for the gate MUST carry the source's decision, action and reason`)).toEqual(pick(srcResolved));
+    const runFailed = r.fork.find((e) => e.type === 'run.failed');
+    const nodeFailed = r.fork.find((e) => e.type === 'node.failed' && onGate(e));
+    expect(
+      { status: r.forkStatus, code: errorCode(runFailed?.payload?.['error']), failedNodeId: runFailed?.payload?.['failedNodeId'], retryable: (nodeFailed?.payload?.['error'] as { retryable?: unknown } | undefined)?.retryable },
+      req(REPLAY_ID, REPLAY_DOC, `${leg}: the fork MUST fail as the source did — failed, run.failed approval_rejected naming the gate, and a non-retryable node.failed`),
+    ).toEqual({ status: 'failed', code: CODE, failedNodeId: NODE_ID, retryable: false });
+  }
+
+  it('a replay fork of a decided reject derives the rejection from the log and never re-decides it (leg A)', async () => {
+    if (!(await gateFamily('interrupt'))) return softSkip('inapplicable', 'interrupt family not advertised (gate recorded under openwop.family.interrupt)');
+    const noReplay = await replayGate();
+    if (noReplay !== null) return softSkip('inapplicable', noReplay);
+    if (!isFixtureAdvertised(FIXTURE)) return softSkip('inapplicable', `fixture ${FIXTURE} is not advertised`);
+    const r = await replayedRejection(FIXTURE, async (runId) => {
+      const res = await resolve(runId, { action: 'reject' });
+      return res === null || res.status < 200 || res.status >= 300 ? `the reject resolve answered ${res?.status ?? 'nothing'} ${readErrorCode(res?.json) ?? ''}` : null;
+    }, 10_000);
+    if ('reason' in r) return softSkip('blocked', r.reason);
+    const again = await resolve(r.forkId, { action: 'accept' });
+    assertDerived(r, 'leg A (decided reject)');
+    expect({ status: again?.status, code: readErrorCode(again?.json) }, req(REPLAY_ID, 'spec/v2/core/errors.md §One code per state', 'a resolve against the replayed gate MUST answer 409 interrupt_already_resolved — the fork holds no open decision')).toEqual({ status: 409, code: 'interrupt_already_resolved' });
+  }, 90_000);
+
+  it('a replay fork of a timed-out gate derives the timeout rejection from the log and never re-arms it (leg B)', async () => {
+    if (!(await gateFamily('interrupt'))) return softSkip('inapplicable', 'interrupt family not advertised (gate recorded under openwop.family.interrupt)');
+    const noReplay = await replayGate();
+    if (noReplay !== null) return softSkip('inapplicable', noReplay);
+    if (!isFixtureAdvertised(TIMEOUT_APPROVE)) return softSkip('inapplicable', `fixture ${TIMEOUT_APPROVE} is not advertised`);
+    const r = await replayedRejection(TIMEOUT_APPROVE, null, scaledTimeoutMs(20_000));
+    if ('reason' in r) return softSkip('blocked', r.reason);
+    assertDerived(r, 'leg B (timeout)');
+    // resolvedBy on a timeout row is the host's subject or absent (RFC 0183 §Unresolved 1), so
+    // the leg asserts the fork carries the SOURCE's value, not that it is absent.
+    const forkResolved = r.fork.find((e) => e.type === 'interrupt.resolved' && onGate(e));
+    const srcResolved = r.source.find((e) => e.type === 'interrupt.resolved' && onGate(e));
+    expect(
+      { action: forkResolved?.payload?.['action'], reason: forkResolved?.payload?.['reason'], resolvedBy: JSON.stringify(forkResolved?.payload?.['resolvedBy'] ?? null) },
+      req(REPLAY_ID, REPLAY_DOC, 'leg B: the derived resolution is the recorded timeout — action timeout, reason timeout, and the source\'s resolvedBy (a host subject or absent), never a caller'),
+    ).toEqual({ action: 'timeout', reason: 'timeout', resolvedBy: JSON.stringify(srcResolved?.payload?.['resolvedBy'] ?? null) });
+  }, 120_000);
 });

package/src/scenarios/v2-replay-suppression-ordinal.test.ts

@@ -1,9 +1,10 @@
 /**
  * `spec/v2/core/replay.md` §Suppression rule 2 — a replay fork resolves a
- * side-effecting node's outcome keyed on `(sourceRunId, nodeId, n)`, where `n`
- * counts the node's `node.started` events through this execution, including
- * retries, later visits and the fork's inherited prefix (openwop#1718, suite
- * 2.43.2, target major 2).
+ * side-effecting node's outcome keyed on `(sourceRunId, nodeId, n)`, where `n` is
+ * one more than the node's recorded terminals (`node.completed`, `node.failed`)
+ * before this execution, the fork's inherited prefix included (openwop#1718,
+ * re-corrected by openwop#1769 from a count of `node.started`, which a pause or a
+ * suspend re-emits; suite 2.43.2, target major 2).
  *
  * The rule said `(nodeId, attempt)`. In a run where a node executes more than
  * once (a loop back over an edge), that key names no single execution: a host
@@ -11,8 +12,8 @@
  * outcome, and one selecting the latest recorded outcome does the same when
  * only one exists.
  *
- * Construction (`conformance-replay-ordinal-loop`): start → effect → wait →
- * effect, so `effect` executes twice. The suite cancels the source inside
+ * Construction (`conformance-replay-ordinal-loop`): start → effect → wait
+ * (`core.conformance.hold`, a reserved PURE node that re-executes live) → effect, so `effect` executes twice. The suite cancels the source inside
  * `wait`, after `effect`'s first execution completed and before its second, and
  * forks `mode: "replay"` at `effect`'s first `node.started`. In the fork:
  *   - `effect` execution 1 (n = 1) has a recorded outcome and resolves from it;
@@ -74,7 +75,7 @@ describe('v2 replay-suppression-ordinal (replay.md §Suppression rule 2, openwop
     if (!(await gateFamily('replay'))) return softSkip('inapplicable', 'replay family not advertised (gate recorded under openwop.family.replay)');
     if (!isFixtureAdvertised(FIXTURE)) return softSkip('inapplicable', `fixture ${FIXTURE} is not advertised — the host does not run cycles with a side-effecting node`);
 
-    const created = await http(() => driver.post('/runs', { workflowId: FIXTURE, inputs: { delayMs: 3000 } }));
+    const created = await http(() => driver.post('/runs', { workflowId: FIXTURE, inputs: { delayMs: 3000 }, configurable: { version: 1, run: { recursionLimit: 20 } } }));
     const runId = (created?.json as { runId?: unknown } | null)?.runId;
     if (created === null || created.status !== 201 || typeof runId !== 'string') return softSkip('blocked', `POST /runs (${FIXTURE}) answered ${created?.status ?? 'nothing'} ${readErrorCode(created?.json) ?? ''}`.trim());
 

package/src/scenarios/v2-webhook-durable-delivery.test.ts

@@ -35,7 +35,8 @@ import { absenceIsUnmeasured, noDeliveryCause, startScopedReceiver, type ScopedR
 import { readErrorCode } from '../lib/error-envelope.js';
 import { blockedDespiteAssertions, softSkip } from '../lib/soft-skip.js';
 import { req } from '../lib/requirement-ids.js';
-import { pastAdvertisedBound, retryTestTimeoutMs, retryWaitCapMs, retryWaitFor, windowClosedNote } from '../lib/webhook-retry-window.js';
+import { pastAdvertisedBound, retryTestTimeoutMs, retryWaitCapMs, retryWaitFor, retryWaitSelection, waitObservation, windowClosedNote } from '../lib/webhook-retry-window.js';
+import { noteObservation } from '../lib/row-observation.js';
 
 export const REQUIRES_HOST_CALLBACK = 'the host POSTs and retries webhook deliveries to the suite-owned scoped receiver behind OPENWOP_WEBHOOK_RECEIVER_URL';
 
@@ -180,6 +181,21 @@ function retryWaitMs(doc: Record<string, unknown>): number {
   return retryWaitFor(advertisedRetryPolicy(doc), RETRY_WAIT_CAP_MS);
 }
 
+/**
+ * Suite 2.44.3 (RFC 0225 witness): say, on the row, which wait path ran and what
+ * was measured. The first attempt is t0; the last attempt lower-bounds the
+ * dead-lettering, and the sink read that found the record upper-bounds it.
+ */
+function observeWait(doc: Record<string, unknown>, attempts: ReadonlyArray<{ readonly at: number }>, sinkSeenAt?: number | null): void {
+  const t0 = attempts[0]?.at;
+  const last = attempts[attempts.length - 1]?.at;
+  noteObservation(waitObservation(retryWaitSelection(advertisedRetryPolicy(doc), RETRY_WAIT_CAP_MS), {
+    attempts: attempts.length,
+    lastAttemptAfterMs: t0 === undefined || last === undefined ? null : last - t0,
+    ...(sinkSeenAt === undefined ? {} : { sinkSeenAfterMs: t0 === undefined || sinkSeenAt === null ? null : sinkSeenAt - t0 }),
+  }));
+}
+
 /**
  * The per-test budget, DERIVED from the wait above (suite 2.0.2).
  *
@@ -287,6 +303,7 @@ describe('RFC 0173 §B — webhook-durable-delivery (gated on webhooks)', () =>
     const ours = () => receiver.attempts.filter((a) => a.runId === runId && a.webhookId === sub.webhookId);
     const retried = await waitFor(() => ours().some((a) => a.status === 204), retryWaitMs(doc));
     const attempts = ours();
+    observeWait(doc, attempts);
     // A zero that is PROVABLY not a verdict about the host records `blocked`
     // with its cause, not `executed-fail` (2.37.0). `absenceIsUnmeasured` is
     // true only when other traffic reached this listener — the path from the
@@ -440,6 +457,7 @@ describe('RFC 0173 §B — webhook-durable-delivery (gated on webhooks)', () =>
       const sink = await driver.get(`/webhooks/${projectBoundId(sub.webhookId)}/dead-letters`);
       inSink = sink.status === 200 && ((sink.json as { deliveries?: unknown[] } | null)?.deliveries ?? []).some((r) => (r as Record<string, unknown>)['runId'] === runId);
     }
+    observeWait(doc, ours(), inSink === true ? Date.now() : null);
     await driver.delete(`/webhooks/${encodeURIComponent(sub.webhookId)}`);
     // "Routed to the sink, not dropped" is a claim about an EXHAUSTED delivery,
     // and until 2.34.1 it was asserted whether or not exhaustion had been
@@ -589,9 +607,11 @@ describe('RFC 0173 §B — webhook-durable-delivery (gated on webhooks)', () =>
     await new Promise((r) => setTimeout(r, 1_000));
 
     const sink = await driver.get(`/webhooks/${projectBoundId(sub.webhookId)}/dead-letters`);
+    const sinkReadAt = Date.now();
     await driver.delete(`/webhooks/${encodeURIComponent(sub.webhookId)}`);
     if (sink.status !== 200) return blockedDespiteAssertions(`the dead-letter read answered ${sink.status}`);
     const rows = ((sink.json as { deliveries?: Array<Record<string, unknown>> } | null)?.deliveries ?? []);
+    observeWait(doc, ours(), rows.length > 0 ? sinkReadAt : null);
     // An empty sink is NOT a pass. Recording one as a pass is exactly the
     // vacuous witness this leg used to produce; say so instead.
     if (rows.length === 0) {

package/src/scenarios/v2-webhook-message-id-stable.test.ts

@@ -33,7 +33,8 @@ import { req } from '../lib/requirement-ids.js';
 import { readErrorCode } from '../lib/error-envelope.js';
 import { blockedDespiteAssertions, softSkip } from '../lib/soft-skip.js';
 import { hitHeader, mintWhsec, startModalReceiver, STANDARD_WEBHOOKS_ID, type ModalHit } from '../lib/webhook-receiver.js';
-import { retryTestTimeoutMs, retryWaitCapMs, retryWaitFor } from '../lib/webhook-retry-window.js';
+import { retryTestTimeoutMs, retryWaitCapMs, retryWaitFor, retryWaitSelection, waitPathNote } from '../lib/webhook-retry-window.js';
+import { noteObservation } from '../lib/row-observation.js';
 import {
   STANDARD_WEBHOOKS_ALG,
   deliveriesFor,
@@ -89,7 +90,9 @@ describe('RFC 0201 §C.10 — webhook-id is stable across retries, distinct acro
       return m;
     };
     const expectedKeys = subs.length * EVENTS.length;
-    const window = retryWaitFor((g.facet['retryPolicy'] as { backoff?: string; maxElapsedMs?: unknown } | undefined) ?? null, CAP_MS);
+    const advertised = (g.facet['retryPolicy'] as { backoff?: string; maxElapsedMs?: unknown } | undefined) ?? null;
+    const window = retryWaitFor(advertised, CAP_MS);
+    noteObservation(waitPathNote(retryWaitSelection(advertised, CAP_MS)));
     await waitFor(() => {
       const m = byKey();
       return m.size >= expectedKeys && [...m.values()].every((a) => a.length >= 2);

package/src/scenarios/v2-webhook-unregister-stops-delivery.test.ts

@@ -45,7 +45,8 @@ import { absenceIsUnmeasured, noDeliveryCause, startScopedReceiver, type ScopedR
 import { readErrorCode } from '../lib/error-envelope.js';
 import { blockedDespiteAssertions, softSkip } from '../lib/soft-skip.js';
 import { req } from '../lib/requirement-ids.js';
-import { retryTestTimeoutMs, retryWaitCapMs, retryWaitFor } from '../lib/webhook-retry-window.js';
+import { retryTestTimeoutMs, retryWaitCapMs, retryWaitFor, retryWaitSelection, waitPathNote } from '../lib/webhook-retry-window.js';
+import { noteObservation } from '../lib/row-observation.js';
 
 export const REQUIRES_HOST_CALLBACK = 'the host POSTs and retries webhook deliveries to the suite-owned scoped receiver behind OPENWOP_WEBHOOK_RECEIVER_URL';
 
@@ -94,6 +95,7 @@ describe('RFC 0215 §B — unregistering stops the attempts (gated on webhooks)'
     const policy = advertisedRetryPolicy(doc);
     if (policy?.maxAttempts === 1) return softSkip('inapplicable', 'webhooks.retryPolicy.maxAttempts is 1 — the host schedules no retry, so there is none for unregistering to stop');
     const windowMs = retryWaitFor(policy, CAP_MS);
+    noteObservation(waitPathNote(retryWaitSelection(policy, CAP_MS)));
 
     const arrivals = new Map<string, number[]>();
     let targetId: string | null = null;

package/src/scenarios/webhook-signed-delivery.test.ts

@@ -453,8 +453,18 @@ describe('webhook-signed-delivery: end-to-end HMAC v1', () => {
     activeServer = receiver;
 
     const ownedTenant = await discoverOwnedTenant(driver);
+    // The LOOPBACK address, over https so the refusal can only come from the
+    // loopback deny list and not from a scheme check. `receiver.url` is the
+    // wrong value here: since 2.37.0 `startScopedReceiver` returns the PUBLIC
+    // FRONT as `url` whenever OPENWOP_WEBHOOK_RECEIVER_URL is set (the tunnel,
+    // plus this exercise's nonce). So this control registered the tunnel, a
+    // host with a working guard correctly accepted it, and the control failed
+    // on a conformant host. Measured on openwop-app production 2026-09-28: the
+    // same host answers `https://127.0.0.1:…` and `https://localhost:…` with
+    // 400 webhook_url_rejected ("denied (loopback / link-local / private-IP)").
+    const loopback = receiver.localUrl.replace(/^http:/, 'https:');
     const reg = await driver.post('/v1/webhooks', {
-      url: receiver.url, // deliberately the raw loopback URL, NOT the tunnel
+      url: loopback, // deliberately the raw loopback listener, NOT the tunnel
       events: ['run.completed'],
       ...(ownedTenant ? { tenantId: ownedTenant } : {}),
     });

package/src/setup.ts

@@ -38,6 +38,7 @@ import { PKG_ROOT_PATH } from './lib/paths.js';
 import { recordRequirement, hasRequirement, journalLength, journalSince } from './lib/requirement-ledger.js';
 import { requirementIdForFile, resolveFileRecord, resolveItRecord, unrecordedTests, type FileTestState, type FinishedTest, type TestFailure } from './lib/scenario-disposition.js';
 import { softSkipDisposition, softSkipDispositionSince, softSkipMark } from './lib/soft-skip.js';
+import { attachObservation, takeNotedObservation } from './lib/row-observation.js';
 import { ItIdAllocator, takeExplicitRequirementId } from './lib/requirement-ids.js';
 import { SPEC_COHERENCE_SCENARIOS, SPEC_COHERENCE_DETAIL } from './lib/spec-coherence.js';
 import type { DiscoveryPayload } from './lib/profiles.js';
@@ -472,6 +473,7 @@ afterEach(({ task }) => {
   // Suite 2.0.0: src/coherence/ rows feed the corpus ledger (scripts/check-spec-coherence.mjs), not a host bundle.
   if (!/[\\/]src[\\/](scenarios|coherence)[\\/]/.test(filepath)) {
     takeExplicitRequirementId();
+    takeNotedObservation();
     return;
   }
   // A leg that soft-skips before its first assertion never calls `req()`, so the
@@ -506,6 +508,11 @@ afterEach(({ task }) => {
     disposition = rec.disposition;
     detail = rec.detail;
   }
+  // Suite 2.44.3 (RFC 0225 witness): an informational note the `it` wrote
+  // (`noteObservation`) rides on the row's detail, on a pass too — the bundle
+  // writer carries it and the witness digest signs it. Taken whatever the
+  // disposition, so a note never leaks into the next `it`.
+  detail = attachObservation(detail, takeNotedObservation());
   try {
     const evidence = takeNotedEvidence();
     // `fold: true` — several `it` legs of one file legitimately share ONE