@openwop/openwop-conformance 2.1.7 → 2.2.1

package/CHANGELOG.md

@@ -1,5 +1,109 @@
 # `@openwop/openwop-conformance` Changelog
 
+## [2.2.1] — 2026-09-16 — RFC 0183's `edit-accept` arm is witnessed, and the MUST the suite already enforced is now the spec's
+
+**Why a patch.** `PUBLISHING.md` §"Versioning alignment": new `it` blocks in an
+existing scenario file are a patch; no scenario file is added.
+
+### Added
+
+- **`interrupt-approval.test.ts` witnesses the `edit-accept` arm of RFC 0183
+  §A.2** — two legs on a new `conformance-approval-edit-accept` fixture: an
+  edit-accept resolve MUST round-trip `action: "edit-accept"` and the
+  `editedArtifactData` supplied, and an edit-accept resolve with no
+  `editedArtifactData` MUST be refused. Until now the refine arm was witnessed
+  and the edit-accept arm was stated in `spec/v2/core/interrupt.md` §Approval and
+  measured nowhere. Gated on the fixture being advertised, like the refine block;
+  a host that does not serve the fixture skips rather than fails.
+
+### Fixed
+
+- **`replay-fanout-suppression` declared ~77.5 s of poll budget inside vitest's
+  30 s wall** (openwop #1355). The second and third poll ceilings were
+  unreachable, `OPENWOP_POLL_TIMEOUT_SCALE` could not extend the wait (the
+  header claimed the opposite), and a slow host and a host that re-fired
+  fan-out on a replay fork failed with the same generic vitest timeout. The
+  test's wall is now derived from its own scaled budgets plus a 15 s margin,
+  so every declared ceiling is reachable and a failure names the predicate. No
+  assertion changed.
+
+### Changed (spec side, no assertion change)
+
+- **The `payload.action` assertion in the refine leg is no longer stricter than
+  the spec.** It asserted a MUST from 2.2.0 while RFC 0183 §A.1 said SHOULD —
+  a `COMPATIBILITY.md` §2.3 violation (a suite MUST NOT be stricter than the spec
+  on wire shape). The spec was raised, not the assertion lowered: RFC 0183 is now
+  `Active` and §A.1 is a MUST conditioned on approval kind, with its normative
+  home in `spec/v2/core/interrupt.md` §Events. RFC 0184 is `Active` too; suite
+  2.2.0 had shipped both RFCs' legs enforced against `Draft` headers, and the
+  corpus gate now refuses that (`check-rfc-status-coherence` rule 7).
+
+## [2.2.0] — 2026-09-16 — the suite wipes the host's mock program store between scenario files, and a bound id travels as one segment
+
+**Why a minor and not the 2.1.8 that was pinned for two days.** `PUBLISHING.md`
+§"Versioning alignment": a conformance scenario addition is a minor bump, and
+#1359 added `v2-bound-id-path-projection` (its header says so). No 2.1.x patch
+ever added a scenario file; this release does, so it is 2.2.0. Nothing was
+published as 2.1.8.
+
+This entry was first written on 2026-09-14 and said *"No scenario behaviour
+changes."* That was true when written and false by the time the tag was cut, so
+the entry is rewritten rather than appended to: two behaviour changes rode this
+version after the pin was bumped, and a package whose changelog denies them is a
+package lying about itself.
+
+### Fixed
+
+- **The suite now wipes the host's conformance-mock program store between
+  scenario FILES** (openwop #1357; half two of two, paired with openwop-app
+  #3870, which adds the seam). `src/setup.ts` posts
+  `POST /v1/host/openwop-app/test/mock-ai/reset` in a per-file `afterAll`.
+  Why: a host that keeps mock programs in a module-level map keyed by `nodeId`
+  retains any program a scenario does not fully drain, and the four
+  envelope-truncation scenarios seed `finishReason: 'length'` programs on
+  purpose — so a later file dispatching on a colliding node consumed the
+  leftovers and failed `envelope_truncation_unrecoverable` with nothing in that
+  file to explain it. Measured on `replay-observable-sequence-determinism`: red
+  in-suite, green alone, five runs on unchanged bases, one red at load 3.4 on an
+  idle box — not contention, and a wrong value rather than a missing result.
+  2.1.7's per-fixture node ids narrowed the collision; this removes the
+  leftover. **Best-effort by design:** a host without the seam answers 404 and
+  the call is swallowed, because a suite must never fail a compliant host for
+  lacking a TEST seam — so on such a host the leak, if it has one, persists
+  silently. Per file rather than per test because programs are seeded for a
+  whole scenario's attempt sequence.
+
+### Changed
+
+- **A tenant-bound id travels as one `~`-escaped path segment** (RFC 0184 §A.1,
+  openwop #1359; §A.5 exactly-once projection, #1360). A bound id is two
+  segments joined by `/`; a path parameter is one. The corpus said `%2F`
+  carried the separator, and a tier-1 host's front door decoded it back to `/`
+  before forwarding, so every bound id was unreachable through its own front
+  door. The escape marker is now `~` (RFC 3986 unreserved — an intermediary has
+  no license to rewrite it): every byte outside `[A-Za-z0-9._-]` becomes
+  `~XX`, total over bytes so a later grammar widening cannot invalidate a
+  projection already on the wire. New `src/lib/bound-id.ts` codec;
+  `v2-id-grammar` asserts the new form; new scenario
+  `v2-bound-id-path-projection` creates one run and asserts the projection is
+  applied exactly once (the codec is deliberately not idempotent — `a~3Ab` →
+  `a~7E3Ab` — so a double projection corrupts silently, and both reporting hosts
+  found a read path that already composed one). **Hosts: a v2 path parameter
+  carrying a bound id is now expected in the `~` form. `%2F` is no longer the
+  contract.**
+- **RFC 0183 requirement rows** (openwop #1358): `requirements.json` gains the
+  ids for the resume actions `interruptResolved` cannot record, and
+  `fixtures.md` documents `conformance-approval-refine`. The RFC is `Draft`; no
+  scenario asserts these rows yet.
+
+### Packaging
+
+- `spec/v2/declaration.json` gains an optional `normativeText` on family
+  entries (RFC 0169, openwop #1351): the path(s) where a family's behaviour is
+  actually written, as distinct from `section`, which names the declaration
+  site. The declaration and the schemas generated from it ship in the tarball,
+  which is what first required the identity bump.
+
 ## [2.1.7] — 2026-09-13 — nine fixtures shared one programmable node id
 
 The mock-AI program seam is keyed by `nodeId` alone — no run, no workflow, no

package/README.md

@@ -117,7 +117,7 @@ Exit code is non-zero on any failed assertion. `--certify` distinguishes: `0`
 
 ## What's Covered
 
-The current suite has 517 scenario files under `src/scenarios/`.
+The current suite has 518 scenario files under `src/scenarios/`.
 - 2026-09-03 (suite `1.157.0 -> 1.158.0`, gap G17): NEW `idempotency-concurrent-claim.test.ts` — drives the new `host-sample-test-seams.md` §25 concurrent duplicate-delivery seam for the RFC 0150 §B / `idempotency.md` §"Concurrent duplicates (Layer 2)" atomic-claim MUST, which is unconditional and had no witness of any kind. Asserts every executor mints the SAME `logicalInvocationId` **before** asserting `delivered === 1` — without the identity check a host passes by minting different ids and never colliding, one effect because nothing raced. Not profile-gated and so not opt-out-able (the obligation is unconditional); an unmounted seam records `blocked`, which is not certifiable. Graduates `layer2-invocation-claim-atomic` reference-impl -> protocol.
 - 2026-08-19 (suite `1.137.0 → 1.138.0`): NEW `durability-poison-exhaustion.test.ts` — RFC 0158 §C.8, the FIRST row of that RFC's conformance table to land. Asserts what `failure-path.test.ts` cannot: not just that deterministically failing work reaches terminal, but that attempts STOP — counted on the log, re-counted after a scaled quiet window, asserted unchanged. A host still redelivering records more. Seam-gated on the existing event-log seam (`blocked` = unobservable, not unmet) and outside every profile floor.
 - 2026-08-19 (suite `1.136.15 → 1.137.0`): NEW `replay-fanout-suppression.test.ts` — capability-gated on `webhooks.supported`, **outside every profile floor**; witnesses `replay.md` §"Host-initiated fan-out is an external effect", which was the largest normative MUST NOT on the replay surface with no scenario and no SECURITY invariant. Three legs in ONE `it` against ONE receiver and ONE subscription — a positive control, the MUST NOT, and a `branch` boundary leg — because "no delivery arrived" passes identically when delivery never worked, so absence is asserted only after presence is proven on that exact wiring. A host with an SSRF guard correctly refuses the loopback receiver and records `blocked`: **unobservable, not unmet.**
@@ -462,7 +462,7 @@ Server-required (added in 1.7.0):
 | ------------- | ----------------------------------------------------------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
 | **Redaction** | [`capabilities.md`](../spec/v1/capabilities.md) §"Secrets" + NFR-7 + §"aiProviders" | Vendor-neutral assertions that the server doesn't leak secret material. Three scenario groups: (a) discovery shape contract — `secrets` + `aiProviders` advertisements are well-formed regardless of `secrets.supported`; when `supported === true`, scopes MUST be non-empty + `resolution === 'host-managed'`; `byok ⊆ supported`. (b) bearer-token redaction — invalid Bearer canary in `Authorization` header is not echoed in the 401 response body. (c) credentialRef echo control — gated on `secrets.supported === true`; canary planted in `configurable.ai.credentialRef` MUST NOT appear in any RunEvent payload (poll-based capture; transport-agnostic). Uses runtime-built canary fixtures (`lib/canaries.ts`) that defeat static secret scanners. 6 scenarios. |
 
-Current source tree: 517 scenario files. Use [`coverage.md`](./coverage.md) for current grade/gap tracking.
+Current source tree: 518 scenario files. Use [`coverage.md`](./coverage.md) for current grade/gap tracking.
 
 ## Remaining Gaps
 

package/dist/spec-artifacts.lock.json

@@ -1,5 +1,5 @@
 {
   "package": "@openwop/spec-artifacts",
-  "version": "2.1.7",
-  "stampSha256": "5594e445dafa6f889a005c00556c2ae241f7a591f45e08eee0b126736bfb04d3"
+  "version": "2.2.1",
+  "stampSha256": "35a007d7fd7df22f277d9b3c6d58c2a723e3a7312eaafd851a0f771c18c844c5"
 }

package/fixtures/conformance-approval-edit-accept.json

@@ -0,0 +1,45 @@
+{
+  "id": "conformance-approval-edit-accept",
+  "name": "Conformance: Approval (edit-accept)",
+  "version": "1.0",
+  "description": "Suspends on an approval gate offering accept|reject|edit-accept. Resume with {action:'edit-accept', editedArtifactData:{...}} exercises RFC 0183 \u00a7A.2's edit-accept arm \u2014 the resolved payload must carry the action and the edited artifact data.",
+  "nodes": [
+    {
+      "id": "gate",
+      "typeId": "core.approvalGate",
+      "name": "Approval Gate",
+      "position": {
+        "x": 0,
+        "y": 0
+      },
+      "config": {
+        "title": "Conformance approval (edit-accept-capable)",
+        "description": "Conformance suite \u2014 please accept to complete the run.",
+        "actions": [
+          "accept",
+          "reject",
+          "edit-accept"
+        ]
+      },
+      "inputs": {}
+    }
+  ],
+  "edges": [],
+  "triggers": [
+    {
+      "id": "manual",
+      "type": "manual",
+      "enabled": true
+    }
+  ],
+  "variables": [],
+  "metadata": {
+    "tags": [
+      "conformance",
+      "hitl"
+    ]
+  },
+  "settings": {
+    "timeout": 0
+  }
+}

package/fixtures/conformance-approval-refine.json

@@ -0,0 +1,45 @@
+{
+  "id": "conformance-approval-refine",
+  "name": "Conformance: Approval (refine)",
+  "version": "1.0",
+  "description": "Suspends on an approval gate offering accept|reject|refine. Resume with {action:'refine', refineFeedback:{...}} exercises RFC 0183 \u00a7A.1/\u00a7A.2 \u2014 the resolved payload must carry the action and its structured feedback.",
+  "nodes": [
+    {
+      "id": "gate",
+      "typeId": "core.approvalGate",
+      "name": "Approval Gate",
+      "position": {
+        "x": 0,
+        "y": 0
+      },
+      "config": {
+        "title": "Conformance approval (refine-capable)",
+        "description": "Conformance suite \u2014 please accept to complete the run.",
+        "actions": [
+          "accept",
+          "reject",
+          "refine"
+        ]
+      },
+      "inputs": {}
+    }
+  ],
+  "edges": [],
+  "triggers": [
+    {
+      "id": "manual",
+      "type": "manual",
+      "enabled": true
+    }
+  ],
+  "variables": [],
+  "metadata": {
+    "tags": [
+      "conformance",
+      "hitl"
+    ]
+  },
+  "settings": {
+    "timeout": 0
+  }
+}

package/fixtures.md

@@ -47,6 +47,8 @@ All fixtures MUST advertise:
 | Delay                                     | `conformance-delay`                                                                             | Verifies poll/SSE behavior over time                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                      | `completed`                                                                         | ≤ 30s (input-controlled)     |
 | Failure                                   | `conformance-failure`                                                                           | Verifies error-event surface                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                              | `failed`                                                                            | ≤ 5s                         |
 | Approval                                  | `conformance-approval`                                                                          | Verifies HITL approval interrupt + resume                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                 | `completed` after resolve                                                           | unbounded (suspends)         |
+| Approval (refine)                         | `conformance-approval-refine`                                                                   | RFC 0183 — refine resolution carries action + refineFeedback                                                        |
+| Approval (edit-accept)                    | `conformance-approval-edit-accept`                                                                 | RFC 0183 — edit-accept resolution carries action + editedArtifactData                                                        |
 | Clarification                             | `conformance-clarification`                                                                     | Verifies HITL clarification interrupt + resume                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                            | `completed` after resolve                                                           | unbounded (suspends)         |
 | Multi-node                                | `conformance-multi-node`                                                                        | Verifies edge ordering + per-node events                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                  | `completed`                                                                         | ≤ 10s                        |
 | Idempotent                                | `conformance-idempotent`                                                                        | Verifies `Idempotency-Key` cache                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                          | `completed`                                                                         | ≤ 5s                         |
@@ -177,6 +179,32 @@ The `messages`-mode stream fixture (AI token streaming) is covered by the determ
 - **Terminal status (after accept)**: `completed`.
 - **Resolve schema**: `{action: "accept" | "reject"}`. Server MUST reject any other shape with 400.
 
+### `conformance-approval-refine`
+
+- **Purpose**: verify a `refine` resolution round-trips its action and structured feedback (RFC 0183 §A.1/§A.2).
+- **Inputs**: none.
+- **Behavior**:
+  1. Run starts and reaches an `approvalGate` offering `accept | reject | refine`.
+  2. Run status MUST be `waiting-approval`.
+  3. Client POSTs `{action: 'refine', refineFeedback: {scope: 'whole', ...}}` to the interrupt.
+  4. The resolved event payload MUST carry `action: 'refine'` and the `refineFeedback` supplied.
+  5. A `refine` resolution supplying no `refineFeedback` MUST be refused.
+- **Why it is separate from `conformance-approval`**: adding `refine` to that fixture's `actions` would change a
+  registered workflow definition every host already serves, forcing a re-registration for a test-only widening.
+  A new id costs one catalog row and no host churn.
+
+### `conformance-approval-edit-accept`
+
+- **Purpose**: verify an `edit-accept` resolution round-trips its action and the edited artifact (RFC 0183 §A.2, the arm `conformance-approval-refine` does not reach).
+- **Inputs**: none.
+- **Behavior**:
+  1. Run starts and reaches an `approvalGate` offering `accept | reject | edit-accept`.
+  2. Run status MUST be `waiting-approval`.
+  3. Client POSTs `{action: 'edit-accept', editedArtifactData: {...}}` to the interrupt.
+  4. The resolved event payload MUST carry `action: 'edit-accept'` and the `editedArtifactData` supplied.
+  5. An `edit-accept` resolution supplying no `editedArtifactData` MUST be refused.
+- **Why it is separate**: same reason as `conformance-approval-refine` — widening a registered fixture's `actions` forces every host to re-register a test-only change.
+
 ### `conformance-clarification`
 
 - **Purpose**: verify HITL clarification interrupt + resume.
@@ -487,6 +515,8 @@ conformance/
     conformance-delay.json
     conformance-failure.json
     conformance-approval.json
+    conformance-approval-refine.json
+    conformance-approval-edit-accept.json
     conformance-clarification.json
     conformance-multi-node.json
     conformance-idempotent.json

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@openwop/openwop-conformance",
-  "version": "2.1.7",
+  "version": "2.2.1",
   "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": "2.1.7"
+    "@openwop/spec-artifacts": "2.2.1"
   }
 }

package/requirements.json

@@ -2,11 +2,11 @@
   "$comment": "GENERATED by conformance/scripts/generate-requirement-registry.mjs — do not edit. One record per it()/test() in src/scenarios. Ids: openwop.it.<file-stem>.<title-slug>[~n] (src/lib/requirement-ids.ts). A record with id null has an interpolated title; its run-time row is keyed by the rendered title and maps here by file+line only. Renamed ids need a row in requirement-aliases.json.",
   "generatedFrom": "src/scenarios/*.test.ts",
   "counts": {
-    "files": 566,
-    "tests": 2156,
-    "withStableId": 2156,
+    "files": 567,
+    "tests": 2161,
+    "withStableId": 2161,
     "interpolatedTitles": 0,
-    "explicitIds": 2111
+    "explicitIds": 2116
   },
   "records": [
     {
@@ -14544,7 +14544,7 @@
     {
       "id": "openwop.it.interrupt-approval.run-suspends-at-gate-accept-resolution-drives-terminal-completed",
       "file": "interrupt-approval.test.ts",
-      "line": 21,
+      "line": 27,
       "title": "run suspends at gate, accept resolution drives terminal completed",
       "explicitId": "openwop.it.interrupt-approval.run-suspends-at-gate-accept-resolution-drives-terminal-completed",
       "citations": [
@@ -14565,7 +14565,7 @@
     {
       "id": "openwop.it.interrupt-approval.400-or-422-when-action-is-not-in-accept-reject",
       "file": "interrupt-approval.test.ts",
-      "line": 50,
+      "line": 56,
       "title": "400 (or 422) when action is not in {accept, reject}",
       "explicitId": "openwop.it.interrupt-approval.400-or-422-when-action-is-not-in-accept-reject",
       "citations": [
@@ -14578,7 +14578,7 @@
     {
       "id": "openwop.it.interrupt-approval.400-404-when-nodeid-does-not-match-an-active-interrupt",
       "file": "interrupt-approval.test.ts",
-      "line": 78,
+      "line": 84,
       "title": "400/404 when nodeId does not match an active interrupt",
       "explicitId": "openwop.it.interrupt-approval.400-404-when-nodeid-does-not-match-an-active-interrupt",
       "citations": [
@@ -14588,6 +14588,82 @@
         }
       ]
     },
+    {
+      "id": "openwop.it.interrupt-approval.a-refine-resolve-round-trips-action-and-the-structured-feedback-it-requires",
+      "file": "interrupt-approval.test.ts",
+      "line": 120,
+      "title": "a refine resolve round-trips `action` and the structured feedback it requires",
+      "explicitId": "openwop.it.interrupt-approval.a-refine-resolve-round-trips-action-and-the-structured-feedback-it-requires",
+      "citations": [
+        {
+          "section": "RFCS/0183-interrupt-resolved-action-fidelity.md §A.1",
+          "requirement": "a gate offering `refine` MUST accept a refine resolution carrying refineFeedback"
+        },
+        {
+          "section": "RFCS/0183-interrupt-resolved-action-fidelity.md §A.1",
+          "requirement": "the resolution MUST be recorded as an event"
+        },
+        {
+          "section": "RFCS/0183-interrupt-resolved-action-fidelity.md §A.1",
+          "requirement": "the resolved payload MUST carry `action: \"refine\"` — the seat RFC 0183 adds"
+        },
+        {
+          "section": "RFCS/0183-interrupt-resolved-action-fidelity.md §A.2",
+          "requirement": "a refine resolution MUST carry the refineFeedback it was given — an action whose meaning is incomplete without it"
+        }
+      ]
+    },
+    {
+      "id": "openwop.it.interrupt-approval.refuses-a-refine-resolution-that-supplies-no-refinefeedback",
+      "file": "interrupt-approval.test.ts",
+      "line": 170,
+      "title": "refuses a refine resolution that supplies no refineFeedback",
+      "explicitId": "openwop.it.interrupt-approval.refuses-a-refine-resolution-that-supplies-no-refinefeedback",
+      "citations": [
+        {
+          "section": "RFCS/0183-interrupt-resolved-action-fidelity.md §A.2",
+          "requirement": "refine without refineFeedback MUST be refused — the feedback is what the action means"
+        }
+      ]
+    },
+    {
+      "id": "openwop.it.interrupt-approval.an-edit-accept-resolve-round-trips-action-and-the-edited-artifact-it-requires",
+      "file": "interrupt-approval.test.ts",
+      "line": 197,
+      "title": "an edit-accept resolve round-trips `action` and the edited artifact it requires",
+      "explicitId": "openwop.it.interrupt-approval.an-edit-accept-resolve-round-trips-action-and-the-edited-artifact-it-requires",
+      "citations": [
+        {
+          "section": "RFCS/0183-interrupt-resolved-action-fidelity.md §A.2",
+          "requirement": "a gate offering `edit-accept` MUST accept an edit-accept resolution carrying editedArtifactData"
+        },
+        {
+          "section": "RFCS/0183-interrupt-resolved-action-fidelity.md §A.1",
+          "requirement": "the resolution MUST be recorded as an event"
+        },
+        {
+          "section": "RFCS/0183-interrupt-resolved-action-fidelity.md §A.1",
+          "requirement": "the resolved payload MUST carry `action: \"edit-accept\"`"
+        },
+        {
+          "section": "RFCS/0183-interrupt-resolved-action-fidelity.md §A.2",
+          "requirement": "an edit-accept resolution MUST carry the editedArtifactData it was given"
+        }
+      ]
+    },
+    {
+      "id": "openwop.it.interrupt-approval.refuses-an-edit-accept-resolution-that-supplies-no-editedartifactdata",
+      "file": "interrupt-approval.test.ts",
+      "line": 247,
+      "title": "refuses an edit-accept resolution that supplies no editedArtifactData",
+      "explicitId": "openwop.it.interrupt-approval.refuses-an-edit-accept-resolution-that-supplies-no-editedartifactdata",
+      "citations": [
+        {
+          "section": "RFCS/0183-interrupt-resolved-action-fidelity.md §A.2",
+          "requirement": "edit-accept without editedArtifactData MUST be refused — the edit is what the action means"
+        }
+      ]
+    },
     {
       "id": "openwop.it.interrupt-approver-routing.capabilities-schema-declares-interrupt-approverrouting",
       "file": "interrupt-approver-routing.test.ts",
@@ -21620,7 +21696,7 @@
     {
       "id": "openwop.it.replay-fanout-suppression.delivers-for-a-live-run-suppresses-for-a-replay-fork-and-delivers-again-for-a-br",
       "file": "replay-fanout-suppression.test.ts",
-      "line": 159,
+      "line": 205,
       "title": "delivers for a live run, suppresses for a replay fork, and delivers again for a branch fork",
       "explicitId": "openwop.it.replay-fanout-suppression.delivers-for-a-live-run-suppresses-for-a-replay-fork-and-delivers-again-for-a-br",
       "citations": [
@@ -25930,6 +26006,49 @@
         }
       ]
     },
+    {
+      "id": "openwop.it.v2-bound-id-path-projection.a-tenant-bound-id-is-readable-at-its-escaped-path-segment-links-carry-that-form",
+      "file": "v2-bound-id-path-projection.test.ts",
+      "line": 51,
+      "title": "a tenant-bound id is readable at its ~-escaped path segment, links carry that form, and a malformed escape is refused",
+      "explicitId": "openwop.requirement.0184.bound-id-path-projection",
+      "citations": [
+        {
+          "section": "spec/v2/core/identity.md §5",
+          "requirement": null,
+          "interpolated": true
+        },
+        {
+          "section": "spec/v2/core/identity.md §5",
+          "requirement": null,
+          "interpolated": true
+        },
+        {
+          "section": "spec/v2/core/identity.md §5",
+          "requirement": "the run read at the projected segment MUST be the run that was created — a host that decodes the escape to a DIFFERENT id has a non-injective decoder"
+        },
+        {
+          "section": "spec/v2/core/identity.md §5",
+          "requirement": null,
+          "interpolated": true
+        },
+        {
+          "section": "spec/v2/core/identity.md §5",
+          "requirement": null,
+          "interpolated": true
+        },
+        {
+          "section": "spec/v2/core/identity.md §5",
+          "requirement": null,
+          "interpolated": true
+        },
+        {
+          "section": "spec/v2/core/identity.md §5",
+          "requirement": null,
+          "interpolated": true
+        }
+      ]
+    },
     {
       "id": "openwop.it.v2-bundle-v3-signed.a-signed-v3-bundle-validates-against-the-closed-root-schema-an-extra-root-key-is",
       "file": "v2-bundle-v3-signed.test.ts",

package/scenario-majors.json

@@ -1,9 +1,9 @@
 {
   "$comment": "GENERATED by conformance/scripts/generate-scenario-majors.mjs (RFC 0168 §D.3). Do not edit; add a file to BOTH_MAJORS in the generator to target both majors.",
   "counts": {
-    "files": 517,
+    "files": 518,
     "v1": 445,
-    "v2": 74
+    "v2": 75
   },
   "majors": {
     "a2a-1-0-agent-card.test.ts": [
@@ -1244,6 +1244,9 @@
     "v2-assurance-downgrade-audited.test.ts": [
       2
     ],
+    "v2-bound-id-path-projection.test.ts": [
+      2
+    ],
     "v2-bundle-v3-signed.test.ts": [
       2
     ],

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": "2.1.7",
-  "corpusTag": "v2.1.7",
+  "version": "2.2.1",
+  "corpusTag": "v2.2.1",
   "files": {
     "api/.redocly.lint-ignore.yaml": "bf5a8350b88a72fa43f59605ed8d903ed24b6cfccda5e45509c9f6ed9ee4e712",
     "api/asyncapi.yaml": "d5ecb9ee6114582be3b1f662c84bfac9ae96dae7bacb853e461168f70a8e1c7d",
     "api/grpc/openwop.proto": "c3e72bb17cba514ee98feb6434e6c9b6ea6795bfd086489ec69fd882dd1ad977",
     "api/openapi.yaml": "39081c59fb696159806b0f2f9a42e7e9ff830d622fcf2ad4159b21357580a955",
     "api/redocly.yaml": "b0604c89b2ca6d5076ec25725c539dad44a741a811fe524439ee6daef8baa09f",
-    "api/seams-v2.yaml": "a848487c183fff6a6f9117932245c2c7765b5b7da8cad1d0a6d02e669ef1b40c",
-    "api/v2/asyncapi.yaml": "66440a6d3f136c5d12c9ebe4dad80e993f7d8fcf5c6203a5694f06ce45f977df",
-    "api/v2/openapi.yaml": "0ca4bdcc6572f427369e1d55b7d3e981668cc424f4bc84f314d6d25170964d51",
+    "api/seams-v2.yaml": "a309a537a5b0aaa20c5f496943bb8cc6642b143eb27b7481ee0b9e1030d00bfb",
+    "api/v2/asyncapi.yaml": "1f7023d581d02dba0427630dfd89ec890941ecbbc31bf8f09ce7ec8b965988b2",
+    "api/v2/openapi.yaml": "f26603003e562b928c148ebaebe49b93c9bd35c01e0996611311114908858e5d",
     "api/v2/redocly.yaml": "1e66b60e6118ad11a823bb620678be464d99dfe50a40e3e6f93ec9429b88b34c",
     "schemas/README.md": "0c0b737ffcf8f30e7d2809cec8a498232de710f41443212922ad8337cdde0b51",
     "schemas/a2a-task-state.schema.json": "c9365918f993f943b4b619d42551eb066a1ed33a08d895d51b395432a5b1f1bc",
@@ -82,7 +82,7 @@
     "schemas/residency.schema.json": "b45fdc91ae88091251914dbd6787a1e2ce687580b91d31a759217b7a1f0010c0",
     "schemas/run-ancestry-response.schema.json": "963a5763b03216aa72fc2854560c778e53994f8cb06ea9a4053e541273c87f5c",
     "schemas/run-diff-response.schema.json": "d9d196730b5d5d8a04a591d21a8fc6b44be7fa134f11827ff7e4f1a86d47e83d",
-    "schemas/run-event-payloads.schema.json": "37cdbdb5038e148056c104a49b8d28a9668479e06d401309d354a8b0a70921a3",
+    "schemas/run-event-payloads.schema.json": "2f2a5192a6262758bbffb358d2044ebe4ad234b388ddcd47b818602897efb93f",
     "schemas/run-event.schema.json": "c70c06932b9831295042836213e617c8e946f20ddd49cf4e889b74d89d2175a8",
     "schemas/run-options.schema.json": "f8241ece6ddc6527de9160d811d42c231b4f77b935f2e107b3c05608545c59ab",
     "schemas/run-orchestrator-decided-event.schema.json": "17a43a3c44680a24d411141357a867b3caed6107d56de88fe2e062eaf93fa2c1",
@@ -149,7 +149,7 @@
     "schemas/v2/goal.schema.json": "146c4182e36d3538778f947e668a92cdb69ec42ee0820104b00063b035e342d7",
     "schemas/v2/heartbeat-evaluated.schema.json": "4086b1ba1d0547c08b038dfb263ac89416a3e406430db1250156fb2f9b74f8d6",
     "schemas/v2/heartbeat-state-changed.schema.json": "eb68d0082aa996650e3a6b8ccf32457f959cebc7e6728cc7c331b9d171a4bef6",
-    "schemas/v2/ids.schema.json": "05c98e3f8a72fa79be123765bd9c2ea7c6620188d69df8524160399225316cd6",
+    "schemas/v2/ids.schema.json": "6481cab16c9b89fab168627502dcbf2ee7489b5a636a4080ea8cfd2670d7b17e",
     "schemas/v2/localized-content-language-settings.schema.json": "92240f19a2b4b2f05ab41a797f703485a28813848035478e72f395edd04c1d35",
     "schemas/v2/localized-content-page-response.schema.json": "ef48329749758afb21e4d3cbc3f017abb911f75726819457f35b9f1390332c75",
     "schemas/v2/localized-content-page.schema.json": "e0f97660121955d6705d935f34e748adb739ac70f8536a009cd8d6f303735151",
@@ -169,7 +169,7 @@
     "schemas/v2/residency.schema.json": "3fd23c9732beb140504e81560fc426eb6764d547e56da5fb92c5a5c7bf04db3a",
     "schemas/v2/run-ancestry-response.schema.json": "334c17f2062dea62726ca1d07b8be3f169cdc422af843c1def5fb317039814c9",
     "schemas/v2/run-diff-response.schema.json": "8bdb2045c274580038bd74781f7d062f5049839797c164a89de4ed89edc53a74",
-    "schemas/v2/run-event-payloads.schema.json": "1d3ccffc4e506d3109c17990e01976ba935e61672562684b9635743d56af3f57",
+    "schemas/v2/run-event-payloads.schema.json": "38dd47929f0245c865e4850aab453e073bb717bc81590d68a91bd7b490536d63",
     "schemas/v2/run-event.schema.json": "1a9946374af9323b71635d00fed22abe02c300fa2cf3ae3e9c30f3e553e408ec",
     "schemas/v2/run-list-response.schema.json": "c2dbc34768dcf64528914a213aa3cb55952e8e3e90a64b04a9423b2a19749d38",
     "schemas/v2/run-options.schema.json": "f0e9f494d80086c2ed2cf9fe87df54e5b4e66601699c4a1089f4edbc7ba58867",
@@ -200,11 +200,11 @@
     "schemas/workspace-file.schema.json": "464de85c2a068243084ee9c1d969bc7cd5d8f7948574e58450d6493c38a0e1e4",
     "spec/v1/alias-detectors.json": "fee4594ef49953953ffcd0b3813300067d16b3e65ebff2aac722034ac9b3f545",
     "spec/v1/capability-declaration-classes.json": "e7729aed5c4b4e1dd02abab0530f14cc95f5d4070fe51fb139e7f5cccefa00c6",
-    "spec/v1/core-standard-manifest.json": "de593dd613393b6d93d90c8e76032466aaa68379a53eaf274fcd921264e51ead",
+    "spec/v1/core-standard-manifest.json": "9d76a090508697628031972795c813a12ff83b53ed1dff2dbaabe14d853b90c2",
     "spec/v1/deprecations.json": "a2022c2d6eb501bf406b5d75fe2d431a3fe0158e8509125d2453985eb1bf511a",
     "spec/v1/deprecations.schema.json": "18c87e78bedc210431f795ae44c5b5d202f2f3317850d5cf86867d4f1fa1cdfb",
     "spec/v1/event-codemap.json": "3da60d884157793a360da532a9fcbbfb5285636db325a74cec94b34622186d97",
-    "spec/v1/event-codemap.schema.json": "d52caf29071e3784d4c20259cdbd5a7b533986548a440bb632a80b82cadd7ea5",
+    "spec/v1/event-codemap.schema.json": "d05933b2e88103aff51a2f774df97b9bda065e0fc88dbd9b114cc47a32189174",
     "spec/v1/extensions.json": "79a60754aa16cbdbdb604f8e5c00038af535a1690a4d07bd5ff7c38f0256a7bc",
     "spec/v1/gaps.json": "96ca051cb20f1a669b355d67a89421c56fef28aa5e6710c9519f592f26af33a3",
     "spec/v1/gaps.schema.json": "8fd83259f556553c9df0f53e7a82ca8c2a4771d471197f69b4e2ae0ceeacfd66",
@@ -212,18 +212,18 @@
     "spec/v1/migrations.schema.json": "886779aa6c22e646db097f5df210adb018a4dd14a7b815465a18c8a7056c8f72",
     "spec/v1/operation-path-manifest.json": "5f5f4e3842669371730ebbd1aace3fb794192615f0018dc144f484ba5db82ac3",
     "spec/v1/spec-gaps.json": "6cc9962c6b969f632e07a78b52a4f61447ff579e2990cbae989866f584a86042",
-    "spec/v2/README.md": "e3d8f6a13338014c349e57bf2b50a398aa1e0aa162c5adda0273e26dd6f48acd",
+    "spec/v2/README.md": "0f8a5c36ca9af5c6aa80b2e24ecf2d6af5ede9964bf70e88839110d9beab5693",
     "spec/v2/core/capabilities.md": "c5107a4964e487c20bbb5254cfe1f706e0d9ba7e34da9070bc43b50b385e7948",
     "spec/v2/core/conformance.md": "622084e3c74c4549087f4d861b1bb98cec201a0c5aec23788d27f53a355fde3b",
     "spec/v2/core/connection-packs.md": "1d759681e2962f5103228c52c45f24607ed4bb9ae73e30605f690c1d5a59e9ee",
     "spec/v2/core/errors.md": "74fdb5862f432a831cf0c2eadbe0c8cca6b2660ec578180ae33db54773a1df28",
-    "spec/v2/core/events.md": "defe0969d6ec7c980f415cc0d78b9d065782ac7e5638b12a44aa4895a35d6126",
+    "spec/v2/core/events.md": "3b7d1ad34f9e438c9825d60c0114b9a21c82a468f15ccbd7567022641a4537ad",
     "spec/v2/core/form-content-packs.md": "6ffaae85d4550be48dd53fb6e2959af8aceb35bd87c5343b722885084f197713",
-    "spec/v2/core/headers.md": "a795c8cd3e5f004b05dad9603d14b92b639a24ccd23b183716cb01097a106119",
+    "spec/v2/core/headers.md": "2dd67d71673539e16612f09f3f2513a384f6566052d7cd8a00885885689854c2",
     "spec/v2/core/idempotency.md": "b86c352d964aebad532727f314e26ed2177d49de382076fa64d3162e2f56118d",
-    "spec/v2/core/identity.md": "dd5299f9cdf54640870e1fffae11edb820cf6396ea298c4ebd14c649c4bd4a87",
+    "spec/v2/core/identity.md": "9d2b823eb899463807452fe5721dcf09881837c161be024175957db27870b243",
     "spec/v2/core/interop.md": "d4c404e469e1d9d9be88bc2c939d777e7d7c1d0419b1251c18770f039838a401",
-    "spec/v2/core/interrupt.md": "d872e0652a938594eeb2a19e86cf64e64e29b234f5d9c4fbe2970ab5fb191b9c",
+    "spec/v2/core/interrupt.md": "f525479a99104655219fa2e51c4cad2a03d8e3be93ff4dac173a684675e57348",
     "spec/v2/core/overview.md": "69377c90bbb87b6bfe4b8be4f1b65810094b0759c8d4f93ca6df92cc530ea023",
     "spec/v2/core/packs.md": "e72a7586a61630cdfd342eaf84e989f864b3e199006f5511ef59ea7c1a1c56f7",
     "spec/v2/core/persistence.md": "2373c2815b25b7b4fe29dfdf3783d3f2f3e637d3abce5085e69f019546302fb6",
@@ -233,11 +233,11 @@
     "spec/v2/core/versioning.md": "d731782eeabb953919c7c7057340b7f9c2916b6186790e6b04c7d8ea60767189",
     "spec/v2/core/webhooks.md": "d360da78336afa837bddf515e289e445808bca02e1e4b493f81fa51ce4a545c4",
     "spec/v2/core/workflow-chain-packs.md": "ef2b07d2d99334eaec0cc370314e2ad2f84ea58deb699e5d92e6e52d6056613f",
-    "spec/v2/declaration.json": "63dd351a87439241bc744ce62fdc498b7621292c11a4ad42c3f41429d72366b9",
-    "spec/v2/declaration.schema.json": "eac5f8080bd572f147bd57d0c4ac8ad73b14536d9236222490ae29a33b21e184",
+    "spec/v2/declaration.json": "a09bac61c1b825e4a77b5abc0f86f1f8c20c381e0650eb08c2f6d0ef0aaec1dc",
+    "spec/v2/declaration.schema.json": "c86e3083fb9789c1958ecb4576cc6f71aa90e6f339521139238543ed08211dd4",
     "spec/v2/errors.json": "f179414a92f30b5dadf26e9e57699649a67324bd7186d5287647d17605d137d3",
     "spec/v2/event-codemap.json": "37869a03cdfd8758b57e6f350641a6ebbd383231cebda51f5473ad85f94ab74c",
-    "spec/v2/event-codemap.schema.json": "ce2d0ca085d7668d7ef745d2560b385ec6881d24a6d4b47c49931986b2ba642d",
+    "spec/v2/event-codemap.schema.json": "b173f2a9bcc0bf9b62a474ce3b91e1431d597e64fd1560fc58662cd8606eaf9c",
     "spec/v2/ext/a2uiSurface/README.md": "176cb0542e1c236ffcd6d682536ba0081242684444e2dec790abd480a44a571e",
     "spec/v2/ext/brand/README.md": "1a80102a9cd766dcb6098c999e552af88f5b83c88951acde18d17cc0b29aa09d",
     "spec/v2/ext/canvas/README.md": "a2d5f28523753e8e85cc9042db39faf9207a2eebce1a8553be22e341f7ea58d9",
@@ -266,12 +266,12 @@
     "spec/v2/facets/runList.schema.json": "6223e1baaebbb9f4faa71bfe968f8d22b4aa88046d0e43e8d5f56409e7f86e90",
     "spec/v2/facets/sandbox.schema.json": "e75f54486d387b2d089791aa93b5093cd1450f1920753a0f97fc23acdfe978ed",
     "spec/v2/facets/webhooks.schema.json": "0b766033ef4b4c6c2ac6f9979a19843ba024d3a1d1576972dcb55aaba52f1b9a",
-    "spec/v2/id-field-bindings.json": "d9d1f8e266660b581f11020b9511a65c677e026b1a71ada870682ad77e18555d",
+    "spec/v2/id-field-bindings.json": "eecadfd11581ba2e075ccb320d85fc033b816edfdd51edb5c5727a181e55dcb4",
     "spec/v2/path-manifest.json": "a6f4e652e866889e8bde815874062f62929aee95dc8e58e7152c56adee262163",
-    "spec/v2/peer-dependency-aliases.json": "d10299280abee08258502925bc327293ee413e0108cd6e6ec75ff6110653308d",
+    "spec/v2/peer-dependency-aliases.json": "10fcab53c65305e9b6bae4a40d30176091dca68739ac8a78291510c77b50ca6b",
     "spec/v2/profiles.json": "0636f19fceae625390003a347e70ef4797d84766b5c24ce8a02cea52aadebca4",
-    "spec/v2/release.json": "2bacef4931ec682b7bf003746eb1f69e3b945bbb93b98d76ddf86128e9922be5",
+    "spec/v2/release.json": "ee596fce1d27e7ec3f7a061b5705ca8022158adee5ba1aa7f321ee9abd97e9fb",
     "spec/v2/retention-floors.json": "eaf3722d95c79947af1d4269ef85117e126518c588cfcf1a2b21b97269f51624"
   },
-  "corpusCommit": "4c50dd0dc8b411090dfeeb3a5bbf9466474807c9"
+  "corpusCommit": "a27e18972293865c72e4a776426718be6e2d77cc"
 }

package/src/lib/bound-id.ts

@@ -0,0 +1,69 @@
+/**
+ * The bound-id path projection (`spec/v2/core/identity.md` §5, RFC 0184 §A.1).
+ *
+ * A tenant-bound id is TWO segments joined by `/`; a URL path parameter is ONE
+ * segment. Something has to carry the separator across, and the corpus used to
+ * say `%2F`. That is what a tier-1 host's front door decoded back to `/` before
+ * forwarding, leaving every bound id unreachable at its own origin
+ * (`v2-created-run-readable.test.ts` records the measurement).
+ *
+ * So the escape marker is `~`, which RFC 3986 §2.3 lists as UNRESERVED. An
+ * intermediary has no license to rewrite an unreserved character, so `~2F`
+ * arrives byte-for-byte. `%2F` has no such protection in practice: handling it
+ * correctly requires a front door to distinguish a percent-encoded RESERVED
+ * octet (§6.2.2.2: MUST NOT decode) from an unreserved one (SHOULD decode), and
+ * deployed front doors do not.
+ *
+ * The codec is TOTAL over bytes rather than conditional on the current id
+ * grammar. That is deliberate: a conditional encoding is only unambiguous while
+ * the grammar holds still, and this grammar has already moved once (the `anon:`
+ * tenant prefix). Encoding every non-passthrough byte means a later widening
+ * cannot invalidate a projection already on the wire.
+ */
+
+/** RFC 3986 unreserved MINUS `~`, which is reserved here as the escape marker. */
+const PASSTHROUGH = /^[A-Za-z0-9._-]$/;
+
+/** Encode a bound id into exactly one path segment. Identity on already-safe input. */
+export function projectBoundId(id: string): string {
+  let out = '';
+  for (const byte of new TextEncoder().encode(id)) {
+    const ch = String.fromCharCode(byte);
+    out += byte < 0x80 && PASSTHROUGH.test(ch) ? ch : `~${byte.toString(16).toUpperCase().padStart(2, '0')}`;
+  }
+  return out;
+}
+
+/**
+ * Decode one path segment back to the bound id.
+ *
+ * Throws on a `~` that does not introduce two hex digits — the wire rule is
+ * `400 validation_error`, and a decoder that silently passed a lone `~` through
+ * would make the codec non-injective in exactly the direction that matters.
+ */
+export function unprojectBoundId(segment: string): string {
+  for (let i = 0; i < segment.length; i++) {
+    if (segment[i] !== '~') continue;
+    if (!/^[0-9A-Fa-f]{2}$/.test(segment.slice(i + 1, i + 3))) {
+      throw new Error(`bound-id projection: '~' at index ${i} is not followed by two hex digits`);
+    }
+    i += 2;
+  }
+  const bytes: number[] = [];
+  let i = 0;
+  while (i < segment.length) {
+    if (segment[i] === '~') { bytes.push(parseInt(segment.slice(i + 1, i + 3), 16)); i += 3; }
+    else { bytes.push(...new TextEncoder().encode(segment[i]!)); i += 1; }
+  }
+  return new TextDecoder('utf-8', { fatal: true }).decode(new Uint8Array(bytes));
+}
+
+/**
+ * The tenant-bound id grammar, mirrored from `ids.schema.json` `$defs.runId`.
+ *
+ * Exported because four scenarios each kept their own copy, and copies drift:
+ * all four still spelled the tenant segment `[A-Za-z0-9._~-]{1,128}` after
+ * `tenantId` grew its `anon:` prefix, so every one of them would have rejected
+ * a runId the schema accepts (RFC 0184 §A.3).
+ */
+export const BOUND_ID = /^(anon:)?[A-Za-z0-9._~-]{1,128}\/[A-Za-z0-9._~-]{16,128}$/;

package/src/scenarios/interrupt-approval.test.ts

@@ -17,6 +17,12 @@ const WORKFLOW_ID = 'conformance-approval';
 const NODE_ID = 'gate';
 const SKIP_NO_FIXTURE = !isFixtureAdvertised(WORKFLOW_ID);
 
+const REFINE_WORKFLOW_ID = 'conformance-approval-refine';
+const SKIP_NO_REFINE = !isFixtureAdvertised(REFINE_WORKFLOW_ID);
+
+const EDIT_ACCEPT_WORKFLOW_ID = 'conformance-approval-edit-accept';
+const SKIP_NO_EDIT_ACCEPT = !isFixtureAdvertised(EDIT_ACCEPT_WORKFLOW_ID);
+
 describe.skipIf(SKIP_NO_FIXTURE)('interrupt: approval accept resumes to `completed`', () => {
   it('run suspends at gate, accept resolution drives terminal completed', async () => {
     const create = await driver.post('/v1/runs', { workflowId: WORKFLOW_ID });
@@ -96,3 +102,164 @@ describe.skipIf(SKIP_NO_FIXTURE)('interrupt: resolving an unknown interrupt retu
     });
   });
 });
+
+/**
+ * RFC 0183 §A.1/§A.2 — a `refine` resolution must be recordable.
+ *
+ * Before RFC 0183, `interruptResolved` was `additionalProperties: false` with no
+ * seat for the resume action and no seat for the feedback a refine REQUIRES, and
+ * `decision` was an unconstrained string whose documented vocabulary is a
+ * different axis (governance: granted|rejected|overridden). A host resolving by
+ * refine could record neither what the approver did nor the feedback they gave.
+ *
+ * Uses its own fixture rather than widening `conformance-approval`, because
+ * adding `refine` to that fixture's `actions` would change a registered workflow
+ * definition every host already serves.
+ */
+describe.skipIf(SKIP_NO_REFINE)('interrupt: refine resolution carries action + refineFeedback (RFC 0183)', () => {
+  it('a refine resolve round-trips `action` and the structured feedback it requires', async () => {
+    const create = await driver.post('/v1/runs', { workflowId: REFINE_WORKFLOW_ID });
+    expect(create.status).toBe(201);
+    const runId = (create.json as { runId: string }).runId;
+
+    const suspended = await pollUntilStatus(runId, 'waiting-approval', { timeoutMs: 10_000 });
+    expect(suspended.currentNodeId).toBe(NODE_ID);
+
+    const refineFeedback = { scope: 'whole', text: 'conformance: please tighten the summary' };
+    const resolve = await driver.post(
+      `/v1/runs/${encodeURIComponent(runId)}/interrupts/${encodeURIComponent(NODE_ID)}`,
+      { resumeValue: { action: 'refine', refineFeedback } },
+    );
+    expect(
+      resolve.status,
+      req('openwop.it.interrupt-approval.a-refine-resolve-round-trips-action-and-the-structured-feedback-it-requires',
+        'RFCS/0183-interrupt-resolved-action-fidelity.md §A.1',
+        'a gate offering `refine` MUST accept a refine resolution carrying refineFeedback',
+      ),
+    ).toBe(200);
+
+    const events = await driver.get(`/v1/runs/${encodeURIComponent(runId)}/events`);
+    expect(events.status).toBe(200);
+    const rows = ((events.json as { events?: Array<Record<string, unknown>> } | undefined)?.events ?? []);
+    const resolved = rows.find((e) => String(e.type).endsWith('interrupt.resolved') || String(e.type).endsWith('approval.received'));
+    expect(
+      resolved,
+      req('openwop.it.interrupt-approval.a-refine-resolve-round-trips-action-and-the-structured-feedback-it-requires',
+        'RFCS/0183-interrupt-resolved-action-fidelity.md §A.1',
+        'the resolution MUST be recorded as an event',
+      ),
+    ).toBeDefined();
+
+    const payload = (resolved?.payload ?? {}) as Record<string, unknown>;
+    expect(
+      payload.action,
+      req('openwop.it.interrupt-approval.a-refine-resolve-round-trips-action-and-the-structured-feedback-it-requires',
+        'RFCS/0183-interrupt-resolved-action-fidelity.md §A.1',
+        'the resolved payload MUST carry `action: "refine"` — the seat RFC 0183 adds',
+      ),
+    ).toBe('refine');
+    expect(
+      payload.refineFeedback,
+      req('openwop.it.interrupt-approval.a-refine-resolve-round-trips-action-and-the-structured-feedback-it-requires',
+        'RFCS/0183-interrupt-resolved-action-fidelity.md §A.2',
+        'a refine resolution MUST carry the refineFeedback it was given — an action whose meaning is incomplete without it',
+      ),
+    ).toBeTruthy();
+  });
+
+  it('refuses a refine resolution that supplies no refineFeedback', async () => {
+    const create = await driver.post('/v1/runs', { workflowId: REFINE_WORKFLOW_ID });
+    expect(create.status).toBe(201);
+    const runId = (create.json as { runId: string }).runId;
+    await pollUntilStatus(runId, 'waiting-approval', { timeoutMs: 10_000 });
+
+    const resolve = await driver.post(
+      `/v1/runs/${encodeURIComponent(runId)}/interrupts/${encodeURIComponent(NODE_ID)}`,
+      { resumeValue: { action: 'refine' } },
+    );
+    expect(
+      [400, 422].includes(resolve.status),
+      req('openwop.it.interrupt-approval.refuses-a-refine-resolution-that-supplies-no-refinefeedback',
+        'RFCS/0183-interrupt-resolved-action-fidelity.md §A.2',
+        'refine without refineFeedback MUST be refused — the feedback is what the action means',
+      ),
+    ).toBe(true);
+  });
+});
+
+/**
+ * RFC 0183 §A.2, the `edit-accept` arm — an edit-accept resolution must be
+ * recordable with the artifact the approver edited. The refine block above
+ * witnesses one of §A.2's two conditional MUSTs; until this block the other was
+ * stated in `spec/v2/core/interrupt.md` §Approval and witnessed nowhere.
+ */
+describe.skipIf(SKIP_NO_EDIT_ACCEPT)('interrupt: edit-accept resolution carries action + editedArtifactData (RFC 0183)', () => {
+  it('an edit-accept resolve round-trips `action` and the edited artifact it requires', async () => {
+    const create = await driver.post('/v1/runs', { workflowId: EDIT_ACCEPT_WORKFLOW_ID });
+    expect(create.status).toBe(201);
+    const runId = (create.json as { runId: string }).runId;
+
+    const suspended = await pollUntilStatus(runId, 'waiting-approval', { timeoutMs: 10_000 });
+    expect(suspended.currentNodeId).toBe(NODE_ID);
+
+    const editedArtifactData = { summary: 'conformance: edited by the approver' };
+    const resolve = await driver.post(
+      `/v1/runs/${encodeURIComponent(runId)}/interrupts/${encodeURIComponent(NODE_ID)}`,
+      { resumeValue: { action: 'edit-accept', editedArtifactData } },
+    );
+    expect(
+      resolve.status,
+      req('openwop.it.interrupt-approval.an-edit-accept-resolve-round-trips-action-and-the-edited-artifact-it-requires',
+        'RFCS/0183-interrupt-resolved-action-fidelity.md §A.2',
+        'a gate offering `edit-accept` MUST accept an edit-accept resolution carrying editedArtifactData',
+      ),
+    ).toBe(200);
+
+    const events = await driver.get(`/v1/runs/${encodeURIComponent(runId)}/events`);
+    expect(events.status).toBe(200);
+    const rows = ((events.json as { events?: Array<Record<string, unknown>> } | undefined)?.events ?? []);
+    const resolved = rows.find((e) => String(e.type).endsWith('interrupt.resolved') || String(e.type).endsWith('approval.received'));
+    expect(
+      resolved,
+      req('openwop.it.interrupt-approval.an-edit-accept-resolve-round-trips-action-and-the-edited-artifact-it-requires',
+        'RFCS/0183-interrupt-resolved-action-fidelity.md §A.1',
+        'the resolution MUST be recorded as an event',
+      ),
+    ).toBeDefined();
+
+    const payload = (resolved?.payload ?? {}) as Record<string, unknown>;
+    expect(
+      payload.action,
+      req('openwop.it.interrupt-approval.an-edit-accept-resolve-round-trips-action-and-the-edited-artifact-it-requires',
+        'RFCS/0183-interrupt-resolved-action-fidelity.md §A.1',
+        'the resolved payload MUST carry `action: "edit-accept"`',
+      ),
+    ).toBe('edit-accept');
+    expect(
+      payload.editedArtifactData,
+      req('openwop.it.interrupt-approval.an-edit-accept-resolve-round-trips-action-and-the-edited-artifact-it-requires',
+        'RFCS/0183-interrupt-resolved-action-fidelity.md §A.2',
+        'an edit-accept resolution MUST carry the editedArtifactData it was given',
+      ),
+    ).toEqual(editedArtifactData);
+  });
+
+  it('refuses an edit-accept resolution that supplies no editedArtifactData', async () => {
+    const create = await driver.post('/v1/runs', { workflowId: EDIT_ACCEPT_WORKFLOW_ID });
+    expect(create.status).toBe(201);
+    const runId = (create.json as { runId: string }).runId;
+    await pollUntilStatus(runId, 'waiting-approval', { timeoutMs: 10_000 });
+
+    const resolve = await driver.post(
+      `/v1/runs/${encodeURIComponent(runId)}/interrupts/${encodeURIComponent(NODE_ID)}`,
+      { resumeValue: { action: 'edit-accept' } },
+    );
+    expect(
+      [400, 422].includes(resolve.status),
+      req('openwop.it.interrupt-approval.refuses-an-edit-accept-resolution-that-supplies-no-editedartifactdata',
+        'RFCS/0183-interrupt-resolved-action-fidelity.md §A.2',
+        'edit-accept without editedArtifactData MUST be refused — the edit is what the action means',
+      ),
+    ).toBe(true);
+  });
+});

package/src/scenarios/replay-fanout-suppression.test.ts

@@ -36,6 +36,24 @@
  * operator raising the scale on a slow host makes this scenario harder to pass,
  * not easier.
  *
+ * CORRECTED 2026-09-15 — the paragraph above described a property this file did
+ * not have, and argued for it at length, which is the most expensive kind of
+ * stale comment. The scale knob DID multiply every poll and quiet window, and it
+ * did NOT touch vitest's own `testTimeout` (30_000, `conformance/vitest.config.ts`).
+ * The declared budgets in this one test already summed to ~77.5 s — 10 s + 30 s +
+ * 30 s of polling plus three quiet windows — so at scale 1 the second and third
+ * poll budgets were UNREACHABLE, and raising the scale made the test die at the
+ * same 30 s wall having spent even less of the window it was told to wait.
+ *
+ * So the knob could not make this scenario harder to pass. It could only make it
+ * fail sooner, on the clock, with vitest's generic "Test timed out in 30000ms" in
+ * place of `pollUntil`'s message naming the run, the predicate and the last
+ * status. A host that genuinely re-fired on replay and a host that was merely
+ * slow produced the same output.
+ *
+ * `TEST_TIMEOUT_MS` below now derives the test's own wall clock from the same
+ * budgets and the same scale, so the paragraph above is true as written.
+ *
  * ── Deliberately NOT in the `openwop-replay-fork` floor ──────────────────────
  * This scenario needs the host to accept a loopback receiver. A host with an
  * SSRF guard on `POST /v1/webhooks` correctly refuses one, and the suite's
@@ -155,6 +173,34 @@ const DELIVERY_GRACE_MS = 1_500;
  */
 const SUPPRESSION_WINDOW_MS = 6_000;
 
+/** Poll ceilings, named so the test's own wall clock can be derived from them
+ *  rather than guessed alongside them. */
+const SOURCE_POLL_MS = 10_000;
+const FORK_POLL_MS = 30_000;
+
+/**
+ * THE TEST'S OWN WALL CLOCK, and it must be computed, not defaulted.
+ *
+ * Every budget in this file is a claim about how long a legitimate host may
+ * take. A `testTimeout` smaller than their sum does not tighten those claims —
+ * it DELETES them, silently, because vitest kills the test before the budget it
+ * was given can be spent. The suite default (30_000) is smaller than the first
+ * two budgets alone.
+ *
+ * It scales with `scaledTimeoutMs` for the same reason: an operator who raises
+ * `OPENWOP_POLL_TIMEOUT_SCALE` is saying "this host is slow, wait longer". A
+ * scaled poll budget under an unscaled wall clock turns that instruction into
+ * its opposite. Both move together or neither should.
+ *
+ * The flat slack covers the run submissions, receiver registration and
+ * assertions between the waits — work with no budget of its own. It is added
+ * AFTER scaling deliberately: it is fixed overhead, not host latency, so
+ * multiplying it would inflate the ceiling for a reason that is not the host's.
+ */
+const TEST_TIMEOUT_MS =
+  scaledTimeoutMs(SOURCE_POLL_MS + FORK_POLL_MS * 2 + DELIVERY_GRACE_MS * 2 + SUPPRESSION_WINDOW_MS)
+  + 15_000;
+
 describe('replay-fanout-suppression: a replay fork MUST NOT fan out re-emitted events', () => {
   it('delivers for a live run, suppresses for a replay fork, and delivers again for a branch fork', async (ctx) => {
     const disco = await driver.get('/.well-known/openwop');
@@ -204,7 +250,7 @@ describe('replay-fanout-suppression: a replay fork MUST NOT fan out re-emitted e
     const create = await driver.post('/v1/runs', { workflowId: 'conformance-noop' });
     expect(create.status, req('openwop.it.replay-fanout-suppression.delivers-for-a-live-run-suppresses-for-a-replay-fork-and-delivers-again-for-a-br', 'webhooks.md §"Register"', 'failed to start conformance-noop')).toBe(201);
     const sourceRunId = (create.json as { runId: string }).runId;
-    await pollUntilTerminal(sourceRunId, { timeoutMs: 10_000 });
+    await pollUntilTerminal(sourceRunId, { timeoutMs: SOURCE_POLL_MS });
     await quietWindow(DELIVERY_GRACE_MS);
 
     expect(
@@ -267,7 +313,7 @@ describe('replay-fanout-suppression: a replay fork MUST NOT fan out re-emitted e
     }
     expect(replay.status, req('openwop.it.replay-fanout-suppression.delivers-for-a-live-run-suppresses-for-a-replay-fork-and-delivers-again-for-a-br', 'webhooks.md §"Register"', 'replay fork should be accepted')).toBe(201);
     const replayRunId = (replay.json as { runId: string }).runId;
-    await pollUntilTerminal(replayRunId, { timeoutMs: 30_000 });
+    await pollUntilTerminal(replayRunId, { timeoutMs: FORK_POLL_MS });
     await quietWindow(SUPPRESSION_WINDOW_MS);
 
     expect(
@@ -315,7 +361,7 @@ describe('replay-fanout-suppression: a replay fork MUST NOT fan out re-emitted e
     }
     expect(branch.status, req('openwop.it.replay-fanout-suppression.delivers-for-a-live-run-suppresses-for-a-replay-fork-and-delivers-again-for-a-br', 'replay.md §"Host-initiated fan-out is an external effect"', 'branch fork should be accepted')).toBe(201);
     const branchRunId = (branch.json as { runId: string }).runId;
-    await pollUntilTerminal(branchRunId, { timeoutMs: 30_000 });
+    await pollUntilTerminal(branchRunId, { timeoutMs: FORK_POLL_MS });
     await quietWindow(DELIVERY_GRACE_MS);
 
     expect(
@@ -326,5 +372,7 @@ describe('replay-fanout-suppression: a replay fork MUST NOT fan out re-emitted e
           + 'means the host keyed on "is a fork" rather than on replay-ness read from the run',
       ),
     ).toBeGreaterThan(0);
-  });
+    // The third positional argument is the whole point of this fix: without it
+    // the suite default (30_000) silently overrides every budget above.
+  }, TEST_TIMEOUT_MS);
 });

package/src/scenarios/v2-bound-id-path-projection.test.ts

@@ -0,0 +1,121 @@
+/**
+ * `spec/v2/core/identity.md` §5 + RFC 0184 §A.1 — a tenant-bound id travels as
+ * ONE path segment under the `~`-escape projection (suite 2.2.0, target major
+ * 2; unaided; creates one run).
+ *
+ * A tenant-bound id is two segments joined by `/`. A path parameter is one. The
+ * corpus used to say `%2F` carries the separator across, and
+ * `v2-created-run-readable.test.ts` records what that cost: a tier-1 host's
+ * hosting layer decoded `%2F` back to `/` before forwarding, the backend
+ * correctly had no route for a literal slash, and EVERY bound id was
+ * unreachable through the host's own front door. One hop behind, the direct
+ * service URL answered 200.
+ *
+ * The projection escapes with `~`, which RFC 3986 §2.3 lists as UNRESERVED — an
+ * intermediary has no license to rewrite it, so `~2F` reaches the origin
+ * byte-for-byte. That is the whole argument: `%2F` needs a front door to tell a
+ * percent-encoded RESERVED octet (§6.2.2.2 says MUST NOT decode) from an
+ * unreserved one (SHOULD decode), and deployed front doors do not.
+ *
+ * **What this file does NOT assert.** The percent-encoded form still MUST work
+ * — `identity.md` §5 keeps it, removing it would be breaking, and
+ * `v2-created-run-readable` already witnesses it. Duplicating it here would red
+ * two rows for one defect and tell a bundle reader nothing new.
+ *
+ * The codec's own edge cases (marker escaping, UTF-8 vs UTF-16, malformed
+ * decode) are a unit concern and live in `src/lib/bound-id.test.ts`, which is
+ * sabotage-checked. This file asserts only what needs a HOST to answer.
+ *
+ * @see spec/v2/core/identity.md §5
+ * @see RFCS/0184-bound-id-path-projection.md §A.1
+ */
+
+import { describe, it, expect } from 'vitest';
+import { driver, type OpenWOPResponse } from '../lib/driver.js';
+import { v2Discovery } from '../lib/v2.js';
+import { readErrorCode } from '../lib/error-envelope.js';
+import { softSkip } from '../lib/soft-skip.js';
+import { req } from '../lib/requirement-ids.js';
+import { projectBoundId } from '../lib/bound-id.js';
+import { BOUND_ID as BOUND } from '../lib/bound-id.js';
+
+const ID = 'openwop.requirement.0184.bound-id-path-projection';
+const DOC = 'spec/v2/core/identity.md §5';
+const NOOP_WORKFLOW_ID = 'conformance-noop';
+
+async function http(fn: () => Promise<OpenWOPResponse>): Promise<OpenWOPResponse | null> {
+  try { return await fn(); } catch { return null; }
+}
+
+describe('v2 bound-id path projection (identity.md §5)', () => {
+  it('a tenant-bound id is readable at its ~-escaped path segment, links carry that form, and a malformed escape is refused', async () => {
+    try { if (!(await v2Discovery())) return softSkip('blocked', 'v2 discovery unreachable'); } catch { return softSkip('blocked', 'v2 discovery unreachable'); }
+
+    const created = await http(() => driver.post('/runs', { workflowId: NOOP_WORKFLOW_ID }));
+    if (created === null) return softSkip('blocked', 'POST /runs unreachable (fetch failed)');
+    if (created.status === 429) return softSkip('blocked', 'POST /runs answered 429 — the run budget, not the wire');
+    const body = (created.json ?? {}) as { runId?: unknown; eventsUrl?: unknown };
+    if (created.status !== 201 || typeof body.runId !== 'string') {
+      return softSkip('blocked', `POST /runs {workflowId: ${NOOP_WORKFLOW_ID}} answered ${created.status} ${readErrorCode(created.json) ?? ''} — the smallest valid create was refused (fixture not seeded?)`.trim());
+    }
+    const runId = body.runId;
+    if (!BOUND.test(runId)) return softSkip('blocked', `the created runId is not tenant-bound (${runId}) — nothing to project`);
+
+    const projected = projectBoundId(runId);
+
+    // The projection MUST leave nothing for an intermediary to decode. If this
+    // fails the encoder is wrong, not the host — assert it before blaming a 404.
+    expect(
+      encodeURIComponent(projected),
+      req(ID, DOC, `the projection MUST contain only RFC 3986 unreserved characters, so no intermediary can rewrite it (${projected})`),
+    ).toBe(projected);
+
+    // ACCEPT SIDE (the new MUST). A 404 here means the host has not implemented
+    // the projection: it is not a routing accident, because the segment that
+    // reached it is byte-identical to the one sent.
+    const read = await http(() => driver.get(`/runs/${projected}`));
+    expect(
+      read?.status ?? null,
+      req(ID, DOC, `GET /runs/{projected} MUST answer 200 — the host MUST accept a tenant-bound id as one ~-escaped segment. Got ${read?.status ?? 'no response'} ${readErrorCode(read?.json) ?? ''} for ${projected} (runId ${runId})`.trim()),
+    ).toBe(200);
+    expect(
+      (read?.json as { runId?: unknown } | undefined)?.runId,
+      req(ID, DOC, 'the run read at the projected segment MUST be the run that was created — a host that decodes the escape to a DIFFERENT id has a non-injective decoder'),
+    ).toBe(runId);
+
+    // EMIT SIDE. A host MUST hand back the form it wants clients to use; a link
+    // still spelling %2F re-creates the front-door defect for every follower.
+    if (typeof body.eventsUrl === 'string' && body.eventsUrl.includes(runId.split('/')[1]!)) {
+      expect(
+        body.eventsUrl.includes('%2F') || body.eventsUrl.includes('%2f'),
+        req(ID, DOC, `a link carrying a tenant-bound id MUST use the ~-escaped projection, not %2F — a front door may decode %2F and strand every client that follows this link (eventsUrl ${body.eventsUrl})`),
+      ).toBe(false);
+      expect(
+        body.eventsUrl.includes(projected),
+        req(ID, DOC, `eventsUrl MUST carry the runId in its projected form ${projected} (got ${body.eventsUrl})`),
+      ).toBe(true);
+    }
+
+    // APPLY-ONCE. The codec is deliberately NOT idempotent — escaping the
+    // marker is what makes it injective — so a host that projects twice breaks
+    // its OWN links. Both reporting hosts found they already compose payload
+    // projections on a single read path, so this is a live shape, not a
+    // thought experiment. Asserting 404 (not merely "not 200") keeps the leg
+    // from passing on a 500.
+    const twice = projectBoundId(projected);
+    const doubled = await http(() => driver.get(`/runs/${twice}`));
+    expect(
+      doubled?.status ?? null,
+      req(ID, DOC, `a DOUBLE-projected segment MUST NOT resolve — the codec is not idempotent, so projecting twice yields a different id and a host that does it strands its own links. Expected 404 for ${twice}, got ${doubled?.status ?? 'no response'}`),
+    ).toBe(404);
+
+    // DECODER RULE. A `~` not introducing two hex digits is malformed input, so
+    // 400 — not 404, which would say "no such run" about a request that never
+    // named one.
+    const malformed = await http(() => driver.get(`/runs/${runId.split('/')[0]}~2`));
+    expect(
+      malformed?.status ?? null,
+      req(ID, DOC, `a path segment whose '~' is not followed by two hex digits MUST be refused 400 validation_error — got ${malformed?.status ?? 'no response'} ${readErrorCode(malformed?.json) ?? ''}`.trim()),
+    ).toBe(400);
+  });
+});

package/src/scenarios/v2-created-run-readable.test.ts

@@ -38,11 +38,11 @@ import { v2Discovery } from '../lib/v2.js';
 import { readErrorCode } from '../lib/error-envelope.js';
 import { softSkip } from '../lib/soft-skip.js';
 import { req } from '../lib/requirement-ids.js';
+import { BOUND_ID as RUN_ID } from '../lib/bound-id.js';
 
 const ID = 'openwop.requirement.0172.created-run-readable';
 const DOC = 'spec/v2/core/runs.md §Create';
 const NOOP_WORKFLOW_ID = 'conformance-noop';
-const RUN_ID = /^[A-Za-z0-9._~-]{1,128}\/[A-Za-z0-9._~-]{16,128}$/;
 
 async function http(fn: () => Promise<OpenWOPResponse>): Promise<OpenWOPResponse | null> {
   try { return await fn(); } catch { return null; }

package/src/scenarios/v2-id-grammar.test.ts

@@ -27,10 +27,10 @@ import { v2Discovery, v2Validator } from '../lib/v2.js';
 import { readErrorCode } from '../lib/error-envelope.js';
 import { softSkip } from '../lib/soft-skip.js';
 import { req } from '../lib/requirement-ids.js';
+import { BOUND_ID as RUN_ID } from '../lib/bound-id.js';
 
 const DOC = 'spec/v2/core/identity.md §5';
 const NOOP_WORKFLOW_ID = 'conformance-noop';
-const RUN_ID = /^[A-Za-z0-9._~-]{1,128}\/[A-Za-z0-9._~-]{16,128}$/;
 const FOREIGN_RUN_ID = 'openwop-conformance-foreign-tenant/foreignopaque0123456789abcdef';
 
 async function discovery(): Promise<Record<string, unknown> | null> {

package/src/scenarios/v2-stream-sse-projection.test.ts

@@ -29,11 +29,11 @@ import { streamEvents } from '../lib/era2-seed.js';
 import { readErrorCode } from '../lib/error-envelope.js';
 import { softSkip } from '../lib/soft-skip.js';
 import { req } from '../lib/requirement-ids.js';
+import { BOUND_ID as RUN_ID } from '../lib/bound-id.js';
 
 const ID = 'openwop.requirement.0171.stream-sse-projection';
 const DOC = 'spec/v2/core/events.md §SSE frames';
 const NOOP_WORKFLOW_ID = 'conformance-noop';
-const RUN_ID = /^[A-Za-z0-9._~-]{1,128}\/[A-Za-z0-9._~-]{16,128}$/;
 
 async function http(fn: () => Promise<OpenWOPResponse>): Promise<OpenWOPResponse | null> {
   try { return await fn(); } catch { return null; }

package/src/setup.ts

@@ -25,6 +25,7 @@
  * `it` here; vitest treats setupFiles differently from scenario files.
  */
 
+import { driver } from './lib/driver.js';
 import { setAdvertisedFixtures, setDiscoveryUnreadable } from './lib/fixtures.js';
 import { setMultiAgentCapabilities } from './lib/multi-agent-capabilities.js';
 import { OtelCollector, setCollector } from './lib/otel-collector.js';
@@ -351,6 +352,43 @@ beforeEach(({ task }) => {
   _itSoftSkipMarks.set(file, softSkipMark()); // rc.56: this test's own softSkip window
   takeExplicitRequirementId(); // clear any override left by a test that threw before afterEach
 });
+/**
+ * Wipe the host's conformance-mock program store between scenario FILES.
+ *
+ * WHY THIS IS NECESSARY AND WAS MISSING. The host keeps the mock's programs in a
+ * module-level map keyed by `nodeId`, with a cursor. Several scenarios here
+ * deliberately seed programs that return `finishReason: 'length'` to exercise
+ * RFC 0033 truncation handling — `envelope-truncation-cap-exhaustion`,
+ * `envelope-truncated`, `envelope-retry-exhausted`,
+ * `envelope-completion-distinguishes-truncation`. A program that a scenario does
+ * not fully DRAIN stays pending for the life of the host process.
+ *
+ * This suite had no way to clear it. The host exposed a seam to SEED
+ * (`POST …/test/mock-ai/program`) and none to reset, while the host-side
+ * `resetMockPrograms` carried a header claiming it was "called between
+ * conformance scenarios" — it was called only by the host's own unit tests.
+ *
+ * The failure it produced accuses the wrong file: a later scenario dispatching
+ * on a colliding nodeId consumes the leftover truncation entries, its run fails
+ * `envelope_truncation_unrecoverable`, and nothing in that scenario explains it.
+ * MEASURED on `replay-observable-sequence-determinism`: red in-suite, green
+ * alone, five runs across unchanged bases, one of the reds at load1 3.4 on an
+ * idle box — so not contention, and a WRONG VALUE rather than a missing result.
+ *
+ * Per FILE rather than per test: programs are seeded for a whole scenario's
+ * sequence and a mid-file wipe would break the multi-attempt scenarios this
+ * store exists to serve. Best-effort — a host that predates the reset seam
+ * answers 404, and a conformance suite must not fail a compliant host for
+ * lacking a TEST seam.
+ */
+afterAll(async () => {
+  try {
+    await driver.post('/v1/host/openwop-app/test/mock-ai/reset', {});
+  } catch {
+    /* seam absent or host down — see above; never fail a scenario for this */
+  }
+});
+
 afterEach(({ task }) => {
   const file = _fileOf(task as { file?: { filepath?: string; name?: string } });
   if (file === null) return;