mandrel 2.55.0 → 2.57.0

package/.agents/workflows/helpers/deliver-story-reference.md

@@ -140,10 +140,8 @@ full-ceremony Story. The lite route's `preserves` field is the machine-readable
 record of those non-negotiables; there is no lite-specific gate bypass.
 
 **Ceremony comes from the landed diff; the dispatch mode comes from the
-run.** Persist stamps a lite cohort's Stories with the `route::lite` label as a
-_human-visible hint only_ (and ledgers the authored verdict — recorded reason
-plus per-Story shape evidence — on the `story-plan-state` checkpoint); the
-label is never the control signal. Ceremony is resolved from the **derived
+run.** Persist stamps no route label (Story #5312 retired the plan-side lite
+claim with its `route::lite` hint). Ceremony is resolved from the **derived
 change level** (`deriveChangeLevel` over the computed change set — digest § 3),
 not from a body-shape read: a footprint intersecting a sensitive-path class
 derives `high`, so the Story keeps its fresh acceptance critic. The light path
@@ -240,35 +238,19 @@ discovers them only after the whole close pipeline has run, at several times
 the cost of one full-suite run in the worktree.
 
 **Run it once, last, so close can credit it.** The run belongs **after** the
-self-eval loop's last fix commit and **after** the hand-off push — the credit
-is keyed on the tree, not on push state, so pushing first keeps the stamp and
-buys the ordering Step 2.5 needs (the capture is backgrounded, and its
-completion ends the turn); redraft rounds run scoped tests. Close skips a gate that already passed at the current HEAD, but a bare
-`npm test` deposits no such record — the suite then runs twice per delivery,
-once here and once in the close gate chain. Pick the invocation by the same
-predicate `close-validation/gates.js` uses to choose its test gate:
-
-```bash
-# CRAP gate enabled (default) + a `test:coverage` script — writes the stamp
-# the close `coverage-capture` gate reads:
-node <main-repo>/.agents/scripts/coverage-capture.js --cwd <workCwd>
-# otherwise — the evidence record the close `test` gate reads. <workCwd> must
-# be ABSOLUTE and the runner exactly `npm test`: both sides hash
-# {cmd, args, cwd}, so a relative path or a wrapper misses the credit.
-node <main-repo>/.agents/scripts/evidence-gate.js --standalone \
-  --scope-id <storyId> --gate test --worktree <workCwd> -- npm test
-```
-
-The credit expires the moment it stops describing the tree: evidence is keyed
-on HEAD, the capture stamp on a content digest of `crap.targetDirs`. A
-self-eval fix — or any commit — invalidates it and close re-runs the suite for
-real, so this never trades away the gate. That keying is exactly why the run
-comes last, and why the push before it is free. Close's own base-sync can
-spend the stamp too when it lands base commits; it now says so out loud rather
-than silently re-running the suite.
-
-**`verify[]` reuses the same stamp.** A `verify[]` entry that is itself a
-full-suite command is reported **credited** against that stamp rather than
+self-eval loop's last fix commit; redraft rounds run scoped tests. A green
+`npm test` in the worktree on `story-<id>` deposits the `test` evidence
+record close reads (Story #5313 — `lib/test-run-credit.js`), keyed on HEAD
+and the tree fingerprint and hashed on the exact command close spawns, so
+close reports the gate as credited at unchanged HEAD instead of re-running
+the suite. The credit expires the moment it stops describing the tree: any
+later commit invalidates it and close re-runs the suite for real, so this
+never trades away the gate. The CRAP gate still runs `coverage-capture.js`
+itself when it needs a fresh artifact — the capture stamp is a claim about
+`coverage/coverage-final.json`, which a bare `npm test` does not produce.
+
+**`verify[]` reuses the same credit.** A `verify[]` entry that is itself a
+full-suite command is reported **credited** against that record rather than
 respawned (`resolveVerifyCredit` in
 [`verify-credit.js`](../../scripts/lib/orchestration/verify-credit.js)), and the
 self-eval gate warns when it sees one: the intended shape is scoped `verify[]`
@@ -294,24 +276,23 @@ round cap, proceed / redraft / block — not an independent additional pass
 over the criteria. The M4-B floor holds: one verdict per cluster, with the
 cluster count owned by the dispatching caller and never by routing.
 
-**One round = N cluster critics → ONE merged verdict → ONE gate call.** The
+**One round = N cluster critics → ONE merged verdict → ONE gate call** (fresh
+critics; an inline-owned verdict is one file scored in one call). The
 clusters are how a round is _authored_; they are not how it is _scored_.
 Concatenate every cluster's records into a single `criteria[]` ordered by
 `index` — exactly one per `acceptance[]` item, under one `storyId`,
 `schemaVersion`, `round` and `commitSha` — and hand that merged file to the
-gate once, with `--expected-criteria` set to the Story's `acceptance[]` count:
+gate once:
 
 ```bash
 node <main-repo>/.agents/scripts/acceptance-eval.js \
-  --story <storyId> --verdict <merged-verdict-path> \
-  --expected-criteria <acceptance[] count>
+  --story <storyId> --verdict <merged-verdict-path>
 ```
 
-The flag is what makes the merge enforceable: `assertCriteriaCoverage` returns
-early on the `null` default, so **omitting it leaves the guard inert** and a
-single cluster's verdict handed over unmerged scores a fraction of the criteria
-and still reports `proceed`. A length mismatch is rejected before scoring and
-consumes no round. Calling the gate once per cluster instead spends a round
+The gate reads the Story's `acceptance[]` count itself (Story #5313), so a
+single cluster's verdict handed over unmerged is rejected before scoring and
+consumes no round; `--expected-criteria` is accepted but redundant. Calling
+the gate once per cluster instead spends a round
 _per cluster_ — a Story past the cluster ceiling would burn its whole redraft
 budget on cluster arithmetic — and N concurrent calls race the Story-scoped
 round ledger. Full per-round mechanics, including the parallel dispatch and the
@@ -342,32 +323,30 @@ node .agents/scripts/update-ticket-state.js --ticket <storyId> --state agent::bl
 
 ## Step 2 — Ceremony detail
 
-**Compute the change set once** with the shared enumerator —
-the same module close uses — and reuse that one list downstream:
+**Compute the change set once** with the shared enumerator — the same module
+close uses — and reuse that one list downstream. `ceremony-derive.js`
+(Story #5313) is that enumeration, the level derivation and the ceremony
+resolution in one call:
 
 ```bash
-node --input-type=module -e '
-  import { computeChangeSet } from "<main-repo>/.agents/scripts/lib/orchestration/change-set.js";
-  const { files } = computeChangeSet({ baseRef: "main", headRef: "story-<storyId>" });
-  console.log(JSON.stringify(files));
-'
+node <main-repo>/.agents/scripts/ceremony-derive.js --story <storyId> --cwd <workCwd>
 ```
 
-Derive the level with
+Its `level` comes from
 [`deriveChangeLevel`](../../scripts/lib/orchestration/review-depth.js) over
 the one computed change-set list: a diff touching a sensitive path registered
 in `.agents/schemas/audit-rules.json` derives `high`, one touching none
-derives `low`, and an unenumerable diff (`files === null`) derives `null`.
-Hand the **same** list to every acceptance critic you spawn (Step 1a) — a
-critic that re-ran its own `git diff` could score against a different set
-than the one that routed it.
+derives `low`, and an unenumerable diff (`files: null`) derives `null`.
+Hand the **same** `files` list to every acceptance critic you spawn (Step 1a)
+— a critic that re-ran its own `git diff` could score against a different
+set than the one that routed it.
 
-Resolve fresh-vs-inline acceptance critics per AC-cluster with
+Its `mode` / `verdictOwner` come from
 [`resolveCeremonyForRisk`](../../scripts/lib/orchestration/ceremony-routing.js)
 (`minimal` → always inline; `strict` → always fresh; `standard` →
-`high`/`null` → `fresh`, `low` → `inline` unless the `freshCriticSampleRate`
-floor forces `fresh`). Review depth reads the same derived level via
-`review-depth.js` inside close, so the two decisions cannot disagree.
+`high`/`null` → `fresh`, `low` → `inline`). Review depth reads the same
+derived level via `review-depth.js` inside close, so the two decisions cannot
+disagree.
 
 **Inline-dispatch override.** When the Story dispatches
 `inline` (`resolveStoryDispatchMode` → `inline`, which is exactly a
@@ -682,13 +661,28 @@ When the watch exits, branch on the exit code:
 
 **Triage authority.** How to classify and remediate a red (or repeatedly slow)
 check — the root-cause-only decision tree for infra/transient and flaky failures
-(reproduce → check `main` → bisect env vs code → fix in-scope or file a
-`meta::framework-gap` issue), the never-rerun / never-quarantine prohibitions,
-and the escalation criteria (three-strikes, the 30-minute wall-clock timebox,
-and the clearly-environmental fast path) — is defined once in
+(reproduce → check `main` → bisect env vs code → fix in-scope, or reach an
+Option-2 verdict and file the intake issue), the never-rerun / never-quarantine
+prohibitions, and the escalation criteria (three-strikes, the 30-minute
+wall-clock timebox, and the clearly-environmental fast path) — is defined once in
 [`.agents/rules/ci-remediation.md`](../../rules/ci-remediation.md). Read it
 before remediating a red check.
 
+**Filing an out-of-scope root cause is one command, never a hand-run `gh issue
+create`:**
+
+```bash
+node <agentRoot>/scripts/file-ci-gap.js --story <storyId> \
+  --verdict <pre-existing|capacity|unreproducible-tier> \
+  --owner <consumer|framework|platform> --evidence "<proof reading>" [--block]
+```
+
+It reads the digest, routes the filing to the repo that owns the fault, updates
+the existing ticket when the signature is a repeat, posts the `friction`
+comment, and with `--block` flips the Story. What it files is an **intake**
+issue, not a Story — `/mandrel-plan <issue number>` graduates it on the next
+planning pass, so the delivery never waits on planning.
+
 ### The auto-merge wait is an internally-blocking step
 
 This is the single most important contract of this workflow, and the seam

package/.agents/workflows/helpers/deliver-story.md

@@ -74,7 +74,7 @@ One branch, one PR to `main`, commits against the inline `acceptance[]` /
    `## Slicing` rows as **intra-session checkpoints** (reference § Step 1).
 2. Implement and commit on the Story branch, iterating with quick advisory
    gates (`typecheck`, `lint`, scoped tests) — the full chain runs in Step 3,
-   and the **one** creditable full-suite run at Step 2.5.
+   and the **one** full-suite run at Step 2.5.
 
 ### Step 1a — Bounded acceptance self-eval loop (**required**)
 
@@ -93,21 +93,17 @@ are reference § Step 2. Hard gates always run in Step 3 — the derived level
 never disables them; do **not** pre-run the chain here — Step 2.5's credited
 suite run is the sole exception.
 
-### Step 2.5 — Push, then the creditable full-suite run, then hand off
+### Step 2.5 — The one full-suite run, the push, then hand off
 
-**Push first.** After the self-eval loop's last fix commit, push
-`story-<storyId>` to `origin`, confirming the remote ref moved: the capture
-below is backgrounded, so *its* completion ends the turn.
+After the self-eval loop's last fix commit, run `npm test` **once** in the
+worktree (**digest § 5**): a green full run deposits the `test` credit close
+reads, keyed on the tree, so only a *later* commit invalidates it. Red →
+fix, commit, re-run.
 
-Then run the full suite **once**, after the push, in the shape close credits
-(**digest § 5**): the credit is keyed on the tree, not on push state, so only
-a *later* commit invalidates it; a bare `npm test` deposits none. Red →
-fix, commit, push, re-capture.
-
-Then (sub-agent dispatch only) return the hand-off — Story id, `workCwd`,
+Push `story-<storyId>` to `origin`, confirming the remote ref moved. Then
+(sub-agent dispatch only) return the hand-off — Story id, `workCwd`,
 branch, pushed head SHA, self-eval verdict, `verify[]` evidence — and stop.
-Do not open the PR or compose a terminal envelope. An inline run captures
-before Step 3.
+Do not open the PR or compose a terminal envelope.
 
 ## Step 3 — Close and land (`single-story-close.js`)
 

package/.agents/workflows/helpers/plan-reference.md

@@ -1,10 +1,10 @@
 # /mandrel-plan — on-demand reference appendix
 
 > **Applies when:** you are executing [`/mandrel-plan`](../mandrel-plan.md) and hit one of the
-> situations below — input-mode derivation, the Gate #1 light handoff,
-> shape-derived complexity routing, tickets-mode supersede authoring, critic
-> dispatch detail, a failed persist, or source-id resolution. The spine stays
-> resident; this file is read on demand.
+> situations below — input-mode derivation, the Gate #1 advisory line,
+> tickets-mode supersede authoring, the operator-invoked pre-mortem, a
+> failed persist, or source-id resolution. The spine stays resident; this
+> file is read on demand.
 
 ## Deriving the input mode
 
@@ -92,117 +92,58 @@ The marker keeps the operator's undelegated decisions findable after the
 fact: reviewing a `--yes` plan means scanning its decisions-made-by-default,
 not re-deriving which assumptions were really the agent's to make.
 
-## Gate #1 → the memory-pool advisory (`memoryPoolAdvisory`)
-
-On a truthy `memoryPoolAdvisory.recommend`, name
-[`/memory-consolidate`](../memory-consolidate.md) at Gate #1, quoting its
-`reasons[]`. Purely advisory: a stale pool degrades recall, it does not make
-the plan wrong, so it never blocks and never reroutes.
-
-## Gate #1 → the light path (in-session handoff)
-
-On a confirmed `deliverLightSuggestion`, `/mandrel-plan` routes into
-[`deliver-light.md`](deliver-light.md) **without ending the session**. Two
-things make that safe, and both are worth understanding before changing it:
-
-1. **The handoff carries the envelope, not the seed.** Gate #1 already holds
-   the interrogated `complexitySignals`; fill the light gate's `--creates`
-   / `--refactors` / `--acceptance` / `--reason` from those. Re-deriving from
-   raw seed text throws away the better signal and can disagree with the
-   suggestion that routed you.
-2. **The gate still runs.** The suggestion is read against seed-time ceilings
-   (`DELIVER_LIGHT_SUGGESTION_CEILINGS` — artifacts, risk hits, sensitive-path
-   classes); the light gate is read against the predicted work's effort and risk
-   (`STORY_SHAPE_CEILINGS` — change kinds, magnitude, uncertainty, deployable
-   span). Two different checks on purpose, so a confirm is not a bypass.
-
-**When the light gate answers `ask-operator`**, the two ceiling sets disagreed.
-Resume `/mandrel-plan` at step 2 (Author) **in this same session** — the interrogation
-is still valid and re-paying for it buys nothing. This bounce-back is not an
-escalation.
-
-**Under `--yes` the offer is recorded and planning proceeds** — it is *never*
-auto-downgraded to light. An unattended run has nobody to confirm the reroute,
-and a suggestion is not a confirmation. The same rule governs unknown triage
-unattended: AFK unknowns are still researched, but no free-form operator
-question is asked — each HITL unknown lands in Key Assumptions marked a
-decision-made-by-default, so the record shows what was decided for the operator
-rather than pretending it was decided with them.
-
-Escalation in the *other* direction — an over-scope prompt on the light path —
-is terminal and requires a fresh session. The rule that separates the two, and
-why it must not be flattened into symmetry:
-[`deliver-light.md` § Why the two directions differ](deliver-light.md).
-
-## Gate #1 → the `/prototype` offer (`uiSurface`)
-
-`complexitySignals.uiSurface` is the second advisory Gate #1 offer, and the
-weaker of the two on purpose: it carries **no routing authority and adds no
-gate**. Both halves are derived from observables already in the checkout — the
-`hasWebSurface` applicability predicate the `target: "web"` audit lenses gate
-on, and whether any predicted path matches a web lens `filePattern` registered
-in `audit-rules.json`. There is no configuration key to set: a project with no
-rendered frontend resolves falsey and the offer never fires.
-
-When it does fire, **name [`/prototype`](../prototype.md) and stop there.**
-`/mandrel-plan` must never invoke it — operator invocation is the entire design, because
-the value is a human looking at a layout before its UI acceptance criteria are
-frozen.
-
-**Under `--yes` the offer is recorded and planning proceeds** — no reroute, no
-prototype written, no gate raised. This is exactly how `deliverLightSuggestion`
-behaves unattended, and for the same reason: an unattended run has nobody to
-review an artifact, so recording the offer is the whole of the right behaviour.
-
-## Shape-derived complexity routing (`complexitySignals`)
-
-Complexity routes on the **objective shape of the authored work**, never on
-seed word count — a detailed prompt can describe trivial work, a terse one
-complex work. The pipeline stages the
-decision:
-
-- **Signals, not routing.** The envelope's `complexitySignals` field is
-  advisory only (`routingAuthority: false`): enumerated-artifact count (with
-  the configured `maxArtifacts` threshold beside it as one input),
-  `planning.riskHeuristics` phrases present in the seed, the repo state of
-  predicted paths (existing paths predict refactors; missing predict
-  creates), and the `audit-rules.json` sensitive-path classes the predicted
-  footprint intersects.
-- **You author the verdict.** Judge the signals: a genuinely trivial scope
-  (small additive footprint, no risk hits, no sensitive class) earns a `lite`
-  claim via `plan-persist.js --route-downgrade-reason "<why>"`. The reason is
-  recorded on every created Story's `story-plan-state` checkpoint, making the
-  judgment auditable; without a recorded reason the conservative default
-  (`full`) stands.
-- **Persist backstops the claim deterministically.** After authoring, the
-  work has measurable shape, so persist validates the `lite` claim against
-  each Story's own shape — distinct change kinds, declared magnitude,
-  uncertainty, deployable/migration span, glob-free footprint, and
-  sensitive-path classes, against the framework `STORY_SHAPE_CEILINGS` (effort
-  and risk, never artifact counts) — and **fails closed to
-  `full`** when any Story exceeds them (the refusal is ledgered on the
-  checkpoint too). The lite route is **not** licence to drop a
-  non-negotiable — every decision's `preserves` field enumerates what still
-  holds: the Story ticket, the PR-to-`main` landing, every repo quality gate,
-  and the security baseline. Those gates run in `single-story-close.js`
-  regardless of route.
-
-**The label is a hint; deliver re-derives.** Persist labels a
-lite cohort's Stories with **`route::lite`** as a *human-visible hint only* —
-`/mandrel-deliver` computes the route from each fetched Story body via the same shape
-function at dispatch, so neither a lost label nor an unread marker can
-misroute delivery: a lite-shaped Story derives `lite` even with the label
-absent, and a sensitive-footprint Story routes `full` and keeps its fresh
-critic even with the label present. The derived route sets ceremony, not
-where the engine runs — sub-agent boots are collapsed by a **single-Story
-run**, never by a trivial shape. The `route::*` axis stays runtime-derived: hand-authored
-`route::*` entries in `labels[]` are dropped by persist.
-
-The knobs (`planning.complexityGate.{enabled, maxArtifacts}`) are documented
-in [`.agents/docs/configuration.md`](../../docs/configuration.md) under
-`### planning`; the defaults live on `DEFAULT_COMPLEXITY_GATE` and the shape
-ceilings on `STORY_SHAPE_CEILINGS` in
-[`lib/orchestration/complexity-gate.js`](../../scripts/lib/orchestration/complexity-gate.js).
+## Gate #1 → the one advisory line
+
+Gate #1 stops for exactly two things — the sharpened plan intent and any HITL
+unknown — and everything else the envelope surfaced collapses to **one
+advisory line** beneath it (Story #5312). Nothing on that line stops the run,
+reroutes it, or is invoked by `/mandrel-plan`; each item names something the
+operator may prefer to do instead, and the run proceeds either way. Under
+`--yes` the line is recorded and planning continues — an unattended run has
+nobody to take an offer.
+
+The line names, in order, whichever of these the envelope carries:
+
+- **`duplicates[]`** — open Stories the seed resembles (never Epics). Name
+  the top one or two by id and title; a plan that duplicates open work is
+  still the operator's call.
+- **Open `intake` rows** (`priorFeedback`) — CI-gap intake filings written by
+  [`file-ci-gap.js`](../../scripts/file-ci-gap.js) when a delivery reached an
+  Option-2 verdict in [`ci-remediation.md`](../../rules/ci-remediation.md).
+  They carry evidence but no `## Spec`, no `acceptance[]` / `verify[]` and no
+  `agent::*` label, so `/mandrel-deliver` cannot take one: graduating it is
+  exactly **tickets mode** (`/mandrel-plan <issue number>`), and a filing
+  that keeps recurring (its `## Occurrences` table is the count) is often the
+  better next Story than the seed in front of you. A `platformGaps[]` row is
+  the same shape with a different owner.
+- **`memoryPoolAdvisory.recommend`** — name
+  [`/memory-consolidate`](../memory-consolidate.md), quoting its
+  `reasons[]`. The one arm left measures the `MEMORY.md` index against the
+  harness's byte cap; a stale pool degrades recall, it does not make the plan
+  wrong.
+- **`complexitySignals.uiSurface`** — name [`/prototype`](../prototype.md)
+  and stop there. The signal carries **no routing authority and adds no
+  gate**: both halves are derived from observables already in the checkout —
+  the `hasWebSurface` applicability predicate the `target: "web"` audit
+  lenses gate on, and whether any predicted path matches a web lens
+  `filePattern` in `audit-rules.json`; a project with no rendered frontend
+  resolves falsey and the offer never fires. `/mandrel-plan` must never invoke
+  it — operator invocation is the entire design, because the value is a human
+  looking at a layout before its UI acceptance criteria are frozen. Under
+  `--yes` the offer is recorded and planning proceeds — no reroute, no
+  prototype written, no gate raised.
+
+## `complexitySignals` are advisory
+
+The envelope's `complexitySignals` field carries the paths the seed predicts,
+their repo state (existing paths predict refactors; missing predict creates)
+and the `audit-rules.json` sensitive-path classes the footprint intersects —
+`routingAuthority: false`, no `route` field. They ground the authoring
+template's pre-resolved `changes[]` and the `/prototype` offer, nothing else.
+Story #5312 deleted the plan-side lite claim that used to read them
+(`--route-downgrade-reason`, the persist shape backstop, the `route::lite`
+hint): every Story lands through the same engine and the same close gates,
+and ceremony is derived from the landed diff at close.
 
 ## Correct-by-construction authoring template
 
@@ -210,30 +151,24 @@ ceilings on `STORY_SHAPE_CEILINGS` in
 **correct-by-construction** skeleton, built from the same repo probe the
 `complexitySignals` ran:
 
-- **`verify[]` placeholders already end with a valid `(tier)` tag.** Keep
-  every filled entry's trailing tag one of `(unit)` / `(contract)` /
-  `(e2e)` / `(validate)` (or use the `manual:<reason>` escape) — a tierless
-  entry is exactly the mechanical persist round-trip the template exists to
-  prevent.
+- **`verify[]` entries are commands.** There is no tier suffix and no
+  `manual:<reason>` escape (Story #5312): write the exact command or test
+  path the deliverer runs and the acceptance critic reads as evidence.
 - **`changes[]` arrive pre-resolved to creates-vs-refactors.** Every path
   the seed predicted is probed against the repo: an existing path is
   emitted with `assumption: "refactors-existing"`, a missing one with
-  `assumption: "creates"`. Trust the pre-resolved assumption — verify
-  against the repo before overriding one (authoring `creates` for a file
-  that exists at base is a validator rejection). The persist gates stay
-  authoritative: they probe the base branch ref, not the working tree.
-- **Keep `## Spec` near contract-level prose.** **Aim for ~250 words; an
-  advisory warning fires past 350** (`SPEC_SOFT_WORD_BUDGET`). Two numbers,
-  two jobs: ~250 is the authoring target — the nudge toward a contract-level
-  Spec (interfaces, invariants, load-bearing constraints; no per-file
-  behavior narration) — while 350 is the slacker threshold at which persist
-  actually warns, so the warning marks a real outlier instead of ordinary
-  variance. Neither fails the persist. The hard fail-closed ceiling
-  (~1500 tokens, `spec-spill.js`) is unchanged.
+  `assumption: "creates"`. The persist gates stay authoritative — they probe
+  the base branch ref, not the working tree — but a `creates` on a path that
+  exists at base, or a `refactors-existing` on one that does not, is a
+  dry-run **warning**, not a rejection; only a `deletes` naming an absent
+  path is refused. A plain-string bullet or a trailing parenthetical is
+  repaired into the object form by probing base, and the repair is reported.
+- **Keep `## Spec` at contract-level prose** — interfaces, invariants,
+  load-bearing constraints; no per-file behavior narration — and as long as
+  the work needs. There is no word or token budget.
 
 A faithfully-filled skeleton — placeholders replaced, pre-resolved entries
-kept, tags valid — passes the persist ticket validators with no
-round-trip.
+kept — passes the persist ticket validators with no round-trip.
 
 ### Authored entry shape
 
@@ -313,18 +248,13 @@ together, and a path two same-wave Stories both write is exactly where that
 promise breaks. Promise and caveat belong on one durable surface — previously
 the caveat was a stderr warning nobody kept.
 
-Two `planning.*` knobs upgrade a conflict class from advisory to a hard
-refusal. **Both default to `false` and are documented, not recommended:**
-
-| Knob | Upgrades | Why it is off |
-| --- | --- | --- |
-| `planning.failOnSharedEditors` | `shared-editor` → `hard` | Co-editing one file is routine and often correct; the delivery scheduler already serializes file-overlapping Stories. |
-| `planning.requireExplicitCrossStoryDeps` | `implicit-cross-story-dep` → `hard` | Path references are matched by substring, so a legitimate mention in prose can read as a dependency. |
-
-Turn one on for a repo where the class is genuinely fatal; expect a refusal to
-name the Stories and the fix (a `depends_on` edge, or folding the shared edit
-into one Story). The sibling knobs `failOnRegistryConflicts`,
-`failOnMissingBddScaffold` and `failOnLargeFanOut` behave the same way.
+Every conflict class is advisory (Story #5312 retired the
+`planning.failOn*` / `requireExplicitCrossStoryDeps` upgrade knobs with the
+registry and fan-out findings): co-editing one file is routine and often
+correct — the delivery scheduler already serializes file-overlapping Stories —
+and a path reference matched by substring can read as a dependency a prose
+mention never meant. A finding names the Stories and the fix (a `depends_on`
+edge, or folding the shared edit into one Story) for the operator to weigh.
 
 ## Tickets mode — authoring `supersedes[]`
 
@@ -359,67 +289,73 @@ total by default — an authored map is the only thing that can say
 `#11-#14 → #20` while `#15 → #21`, which a blanket "superseded by
 this plan-run" reference could not.
 
-## Critic dispatch detail
+## The pre-mortem critic — operator-invoked
+
+The maker-blind **pre-mortem** critic is not a step of the spine (Story #5312
+retired step 2.5 with the consolidation critic, whose one deterministic input
+was a `## Delivery Slicing` table no Story carries). Run it when the operator
+asks for it, after Author and before Persist — the last point a finding folds
+into a re-author:
+
+```bash
+node .agents/scripts/plan-critics.js \
+  --stories temp/plan-<slug>/stories.json \
+  [--tech-spec temp/plan-<slug>/techspec.md]
+```
 
-The **pre-mortem** critic fires on any of three deterministic triggers: the
-draft ticket count reaching half the reviewability budget, a
-`planning.riskHeuristics` phrase matching the plan text, or the
-**external-dependency** probe finding an out-of-repo marker — a
-scoped package the plan names that no repo manifest declares, a cross-repo
-`github.com/<owner>/<repo>` reference, or an endpoint named as a service
-prerequisite. That third trigger is what gives the default N=1 plan a cheap
-viability check, since the size trigger is unreachable at one ticket and this
-repo's resolved `riskHeuristics` is empty. The probe is conservative — explicit
-markers only, so a plan naming no such artifact dispatches exactly as before.
+It fires on one deterministic trigger: the **external-dependency** probe
+finding an out-of-repo marker — a scoped package the plan names that no repo
+manifest declares, a cross-repo `github.com/<owner>/<repo>` reference, or an
+endpoint named as a service prerequisite. The probe is conservative —
+explicit markers only. It exits 0 on **any** verdict (verdicts route work,
+they do not gate) and exits **1** only on a usage/IO error — no critic ran.
 
 ```jsonc
 {
-  "consolidation": { "critic": "consolidation", "dispatch": false, "reasons": ["…"] },
-  "premortem": { "critic": "pre-mortem", "dispatch": true, "reasons": ["…"] },
-  "textHygiene": { "critic": "text-hygiene", "findings": [] }
+  "premortem": { "critic": "pre-mortem", "dispatch": true, "reasons": ["…"] }
 }
 ```
 
-The verdict's third entry, `textHygiene`, is advisory-only: it
-carries deterministic body lints (`dangling-citation` / `open-question` /
-`slicing-mass`) with no dispatch semantics — it spawns nothing and never
-gates the run. Fold `textHygiene.findings[]` into the re-author round the
-same way critic findings fold in: fix each named defect in `stories.json`
-(anchor or inline the citation, resolve the question into a declarative
-assumption, thin the Slicing checkpoint) and re-run the critic step. Empty
-`findings` add nothing to the round.
-
-**Dispatch shape.** When `delivery.routing.roleScopedAgents` is enabled (the
-**default**), dispatch each firing critic with `subagent_type: plan-critic` —
-it boots on the role-scoped [`plan-critic`](../../agents/plan-critic.md)
-context (its own system prompt, no `CLAUDE.md` @-closure) that carries the
-maker-blind invariant, the `consolidation` and `pre-mortem` charters, and the
-output shape standalone. When the kill-switch is off
-(`roleScopedAgents: false`) or the host cannot spawn at this depth, fall back
-to a generic sub-agent and hand it the same charter (the `consolidation` /
-`pre-mortem` definitions in [`plan-critic.md`](../../agents/plan-critic.md)).
-**When both critics fire, dispatch them in a single turn.** Consolidation and
-pre-mortem read the same immutable draft, share no write path, and neither
-consumes the other's verdict — the textbook independent fan-out of
-[`parallel-tooling.md`](parallel-tooling.md) Rule 3. Issue both `Agent` calls
-together in one assistant turn rather than awaiting the first verdict before
-spawning the second; serialized critics double the round's wall clock and buy
-nothing, because you fold both verdicts into the same re-author round anyway.
-
-Either way the critic is **maker-blind**: hand it the draft artifacts
-(`stories.json`, and `techspec.md` when present) — never the authoring
-transcript or the reasons the planner believed its own draft is sound. A
-critic that reads the maker's case grades the case, not the draft.
+On `dispatch: true`, dispatch **one fresh-context, maker-blind sub-agent**.
+When `delivery.routing.roleScopedAgents` is enabled (the **default**), use
+`subagent_type: plan-critic` — it boots on the role-scoped
+[`plan-critic`](../../agents/plan-critic.md) context (its own system prompt,
+no `CLAUDE.md` @-closure) that carries the maker-blind invariant, the
+`pre-mortem` charter, and the output shape standalone. When the kill-switch
+is off (`roleScopedAgents: false`) or the host cannot spawn at this depth,
+fall back to a generic sub-agent and hand it the same charter. Either way the
+critic is **maker-blind**: hand it the draft artifacts (`stories.json`, and
+`techspec.md` when present) — never the authoring transcript or the reasons
+the planner believed its own draft is sound. A critic that reads the maker's
+case grades the case, not the draft. Fold surviving findings into Gate #2 or
+a re-author round.
 
 ## What `--dry-run` actually gates
 
 `plan-persist.js --dry-run` is the same command with GitHub writes suppressed,
-and every gate runs before the first `createIssue` would fire — the validator,
-the body parse, the DAG, the capacity and Spec-budget ceilings, the
-reachability check, the split and supersede partitions, and the Tech Spec fold.
-That is the whole point of running it first: a dry run that comes back clean
-has already paid for every deterministic refusal, so the real persist has
-nothing left to discover except network failure.
+and every gate runs before the first `createIssue` would fire. Since
+Story #5312 the gates split two ways, and the dry-run is where the second
+half is read:
+
+**Hard — the run refuses:** a body that does not parse, a ticket that is not
+a Story, an empty `acceptance[]` or `verify[]`, an unknown or cyclic
+`depends_on`, the acceptance partition at N>1, the supersede partition, a
+forbidden commit-subject prefix, and a `deletes` entry naming a path absent
+at base.
+
+**Warnings — listed, then the persist proceeds:** a `creates` on a path that
+exists at base or a `refactors-existing` on one that does not (including a
+path the base branch deleted or renamed, named with the removing commit), a
+goal or acceptance path absent at base, a `verify[]` command naming an absent
+test file, and an `open-question` in a body (`Flag if…`, `TBD`, a trailing
+`?`). The list also names every `changes[]` **repair** the run applied — a
+plain-string bullet or a trailing parenthetical rewritten into
+`{ path, assumption }` by probing base. The same list rides the result
+envelope as `warnings[]` and `repairs[]`, so a `--chain-on-clean` run loses
+nothing.
+
+A dry run that comes back clean has paid for every deterministic refusal, so
+the real persist has nothing left to discover except network failure.
 
 ## The container Epic (Gate #3)