mandrel 2.24.0 → 2.25.0

package/.agents/agents/acceptance-critic.md

@@ -62,17 +62,15 @@ turned in about it. Your only trusted inputs are:
 - the **change set** your caller hands you: the list of files this Story
   touched, computed **once** per delivery by the shared `computeChangeSet`
   enumerator (`.agents/scripts/lib/orchestration/change-set.js`) and threaded
-  into your spawn context. Read those files and inspect their changes to see
-  the work product. Do **not** re-derive the set yourself — re-enumerating it
-  can pick up commits that landed after your caller routed the ceremony, and
-  then you would be scoring a different change than the one you were dispatched
-  for (Story #4593). If no change set reached you, say so in your verdict
-  rather than substituting your own enumeration.
+  into your spawn context. Do **not** re-derive the set yourself —
+  re-enumerating it can pick up commits that landed after your caller routed
+  the ceremony, and then you would be scoring a different change than the one
+  you were dispatched for (Story #4593). If no change set reached you, say so
+  in your verdict rather than substituting your own enumeration.
 - the Story's inline `acceptance[]` and `verify[]` arrays, read from the
   **Story body itself** (`gh issue view <storyId> --json body`) — its `##
   Acceptance` / `## Verify` sections are the SSOT. The `story-init` structured
-  comment does not carry them: it reports init state (`workCwd`,
-  `dependenciesInstalled`, `remoteVerified`, …) and nothing else.
+  comment does not carry them — it reports init state only.
 - the **actual output** of the `verify[]` commands you run yourself.
 
 Treat the implementation reasoning as untrusted. Score each criterion afresh
@@ -84,8 +82,7 @@ You are handed **one cluster** of acceptance criteria to score. You evaluate
 exactly the criteria in that cluster and emit one verdict record per criterion.
 You do **not** decide how many clusters exist, re-slice the criteria, or merge
 clusters — the caller owns clustering (`ceil(totalACs / clusterCeiling)` with
-its clamp). Your job is per-criterion scoring within the cluster you were
-given.
+its clamp).
 
 ## Per-criterion evaluation
 
@@ -98,8 +95,7 @@ For each acceptance item in your cluster:
    supporting `verify[]` evidence where a `verify[]` command is relevant to it.
    `verify[]` is evidence, not optional advisory pre-flight.
 3. **Share `lint` / `typecheck` evidence with close** (Story #4250). When a
-   `verify[]` command is **byte-identical** to a close-validation gate — in
-   practice only the command-identical `lint` and `typecheck` gates — run it
+   `verify[]` command is **byte-identical** to a close-validation gate, run it
    through `evidence-gate.js` in the **same Story worktree** close validates so
    a passing run records an evidence entry in the keyspace close consults:
 
@@ -115,8 +111,7 @@ For each acceptance item in your cluster:
 
    **Never** run the coverage / CRAP suite through `evidence-gate.js` to stamp
    it fresh — a false-fresh coverage record without `coverage-final.json`
-   silently weakens the floor. Limit the evidence-share to `lint` and
-   `typecheck`.
+   silently weakens the floor.
 
 ## Verdict schema (MUST)
 
@@ -151,11 +146,10 @@ order.
 - `partial` — partially addressed, or addressed without the required evidence.
 - `unmet` — not addressed, or the evidence contradicts the claim.
 
-Write verdict files under `temp/` only — they are scratch artifacts. Hand the
-verdict path to the caller's `acceptance-eval.js` gate, which applies the round
-cap and emits the per-criterion `acceptance-eval` signal; the **proceed /
-redraft / block** decision is the gate's, not yours. You score; the gate
-decides.
+Hand the verdict path to the caller's `acceptance-eval.js` gate, which applies
+the round cap and emits the per-criterion `acceptance-eval` signal; the
+**proceed / redraft / block** decision is the gate's, not yours. You score; the
+gate decides.
 
 ## Boundaries
 

package/.agents/agents/auditor.md

@@ -50,9 +50,9 @@ the step-by-step. This shared core binds every role:
 You are an **audit lens worker**: you run one read-only audit lens over a
 scoped surface, filter your own findings, and return a report path plus an
 Executive Summary. Follow the `audit-<lens>.md` workflow your caller hands you
-for the lens-specific dimensions, detection batteries, applicability gates,
-and report additions; this delta governs what holds across every lens. The
-shared long-form contract is
+for its dimensions, detection batteries, applicability gates and report
+additions; this delta governs every lens. The shared long-form
+contract is
 [`helpers/audit-lens-core.md`](../workflows/helpers/audit-lens-core.md) — this
 file is its standalone-agent form.
 
@@ -61,15 +61,14 @@ file is its standalone-agent form.
 - This is a **read-only** analysis. Do **not** modify application code, styles,
   configuration, dependencies, branches, or labels, and never open a PR.
 - The **only** write you perform is the report artifact at
-  `{{auditOutputDir}}/audit-<lens>-results.md`, plus — where and only where the
-  lens body explicitly declares it — a single measurement/baseline artifact it
-  names (e.g. performance's `perf-baseline.json`).
+  `{{auditOutputDir}}/audit-<lens>-results.md`, plus — only where the lens body
+  explicitly declares it — a single measurement/baseline artifact it names
+  (e.g. `perf-baseline.json`).
 - Running **non-mutating** measurements/scanners the lens calls for (profilers,
-  timers, `npm audit`, `actionlint`, read-only ORM status commands) is
-  permitted; running anything that installs, mutates git/labels, edits source,
-  or connects to a production database is forbidden. A lens that names a
-  stricter carve-out (data-model's no-database rule, quality's "do not run the
-  suite") tightens this for that lens.
+  timers, `npm audit`, read-only status commands) is permitted; running
+  anything that installs, mutates git/labels, edits source, or connects to a
+  production database is forbidden. A lens that names a stricter carve-out
+  tightens this for that lens.
 
 ## Scope
 
@@ -77,17 +76,13 @@ Your caller supplies the change-set file list (the lens's `{{changedFiles}}`
 fence). When it is a populated file list, restrict analysis to those files and
 their direct dependencies. When it is the literal `{{changedFiles}}` token,
 there is no scope filter — run the lens codebase-wide. A lens whose body
-declares a deviation (documentation's target-set intersection, navigability's
-whole-route-tree evaluation) follows its own Scope section instead. When a
-surface is absent or inapplicable, say so in the report and emit the lens's
-not-applicable / empty result rather than inventing findings.
+declares a deviation follows its own Scope section instead.
 
 ## Findings schema — the finding-block skeleton (MUST stay parseable)
 
 Write the report with an `## Executive Summary` and a `## Detailed Findings`
 section. Every finding under Detailed Findings uses this shared skeleton; the
-lens may **add** fields (WCAG criterion, CWE ID, `Baseline MUST`, `Evidence`,
-`Route / Door` + `Persona(s)`) and may relabel `Severity` ↔ `Impact` and
+lens may **add** fields and may relabel `Severity` ↔ `Impact` and
 `Dimension` ↔ `Category` ↔ `Type`, but never drops a shared field — the
 `audit-to-stories` parser depends on this shape:
 
@@ -95,7 +90,7 @@ lens may **add** fields (WCAG criterion, CWE ID, `Baseline MUST`, `Evidence`,
 ### `path/to/primary-file.ext` — [Short title of the issue]
 
 - **Dimension:** [the lens-specific dimension]
-- **Severity:** [Critical | High | Medium | Low]
+- **Severity:** [Critical | High | Medium | Low | Info]
 - **Location:** `path/to/primary-file.ext:line`
 - **Current State:** [the specific file/line and why it is problematic]
 - **Recommendation & Rationale:** [how to remediate and why it matters]
@@ -113,10 +108,12 @@ and a surviving **Critical** halts the delivery gate:
 - **Critical** — an active, exploitable, or data-losing defect that must be
   fixed before the change ships.
 - **High** — a serious correctness/security/maintainability risk to fix
-  promptly; does not by itself block the release.
+  promptly; does not block the release.
 - **Medium** — a real problem worth scheduling; contained blast radius or a
   workaround exists.
 - **Low** — minor or cosmetic; fix opportunistically.
+- **Info** — the floor: a grounded observation asking for no scheduled work
+  (accepts `Informational`). Never a home for findings that fail the bar below.
 
 ## Self-cross-check bar (mandatory before you write the report)
 

package/.agents/agents/story-worker.md

@@ -46,10 +46,10 @@ the step-by-step. This shared core binds every role:
 # story-worker — Story delivery boot context
 
 You are a **Story delivery worker**: you take one Story from init through
-implementation to a landed PR, then return. Follow the
-`helpers/deliver-story` workflow prose your caller hands you for the
-step-by-step; this delta states the non-negotiable MUSTs that hold across
-every step. Treat a blocking tool-permission prompt as a harness condition —
+implementation to a **pushed branch**, then return. You do **not** close it —
+your caller owns the close-and-land tail. Follow the `helpers/deliver-story`
+workflow prose your caller hands you; this delta states the non-negotiable
+MUSTs. Treat a blocking tool-permission prompt as a harness condition —
 transition to `agent::blocked` rather than waiting on an approval that
 cannot come.
 
@@ -59,10 +59,9 @@ cannot come.
    `node .agents/scripts/single-story-init.js --story <storyId>` from the
    **main checkout**, synchronously with the Bash maximum timeout — a
    per-worktree install can take minutes; do not background it.
-2. Capture `workCwd` and `dependenciesInstalled` from the flat init
-   envelope. When worktree isolation is on, work only inside the absolute
-   `workCwd`; the main checkout's HEAD is never moved by you. Because cwd
-   may reset between calls, anchor every subsequent path at `workCwd`.
+2. Capture `workCwd` and `dependenciesInstalled` from the init envelope.
+   Work only inside the absolute `workCwd`; never move the main checkout's
+   HEAD. Because cwd may reset between calls, anchor every path at `workCwd`.
 
 ## Verify branch before every commit (MUST)
 
@@ -80,32 +79,29 @@ state) to restore the branch first.
 
 Author Conventional Commit subjects directly on `story-<storyId>` per
 [`git-conventions.md`](../rules/git-conventions.md): imperative mood,
-≤100 chars, referencing the Story via `(refs #<storyId>)`. The `commit-msg`
-Husky hook runs commitlint — never bypass it with `--no-verify` /
-`--no-gpg-sign`. If a hook fails, fix the cause and add a follow-up commit;
-do not amend the rejected commit.
+≤100 chars, referencing the Story via `(refs #<storyId>)`. Never bypass the
+`commit-msg` hook with `--no-verify` / `--no-gpg-sign`. If a hook fails, fix
+the cause and add a follow-up commit; never amend the rejected one.
 
 ## Docs context — digest first
 
 Do **not** re-read every file in `project.docsContextFiles`. Read the
 `docsDigestPath` digest your caller passes, then pull full files on demand
-at the line numbers it names. A null `docsDigestPath` means no per-Story
-docs mandate — read a full doc only when the Story's own context points at
-one.
+at the line numbers it names. A null `docsDigestPath` means no docs
+mandate — read a full doc only when the Story's context points at one.
 
 ## Close gates — do not pre-run
 
 `single-story-close.js` runs the canonical close-validation chain
 (**typecheck, lint, test, format, maintainability, coverage, crap**) before
-it merges. Advisory pre-flight while iterating on a fix is fine, but the
-close pipeline is the authoritative gate. The acceptance self-eval loop may
-share `lint` / `typecheck` evidence with close via `evidence-gate.js`;
-never stamp coverage / CRAP fresh that way.
+it merges. Advisory pre-flight is fine; the close pipeline is the
+authoritative gate. The acceptance self-eval loop may share `lint` /
+`typecheck` evidence with close via `evidence-gate.js`; never stamp
+coverage / CRAP fresh that way.
 
 Before trusting a gate's output — or diagnosing a red one — read
 [`known-tooling-behavior.md`](../rules/known-tooling-behavior.md): measured
-cases where a command prints what it does not mean (lint exits 1 under a
-`0 error(s)` summary; a green `check-baselines.js` is not `baselines`).
+cases where a command prints what it does not mean.
 
 ## Acceptance self-eval before close (MUST)
 
@@ -115,15 +111,14 @@ run the bounded acceptance self-eval loop
 It scores the change set you computed **once** and injected into the critic
 — never one the critic re-derives (Story #4593) — against each
 `acceptance[]` item, consuming `verify[]` output as required evidence. Gate
-outcomes: **proceed** → flip to `closing` and close; **redraft** → fix the
-flagged criteria, commit, re-eval; **block** → take the blocked path below.
-Never silently proceed to close.
+outcomes: **proceed** → flip to `closing`, push, hand off; **redraft** → fix
+the flagged criteria, commit, re-eval; **block** → take the blocked path
+below. Never silently hand off an unscored branch.
 
 ## Lifecycle: progress & blocked (MUST)
 
 - **Progress.** Relay one terse line per phase transition (e.g.
-  `Story #<id>: implementing → closing`); your commits on `story-<id>` and
-  those lines are the progress surface.
+  `Story #<id>: implementing → closing`).
 - **Blocked.** When you genuinely cannot proceed, transition the Story to
   `agent::blocked`, post a `friction` comment naming the decision needed
   (or the unmet criteria and their evidence), and **exit non-zero**.
@@ -134,31 +129,23 @@ Never silently proceed to close.
 
 The Story's init envelope carries `remoteVerified` + `remoteProbe`. When
 `remoteVerified` is `false`, transition the Story to `agent::blocked`
-quoting `remoteProbe.detail` and stop. Implementing the Story inline
-outside the worktree / branch / PR path — or committing it to local `main`
-— is expressly **forbidden**; the close pipeline's push
-(`single-story-close.js`) is the only sanctioned landing.
-
-## Hold the turn until the envelope arrives (MUST)
-
-Run close in the **foreground** and wait for it. Never background it,
-never delegate it to a child, and never end your turn while it is still
-running — "close is running" is not a return value. Ending early strands
-the envelope in a turn nobody reads and costs your caller a recovery
-cycle plus a full resume of you. Close does persist a copy to
-`temp/orchestration/story-deliver-terminal-<id>.json`; that is your
-caller's fallback, not your licence to return before the verdict.
-
-## Return schema
-
-The return contract is
-[`story-deliver-terminal.schema.json`](../schemas/story-deliver-terminal.schema.json)
-— the SSOT for every field (Story #4543); do not restate them.
-`single-story-close.js` emits a validated envelope between its
-`--- STORY DELIVER TERMINAL ---` markers — **relay it**, never
-hand-compose one. Status ↔ exit code: `landed` → 0; `pending` → 3
-(**resumable, not a failure** — its `nextCommand` resumes it; the only
-sanctioned no-merge ending); `blocked` / `failed` → exit non-zero via the
-blocked path above. Stranded? Probe, don't guess:
-`node .agents/scripts/deliver-recover.js --story <id>` (read-only, prints
-one next command).
+quoting `remoteProbe.detail` and stop. A PR opened by
+`single-story-close.js` is the only sanctioned landing.
+
+## Your turn ends at a pushed branch (MUST)
+
+You do **not** run close. Push `story-<storyId>` to `origin` — confirming
+the remote ref moved — and return. The dispatching orchestrator runs
+`single-story-close.js` in its own session, serialized against your
+siblings. Do not open the PR, do not flip `agent::done`, and do not spawn
+a child to close on your behalf.
+
+## Return contract — the hand-off report
+
+Return a short, literal hand-off your caller can act on: the Story id,
+`workCwd`, the branch name, the pushed head SHA, the self-eval verdict,
+and the `verify[]` evidence you gathered.
+Say plainly that the branch is pushed and unclosed. Never hand-compose a
+terminal envelope — that document belongs to close, and inventing one
+makes an unlanded Story look landed. If the push itself fails, take the
+blocked path above rather than returning a hand-off you cannot back.

package/.agents/audit-checklists/architecture.md

@@ -6,7 +6,7 @@
 
 # Architecture & Clean Code Audit — authoring checklist
 
-> Audit architectural boundaries, module coupling, and layering violations; emit a structured findings report keyed to High/Medium/Low severity.
+> Audit architectural boundaries, module coupling, layering violations, and shipped-but-uncalled seams; emit a structured findings report keyed to the canonical severity scale.
 
 Self-check your change against this lens's concerns before you ship:
 
@@ -21,3 +21,7 @@ Self-check your change against this lens's concerns before you ship:
 - [ ] Medium
 - [ ] Low
 - [ ] Automated Architecture Guardrails
+- [ ] Shipped-But-Never-Wired Seams (mandatory).
+- [ ] A produced-but-never-consumed artifact.
+- [ ] An optional field nothing populates.
+- [ ] An exported seam with no in-tree caller

package/.agents/audit-checklists/quality.md

@@ -18,3 +18,6 @@ Self-check your change against this lens's concerns before you ship:
 - [ ] Test Data Management
 - [ ] Performance & Execution
 - [ ] Requirement Alignment
+- [ ] Unwired Seams — Coverage Without a Caller (mandatory).
+- [ ] A produced-but-never-consumed artifact
+- [ ] An optional field nothing populates

package/.agents/docs/agentrc-reference.json

@@ -100,7 +100,8 @@
       "watch": {
         "pollIntervalMs": 10000,
         "maxPolls": 180,
-        "maxResumes": 3
+        "maxResumes": 3,
+        "attachWindowMs": 1200000
       },
       "autoMerge": "trust-ci"
     },

package/.agents/docs/configuration.md

@@ -278,10 +278,11 @@ top-level keys are validation errors.
 | `acceptanceEval.maxRounds` | No | `integer` | — | Maximum number of redraft rounds before escalation. Default 2; clamped into [1, hard ceiling] by lib/config/acceptance-eval.js so the cap can never be disabled (maxRounds: 0 clamps up to 1). |
 | `acceptanceEval.clusterCeiling` | No | `integer` | — | Epic #4475 (M4-B). Max acceptance criteria one single-delivery acceptance critic scores in a single fresh-context pass. Single delivery clusters the Epic ## Acceptance Table ACs into ceil(totalACs / clusterCeiling) groups and spawns one maker-blind critic per cluster, restoring the distributed acceptance coverage the per-Story critic fan-out gave for free. Default 4; clamped into [1, 8] by lib/config/acceptance-eval.js so a large value cannot collapse the fan-out to a single diluted critic. Ignored on the fan-out route. |
 | `ci` | No | `object` | — | Nested configuration block. |
-| `ci.watch` | No | `object` | — | Story #4356 (Epic #4355). Poll-loop tuning for the merge/CI watch. pollIntervalMs is the cadence between check probes; maxPolls caps total probes before the watcher gives up; maxResumes caps how many times the watcher may resume after a transient stall. |
+| `ci.watch` | No | `object` | — | Story #4356 (Epic #4355). Poll-loop tuning for the merge/CI watch. pollIntervalMs is the cadence between check probes; maxPolls caps total probes before the watcher gives up; maxResumes caps how many times the watcher may resume after a transient stall; attachWindowMs bounds the wait for a required context to attach at all. |
 | `ci.watch.pollIntervalMs` | No | `integer` | — | — |
 | `ci.watch.maxPolls` | No | `integer` | — | — |
 | `ci.watch.maxResumes` | No | `integer` | — | — |
+| `ci.watch.attachWindowMs` | No | `integer` | — | Story #4890. How long (ms) the watch keeps re-resolving an EMPTY `gh pr checks --required` set before it stops waiting for a required context to attach. A ruleset attaches its contexts asynchronously and the arrival latency is set by the slowest one, so a required context that is an aggregator job gated on every other tier is the last to appear — measured at 16m52s on this repository. Default 1200000 (20 minutes). Raise it for a repository whose contexts arrive later still; exhausting the window is never reported as a red check (the watch exits 2, not-yet-started). |
 | `ci.autoMerge` | No | `"trust-ci"` \| `"strict"` | — | Story #4356 (Epic #4355). Merge posture. 'trust-ci' (default) merges once required checks pass; 'strict' additionally requires a clean review gate. |
 | `routing` | No | `object` | — | v2 delivery-spawn routing: role-scoped boot contexts and maker-checker sampling. The v1 singleDelivery epic-route kill-switch was removed in Stage 6. |
 | `routing.roleScopedAgents` | No | `boolean` | — | Epic #4478 (M7-B). Kill-switch for the role-scoped boot contexts. When true (default), a converted delivery spawn (`story-worker`, `acceptance-critic`) boots on its own `.claude/agents/<role>.md` system prompt instead of re-paying the full CLAUDE.md @-import closure (≈50KB → ≈8KB per spawn — the payoff of the context diet). When false, every converted spawn falls back to `subagent_type: general-purpose` — the instant, code-rollback-free per-consumer revert, and the universal escape for hosts that ignore `.claude/agents/`. The fallback is the full-closure agent that ran before M7-B, so flipping it off never drops a gate. |

package/.agents/docs/workflows.md

@@ -37,7 +37,7 @@ description, edit the workflow file’s front-matter and regenerate.
 | Command | Description |
 | --- | --- |
 | `/audit-accessibility` | Audit WCAG accessibility conformance (static-first) with an optional runtime verification pass, and produce a structured findings report |
-| `/audit-architecture` | Audit architectural boundaries, module coupling, and layering violations; emit a structured findings report keyed to High/Medium/Low severity. |
+| `/audit-architecture` | Audit architectural boundaries, module coupling, layering violations, and shipped-but-uncalled seams; emit a structured findings report keyed to the canonical severity scale. |
 | `/audit-clean-code` | Audit code smells, dead code, complexity hotspots, and maintainability-index outliers; emit a structured findings report. |
 | `/audit-data-model` | Audit the persistence layer as a first-class artifact — model↔migration↔seed drift, constraint completeness, migration hygiene, type fidelity, and access-pattern fit; gated by a persistence-layer applicability probe so DB-less repos skip cleanly. |
 | `/audit-dependencies` | Audit `package.json` for unused, outdated, and major-version-stale dependencies; surface Node-engine drift and propose upgrade batches. |

package/.agents/rules/ci-remediation.md

@@ -19,8 +19,10 @@ exactly one of two ways, and no others:
    caused by the diff under review — **verify the same check against an
    unmodified `main` checkout**; if it also fails on `main` the defect is
    pre-existing and belongs in a separate change — then fix it at source,
-   commit on `story-<storyId>`, push, and re-run the watcher. Auto-merge stays
-   armed across retries. Route deterministic per-check failures (lint/format,
+   commit on `story-<storyId>`, push, and re-run the watcher. Auto-merge is
+   **disarmed on the first red** and re-armed only by a green on a **new head
+   SHA**, so the fix must be a new commit. Route deterministic per-check
+   failures (lint/format,
    maintainability/CRAP baseline drift, test failure, coverage threshold)
    through the fix table in
    [`deliver-story-reference.md` § Step 4](../workflows/helpers/deliver-story-reference.md#step-4--ci-watch--fix-recovery);
@@ -39,6 +41,51 @@ through the same two options; bisect environment (runner OS, Node version,
 concurrency, a platform-conditional branch, an external service) vs. code (an
 order-dependent test, a race, a shared-state assumption) to decide which.
 
+## Verdicts
+
+Every red check reaches exactly one verdict, and each verdict routes to one of
+the two options above. Name the verdict you reached in the `friction` comment.
+
+| Verdict | Evidence | Routes to |
+| --- | --- | --- |
+| **defect-in-diff** | The failure reproduces on the branch and not on an unmodified `main` | Option 1 — fix at source |
+| **pre-existing** | The same check fails on an unmodified `main` too | Option 2 — file `meta::framework-gap`; remediate here only if it blocks this delivery |
+| **capacity** | Proven exhaustion of a runner resource, not a property of the diff (see below) | Option 2 — file `meta::framework-gap` **and** escalate to the operator |
+
+### The `capacity` verdict
+
+A job can fail because the runner ran out of something, not because the code is
+wrong: no runner could be provisioned, the disk or memory ceiling was hit, a
+process/PTY/file-descriptor limit was exhausted, the job wall-clock timed out
+with no progress, or a self-hosted pool was saturated. Nothing on the branch
+causes it and nothing on the branch can fix it.
+
+This verdict exists because the rule previously offered no landing for that
+case. The honest reading of "a red check is a defect until proven otherwise" is
+that capacity failures are the *otherwise* — but with no verdict for them the
+only shapes on offer were "fix the diff" (impossible) and "it's flaky, re-run
+it" (forbidden), so the rule got broken rather than followed. Naming the verdict
+removes the incentive to launder a capacity failure as a rerun.
+
+**Capacity must be proven, not inferred.** A green on re-run is the single
+weakest form of evidence for it and never establishes it — that is precisely the
+observation a flaky test produces. Cite the resource and the reading: the log
+line naming the exhausted limit (an OOM kill, `ENOSPC`, `EMFILE`,
+`forkpty/sudo: Device not configured`, a provisioning error, a no-output
+timeout), plus the fact that the failure is not specific to this diff. Absent
+that reading the verdict is **flaky, not capacity**, and it routes to Option 1.
+
+On a `capacity` verdict: file the `meta::framework-gap` issue with the run link,
+the failure signature, and the resource reading; flip the Story to
+`agent::blocked` with a `friction` comment naming the verdict; and hand back to
+the operator, who owns the runner pool. Do not sit in a retry loop waiting for
+capacity to return.
+
+**Rerunning a failed job to reach green stays forbidden under every verdict,
+`capacity` included.** The verdict changes who owns the fix and where it is
+filed; it never licenses a re-run, and it is not a route to a green bar. A
+capacity-blocked delivery ends `agent::blocked` — not merged.
+
 ## Verifier
 
 The check is resolved only when it is **green with zero reruns of the failed
@@ -48,6 +95,23 @@ re-run a failed job to "see if it goes green," and you may **not** skip,
 `.only`, or quarantine a flaky test to get a green bar. Both mask the defect
 and are prohibited by this rule.
 
+**The enforcement point is
+[`pr-watch-with-update.js`](../scripts/pr-watch-with-update.js)**, and it acts
+on the **first red** — GitHub's native auto-merge fires server-side, so it
+races any attempt to detect a rerun-green and block it after the fact. On the
+first red the watcher **disarms native auto-merge** (a disarm failure is a
+blocker, not a warning) and records the PR **head SHA** in the digest
+alongside the failing check-run identity. On green it adjudicates:
+
+- **Same head SHA** → the green came from re-running the failed job. The
+  watcher exits non-zero, flips the Story to `agent::blocked` with a
+  `friction` comment, and requires the `meta::framework-gap` issue (run link +
+  failure signature, both already in the digest) before the delivery proceeds.
+- **New head SHA** → fix at source. The digest is retired, auto-merge is
+  re-armed, and the delivery continues unobstructed.
+
+A delivery that never went red has no digest and is untouched.
+
 ## Escalation
 
 Flip the ticket to `agent::blocked`, post a `friction` comment (naming the
@@ -65,4 +129,5 @@ operator under **any** of:
   registry/network outage, a branch-protection or CI misconfiguration, an
   expired credential) — file the `meta::framework-gap` issue (with run link +
   signature) and escalate on the first encounter rather than burning iterations
-  trying to code around it.
+  trying to code around it. A proven-capacity failure is this case: reach the
+  `capacity` verdict above and escalate on the first encounter.

package/.agents/schemas/agentrc.schema.json

@@ -1367,7 +1367,7 @@
           "properties": {
             "watch": {
               "type": "object",
-              "description": "Story #4356 (Epic #4355). Poll-loop tuning for the merge/CI watch. pollIntervalMs is the cadence between check probes; maxPolls caps total probes before the watcher gives up; maxResumes caps how many times the watcher may resume after a transient stall.",
+              "description": "Story #4356 (Epic #4355). Poll-loop tuning for the merge/CI watch. pollIntervalMs is the cadence between check probes; maxPolls caps total probes before the watcher gives up; maxResumes caps how many times the watcher may resume after a transient stall; attachWindowMs bounds the wait for a required context to attach at all.",
               "properties": {
                 "pollIntervalMs": {
                   "type": "integer",
@@ -1380,6 +1380,11 @@
                 "maxResumes": {
                   "type": "integer",
                   "minimum": 0
+                },
+                "attachWindowMs": {
+                  "type": "integer",
+                  "minimum": 1,
+                  "description": "Story #4890. How long (ms) the watch keeps re-resolving an EMPTY `gh pr checks --required` set before it stops waiting for a required context to attach. A ruleset attaches its contexts asynchronously and the arrival latency is set by the slowest one, so a required context that is an aggregator job gated on every other tier is the last to appear — measured at 16m52s on this repository. Default 1200000 (20 minutes). Raise it for a repository whose contexts arrive later still; exhausting the window is never reported as a red check (the watch exits 2, not-yet-started)."
                 }
               },
               "additionalProperties": false

package/.agents/schemas/baselines/baseline-envelope.schema.json

@@ -25,6 +25,10 @@
       "type": "string",
       "description": "Optional per-kind stamp identifying the SCORING SEMANTICS that produced these rows — the dimension kernelVersion cannot express, because a kind's own scoring can change while the upstream package it stamps does not. A baseline whose stamp differs from the running scorer's is incomparable and the gate fails closed with re-baseline guidance rather than comparing across the boundary."
     },
+    "tsTranspilerVersion": {
+      "type": "string",
+      "description": "Optional per-kind stamp identifying the TypeScript transpiler that produced the sourcemap these rows' line coordinates were resolved through (Story #4866). Only meaningful for kinds whose rows carry a line coordinate derived from transpiled sources; a transpiler change can move every such coordinate, and for kinds that key rows on a line that coordinate is half the row identity."
+    },
     "rollup": {
       "type": "object",
       "description": "Aggregate metrics keyed by component. The '*' key is reserved for the whole-repo rollup and is REQUIRED; additional keys MAY be present, one per declared component.",

package/.agents/schemas/baselines/crap.schema.json

@@ -48,7 +48,11 @@
           "path": { "type": "string" },
           "method": { "type": "string" },
           "startLine": { "type": "integer", "minimum": 1 },
-          "crap": { "type": "number", "minimum": 0 }
+          "crap": { "type": "number", "minimum": 0 },
+          "coordinateSystem": {
+            "enum": ["transpiled"],
+            "description": "Coordinate provenance of startLine (Story #4866). Written ONLY when the row kept transpiled coordinates because its sourcemap lookup did not resolve — the absent-means-'original' default keeps a pure-JavaScript baseline byte-identical, since JavaScript coordinates already ARE original-source coordinates. A row carrying this key is not comparable to one without it: the compare refuses to resolve the two through its nearest-line drift heuristic rather than manufacturing a regression from the gap between two coordinate systems."
+          }
         }
       }
     }

package/.agents/schemas/crap-report.schema.json

@@ -23,6 +23,9 @@
         "regressions",
         "newViolations",
         "drifted",
+        "incomparable",
+        "provenanceMismatched",
+        "unscorable",
         "removed",
         "skippedNoCoverage",
         "scope",
@@ -34,6 +37,21 @@
         "regressions": { "type": "integer", "minimum": 0 },
         "newViolations": { "type": "integer", "minimum": 0 },
         "drifted": { "type": "integer", "minimum": 0 },
+        "incomparable": {
+          "type": "integer",
+          "minimum": 0,
+          "description": "Story #4866: scanned methods the compare REFUSED to resolve because their line coordinate provenance differs from the matching baseline rows'. Reported, never scored — resolving them through the nearest-line drift heuristic is what manufactures a regression no change can satisfy."
+        },
+        "provenanceMismatched": {
+          "type": "integer",
+          "minimum": 0,
+          "description": "Story #4871: scanned methods whose baseline candidates included at least one row expressed in a DIFFERENT line coordinate system, counted BEFORE the compare's provenance filter discards that evidence. This — not the drifted-row count, which by construction only ever counts rows whose coordinates agreed — is the quantity the unsound-basis backstop compares against its threshold."
+        },
+        "unscorable": {
+          "type": "integer",
+          "minimum": 0,
+          "description": "Story #4871: methods the scan could not score because no coverage artifact covered them. Reported rather than computed from an assumed zero percent coverage (which would drive CRAP maximal on untouched files), and excluded from the denominator of the unsound-basis ratio."
+        },
         "removed": { "type": "integer", "minimum": 0 },
         "skippedNoCoverage": { "type": "integer", "minimum": 0 },
         "scope": {
@@ -97,6 +115,25 @@
           }
         }
       }
+    },
+    "diagnostics": {
+      "type": "array",
+      "description": "Story #4866: named, gate-level findings emitted INSTEAD of per-method verdicts when the gate has established that no verdict it could produce would be meaningful — an incomparable baseline, or a comparison basis whose drifted-row ratio proves the two sides disagree on line coordinates. Present only in that case; `violations` is then empty and the preview gate exits 0.",
+      "items": {
+        "type": "object",
+        "required": ["name", "message"],
+        "additionalProperties": false,
+        "properties": {
+          "name": {
+            "type": "string",
+            "enum": [
+              "crap-unsound-comparison-basis",
+              "crap-baseline-incompatible"
+            ]
+          },
+          "message": { "type": "string" }
+        }
+      }
     }
   }
 }