hatch3r 2.1.1 → 2.2.0

package/dist/content/rules/hatch3r-agent-orchestration.md

@@ -39,7 +39,7 @@ Score task complexity per the `hatch3r-deep-context` rule before Phase 1. Apply
 - **Tier 2 hard gate (B1).** Before Phase 2, run `hatch3r-researcher` with `requirements-elicitation:quick` mode to detect ambiguity. The researcher sub-agent emits the elicited questions as a structured list in its result — it does NOT call the platform-native question tool (a spawned sub-agent has no interactive surface). The **orchestrator** renders that list to the user via the platform-native question tool (per `agents/shared/user-question-protocol.md`), mirroring `commands/hatch3r-workflow.md` Phase 1 "Present the `requirements-elicitation` questions inline and await answers". Orchestrator awaits answers and integrates them into the Phase 1 brief; do not begin Phase 2 with unresolved questions. Tier 1 is exempt only when scope is single-file, single-concern, and acceptance is testable from the user message alone.
 - **Tier 3 (Deep):** Present Pre-Implementation Summary and ASK for confirmation. Do NOT proceed until all unresolved questions are answered.
 
-**Tier-to-Phase-4 specialist depth mapping** (Finding D7-M9 / D7-SA7.4-1). Deep-context tier drives Phase 1 researcher depth; the same tier drives Phase 4 specialist depth so quality coverage scales with task risk: Tier 1 → run only the always-mode floor (`hatch3r-security` + `hatch3r-testability`) at `quick` depth — UI/perf/maintainability/etc. specialists are skipped per Phase Skip Criteria. Tier 2 → always-mode floor at `standard` depth + every triggered conditional specialist at `quick` depth. Tier 3 → every applicable specialist at `deep` depth — full WCAG AA / OWASP ASI / CWV / mandate-map sweep with N=3 sampling on always-mode specialists when `floor:security` items are touched (per `agents/shared/quality-charter.md` -> Non-Determinism Budget). The depth signal rides on the specialist prompt as the explicit field `depth: quick | standard | deep`; specialists read it via the shared `agents/shared/quality-specialist-frame.md`. Tier also drives **model class** as a first-order effort lever (Tier 1 → economy, Tier 2 → default, Tier 3 → strongest), resolved per-adapter against `models.default` / `src/models/resolve.ts` and ignored where an adapter has no model-routing surface — see `hatch3r-deep-context` -> Tier Assignment.
+**Tier-to-Phase-4 specialist depth mapping** (Finding D7-M9 / D7-SA7.4-1). Deep-context tier drives Phase 1 researcher depth; the same tier drives Phase 4 specialist depth so quality coverage scales with task risk: Tier 1 → run only the always-mode floor (`hatch3r-security` + `hatch3r-testability`) at `quick` depth — UI/perf/maintainability/etc. specialists (mandatory-on-match included) are skipped per Phase Skip Criteria. Tier 2 → always-mode floor at `standard` depth + every triggered conditional specialist at `quick` depth, and a triggered mandatory-on-match specialist (`hatch3r-ui` CQ1 / `hatch3r-ux` CQ2) MUST spawn as its own dedicated sub-agent instance (never merged into one spawn) — skipping a triggered one at Tier 2/3 is a gate failure. Tier 3 → every applicable specialist at `deep` depth (same mandatory-on-match mandate) — full WCAG AA / OWASP ASI / CWV / mandate-map sweep with N=3 sampling on always-mode specialists when `floor:security` items are touched (per `agents/shared/quality-charter.md` -> Non-Determinism Budget). The depth signal rides on the specialist prompt as the explicit field `depth: quick | standard | deep`; specialists read it via the shared `agents/shared/quality-specialist-frame.md`. Tier also drives **model class** as a first-order effort lever (Tier 1 → economy, Tier 2 → default, Tier 3 → strongest), resolved per-adapter against `models.default` / `src/models/resolve.ts` and ignored where an adapter has no model-routing surface — see `hatch3r-deep-context` -> Tier Assignment.
 
 ## Mandatory Delegation Directives
 
@@ -49,7 +49,7 @@ Spawn `hatch3r-researcher` before implementing any task (skip only for trivial s
 
 ### Research Completeness Checklist
 
-Before Phase 1 to Phase 2 handoff, verify all four: (1) **affected files identified** (create/modify/delete listed); (2) **blast radius assessed** (downstream consumers + integration points documented); (3) **existing tests located** (or absence noted); (4) **dependencies mapped** (internal + external). If any item is unconfirmed, re-run researcher with additional modes or surface to user.
+Before Phase 1 to Phase 2 handoff, verify all five: (1) **affected files identified** (create/modify/delete listed); (2) **blast radius assessed** (downstream consumers + integration points documented); (3) **existing tests located** (or absence noted); (4) **dependencies mapped** (internal + external); (5) **shared contracts inventoried** (every contract the change will mutate — exported symbol, persisted name, wire field, event schema, shared constant — listed with a first-pass consumer grep count per `rules/hatch3r-contract-census.md` → Shared-Contract Taxonomy). If any item is unconfirmed, re-run researcher with additional modes or surface to user.
 
 ### Implementation Delegation
 
@@ -105,20 +105,20 @@ For multi-sub-task implementations, the implementer performs a lightweight mini-
 **Phase 3 — Review Loop:**
 
 1. Spawn `hatch3r-reviewer` with diff, acceptance criteria, and blast-radius summary.
-2. Critical/Warning findings: spawn `hatch3r-fixer` with full reviewer output, then re-review. Repeat until 0 Critical + 0 Warning, or max 4 iterations (matches `DEFAULT_MAX_REVIEW_ITERATIONS` in `src/pipeline/reviewLoop.ts`, raised 3→4 in Cycle 7.5 W2B2 H26 to reach the oscillation detector in default config; kept in sync by `src/__tests__/pipeline/reviewLoop.test.ts`, CI-enforced).
+2. Critical/Warning findings: spawn `hatch3r-fixer` with full reviewer output, then re-review. Repeat until 0 Critical + 0 Warning, or max 4 iterations (matches `DEFAULT_MAX_REVIEW_ITERATIONS` in `src/pipeline/reviewLoop.ts`, raised 3→4 in Cycle 7.5 W2B2 H26 to reach the oscillation detector in default config; kept in sync by `src/__tests__/pipeline/reviewLoop.test.ts`, CI-enforced). Before each fixer dispatch, append the W1 write-ahead rows; after each re-review, append the W2 disposition rows (`rules/hatch3r-findings-ledger.md` → Write Points).
    - **Re-run honor-rule (anti-self-approval, F15.2-H2 / D13-SA13.2-F6 / D15-SA15.2-F3).** The fixer's `Reviewer re-run required` boolean is advisory; the orchestrator derives the authoritative value `reRunRequired = (fixer Files changed list is non-empty)`. When `true`, another `hatch3r-reviewer` pass is MANDATORY before the loop may exit clean — a fixer `Status: SUCCESS` with a non-empty `Files changed` can never self-approve to a clean exit. A `Reviewer re-run required: false` printed alongside a non-empty `Files changed` is overridden to `true` and noted as a self-declared protocol violation. The `Files changed` list is SSOT-bound: it is attested by the same fixer `delegation_proof_id` the orchestrator quotes in its End-of-Turn Delegation Attestation, so the signal cannot be forged without spawning the fixer.
 3. **Confirmation pass** after clean review: lightweight re-review checking only (a) no new test failures vs Phase 2 baseline, (b) no type errors introduced, (c) acceptance criteria still met. Does not re-run the full checklist.
    - **Runtime calibration second pass (orchestrator-owned).** At this would-be-clean exit, the orchestrator — not the stateless reviewer — evaluates the `rules/hatch3r-reviewer-calibration.md` trigger: read the cross-run `consecutive_clean_pass_count` from `.hatch3r/calibration-state.json`, increment it for this clean run, and fire a second-pass review (different model class, else same-class re-roll) when the post-increment count is a multiple of `N` (default 5) OR — for a high-risk diff (`floor:security` / auth / CQ3-security-dispatch files) — on the first clean PASS. A divergent second pass reverts the exit to step 2 (`REQUEST CHANGES`) and resets the count to 0; persist the count (atomic write via `src/merge/safeWrite.ts`) and append the calibration-log record. A REQUEST CHANGES or DESIGN_OBJECTION at any iteration also resets the persisted count to 0.
-4. Max iterations reached: surface to user with a structured summary (iteration count, remaining Critical findings with file:line, remaining Warnings, fix-manually-vs-accept-risk recommendation). Never present raw reviewer output unsummarized.
+4. Max iterations reached: surface to user with a structured summary (iteration count, remaining Critical findings with file:line, remaining Warnings, fix-manually-vs-accept-risk recommendation). Never present raw reviewer output unsummarized. Reconcile the ledger to the run-exit invariant (W3) before surfacing; the structured summary lists the open `finding_id`s.
 5. **Review gate confidence signal:** on a clean verdict, record the iteration count in `PipelineContext.reviewResult.iterations`. Clean-on-first-pass signals higher confidence than clean-after-multiple-iterations; Phase 4 and the orchestrator factor this into risk assessment.
 
 **Phase 4 — Final Quality** (after review loop is clean):
 
-Launch Phase 4 specialists in parallel, bounded by an orchestrator-honored fan-out width `max_phase4_parallel` (default `8` — covers the empirical maximum of applicable specialists per the trigger table, so a typical Tier 3 change fans out in at most 2 batches). This bound is LLM-honored orchestrator guidance, not a code-enforced cap: the host Task tool is the actual dispatcher and applies no platform fan-out limit, so no hatch3r module reads an env var or clamps the count — the orchestrator self-limits per this prose. The bound exists for upstream provider rate-limit headroom (RPM/TPM) — a true dependency edge — NOT per-orchestrator context cost; token cost never serializes independent work (P8 dominates P7). A non-rate-limited orchestrator MAY raise the width up to the full applicable-specialist set. **Rate-limit back-off (orchestrator-LLM guidance):** when the orchestrator observes ≥3 consecutive rate-limit-class transient failures, reduce the active fan-out width by 1 for the next batch and record the back-off in the Iteration Summary; never silently cap a healthy run. When applicable specialists exceed the bound, batch by severity-descending priority `CRITICAL → HIGH → MEDIUM → LOW` (severity is the worst-case finding class the specialist surfaces: always-on testability (CQ5) / security (CQ3) → CRITICAL, conditional CQ1/CQ4/CQ7 (ui/reliability/performance) → HIGH, docs/lint → MEDIUM, low-impact → LOW); within a bucket, dispatch in trigger-table order. Each batch runs to completion before the next starts; the validation pass runs once after the final batch. The applicable specialists and their trigger conditions are listed in the Phase 4 Specialist Trigger Table below.
+Launch Phase 4 specialists in parallel, bounded by an orchestrator-honored fan-out width `max_phase4_parallel` (default `8` — covers the empirical maximum of applicable specialists per the trigger table, so a typical Tier 3 change fans out in at most 2 batches). This bound is LLM-honored orchestrator guidance, not a code-enforced cap: the host Task tool is the actual dispatcher and applies no platform fan-out limit, so no hatch3r module reads an env var or clamps the count — the orchestrator self-limits per this prose. The bound exists for upstream provider rate-limit headroom (RPM/TPM) — a true dependency edge — NOT per-orchestrator context cost; token cost never serializes independent work (P8 dominates P7). A non-rate-limited orchestrator MAY raise the width up to the full applicable-specialist set. **Rate-limit back-off (orchestrator-LLM guidance):** when the orchestrator observes ≥3 consecutive rate-limit-class transient failures, reduce the active fan-out width by 1 for the next batch and record the back-off in the Iteration Summary; never silently cap a healthy run. When applicable specialists exceed the bound, batch by severity-descending priority `CRITICAL → HIGH → MEDIUM → LOW` (severity is the worst-case finding class the specialist surfaces: always-on testability (CQ5) / security (CQ3) → CRITICAL, mandatory-on-match CQ1/CQ2 (ui/ux) + conditional CQ4/CQ7 (reliability/performance) → HIGH, docs/lint → MEDIUM, low-impact → LOW); within a bucket, dispatch in trigger-table order. Each batch runs to completion before the next starts; the validation pass runs once after the final batch. The applicable specialists and their trigger conditions are listed in the Phase 4 Specialist Trigger Table below. The width bound and the rate-limit back-off never drop a triggered mandatory-on-match specialist — `hatch3r-ui`/`hatch3r-ux` dispatch as dedicated instances in the first HIGH batch.
 
 **Specialist Prompt Enrichment:** When spawning Phase 4 specialists, include the Phase 2 `filesChanged` list (focus on affected code), the Phase 3 review verdict summary (avoid re-flagging reviewed issues), and `researchFindings.blastRadius` (assess downstream impact).
 
-**Runtime trigger evaluation (D6-M11):** the orchestrator harness calls `shouldTriggerSpecialist(specialist, changedFiles, projectType)` from `src/pipeline/pipelineContext.ts` to evaluate whether each specialist applies to the current change set. The function returns `{ triggered: boolean, reasons: string[] }` and consumes the same `SPECIALIST_TRIGGER_TABLE` that the prose table below mirrors. Treat the prose as a quick reference; treat the TS predicate as the authoritative gate. `npm run validate:specialist-roster` enforces parity.
+**Runtime trigger evaluation (D6-M11):** the orchestrator harness calls `shouldTriggerSpecialist(specialist, changedFiles, projectType)` from `src/pipeline/pipelineContext.ts` to evaluate whether each specialist applies to the current change set. The function returns `{ triggered: boolean, reasons: string[], mandatory?: boolean }` and consumes the same `SPECIALIST_TRIGGER_TABLE` that the prose table below mirrors. Treat the prose as a quick reference; treat the TS predicate as the authoritative gate. `npm run validate:specialist-roster` enforces parity. The orchestrator treats `mandatory: true` as non-skippable at Tier 2/3.
 
 **Phase 4 Specialist Trigger Table:**
 
@@ -128,8 +128,8 @@ Launch Phase 4 specialists in parallel, bounded by an orchestrator-honored fan-o
 | `hatch3r-lint-fixer` | Conditional | Lint/type errors present |
 | `hatch3r-architect` | Conditional | Architectural decisions, new modules/services |
 | `hatch3r-devops` | Conditional | CI/CD or infrastructure changes |
-| `hatch3r-ui` (CQ1) | Conditional | UI component / theme / token files modified (`*.{tsx,jsx,vue,svelte}`, `tailwind.config.*`, design-token registries) |
-| `hatch3r-ux` (CQ2) | Conditional | Flow / modal / route-transition / error-state files modified; microcopy or i18n strings changed |
+| `hatch3r-ui` (CQ1) | Mandatory-on-match | UI component / theme / token files modified (`*.{tsx,jsx,vue,svelte}`, `tailwind.config.*`, design-token registries) |
+| `hatch3r-ux` (CQ2) | Mandatory-on-match | Flow / modal / route-transition / error-state files modified; microcopy or i18n strings changed |
 | `hatch3r-security` (CQ3) | Always | Any code change (always-mode floor — absorbs legacy security-auditor scope); auth / JWT / OAuth / WebAuthn code modified; release workflow modified; cookie or session handling modified; dependency files modified |
 | `hatch3r-reliability` (CQ4) | Conditional | Service handler / OTel / SLO / retry / circuit-breaker code modified; Kubernetes probe manifests modified |
 | `hatch3r-testability` (CQ5) | Always | Any code change (always-mode floor — absorbs legacy test-writer scope); test code added or modified; mandate-map feature class introduced; coverage threshold or runner config modified |
@@ -173,9 +173,9 @@ Default is **linear per task** (Phase 1 → 2 → 3 → 4 serially) — `Pipelin
 
 ### Three Conditions to Parallelize
 
-ALL three must hold: (1) **read-only or disjoint writes** (no conflict zone); (2) **deterministic aggregation** (outputs merge without orchestrator intervention — tests pass-if-all-pass, findings union); (3) **no shared mutable state** (agents that mutate `PipelineContext.state`/`featureFlags`/`metadata` serialize; parallel agents only READ).
+ALL three must hold: (1) **read-only or disjoint writes, at file AND contract granularity** (no conflict zone: two lanes are NOT disjoint when their file-disjoint diffs mutate the same shared contract — exported symbol, persisted collection/field name, wire field, event schema, shared constant; on contract overlap, assign one seam owner or serialize per `rules/hatch3r-contract-census.md` → Seam-Owner Protocol); (2) **deterministic aggregation** (outputs merge without orchestrator intervention — tests pass-if-all-pass, findings union); (3) **no shared mutable state** (agents that mutate `PipelineContext.state`/`featureFlags`/`metadata` serialize; parallel agents only READ).
 
-**Parallel-safe:** Phase 4 specialists; intra-Phase-1 researcher modes on a self-contained task; per-module Phase 2 fan-out on disjoint `affectedFiles` (merged post-Phase-2); Tier 2/3 elicitation researchers (outputs tagged with confidence + perspective). **NOT parallel-safe:** cross-phase execution (each phase depends on the prior's output); Phase 3 review-loop iterations (reviewer → fixer → re-reviewer serial); overlapping-file implementers (serialize or use a merge-conflict gate); Phase 4 validation re-review.
+**Parallel-safe:** Phase 4 specialists; intra-Phase-1 researcher modes on a self-contained task; per-module Phase 2 fan-out on disjoint `affectedFiles` AND disjoint contract sets (merged post-Phase-2); Tier 2/3 elicitation researchers (outputs tagged with confidence + perspective). **NOT parallel-safe:** cross-phase execution (each phase depends on the prior's output); Phase 3 review-loop iterations (reviewer → fixer → re-reviewer serial); overlapping-file implementers (serialize or use a merge-conflict gate); contract-overlapping implementers even when file-disjoint (one seam owner mutates, peers consume — `rules/hatch3r-contract-census.md`); Phase 4 validation re-review.
 
 **Cost-Dominance Principle.** Token cost of sub-agent invocation never justifies serialization of independent work. The three safety conditions govern WHEN parallelism is safe; cost does not govern WHETHER to parallelize. When in doubt, fan out. Serialization is only valid on true dependency edges.
 
@@ -183,13 +183,13 @@ ALL three must hold: (1) **read-only or disjoint writes** (no conflict zone); (2
 
 ### Concurrent Invocation Handling
 
-Two top-level pipelines running at once (e.g. `hatch3r-workflow` in one shell, `hatch3r-board-pickup` in another) are bounded by the same three conditions, applied cross-pipeline (D7-SA7.5-F7.5.6): (1) **Advisory lock-note (best-effort, NOT atomic — D7-27)** — before Phase 1, the orchestrator writes/reads `.hatch3r/.lock` (JSON: `pid`, `command`, `branch`, `correlation_id`, `started_at`); clear on completion/abort; treat a note older than 6h as stale. This is an advisory coordination note, not a mutual-exclusion primitive: there is no `hatch3r` lock verb and no atomic acquire in `src/` (the only cross-process lock that ships is `acquireWriteLock` in `src/merge/safeWrite.ts`, scoped to single-file atomic writes, not top-level pipelines), so the LLM orchestrator's read-then-write is TOCTOU by construction and `src/cli/commands/status.ts` already labels this file "advisory". It exists to surface a likely collision, never to guarantee exclusion. (2) **Detect-then-warn** — if a live note names the same branch or an open `.hatch3r/hatch.json` board transaction, WARN and ASK (proceed on a separate branch / wait / abort); never silently co-mutate shared state. (3) **True isolation via worktree, not the lock-note (D7-27)** — when concurrency must actually be conflict-free rather than merely warned (the parallel-implementer path), route each pipeline through `hatch3r worktree-setup <name>` — the isolation primitive hatch3r already ships (`src/cli/commands/worktreeSetup.ts`; `commands/board/pickup-delegation-multi.md` already uses it per implementer) — so the pipelines write disjoint working trees and integrate back on completion, satisfying the disjoint-writes safety condition without relying on the non-atomic note. (4) **Cache-sharing** — `.hatch3r/learnings/` is read-many, write-once-at-completion with timestamp-ordered conflict resolution. Cross-task context sharing is bounded by the no-shared-mutable-state condition: learnings consolidate at pipeline completion; mid-pipeline writes are out of scope to preserve parallel-safety determinism (D7-SA7.5-F7.5.8). Each command's Guardrails cite this subsection.
+Two top-level pipelines running at once (e.g. `hatch3r-workflow` in one shell, `hatch3r-board-pickup` in another) are bounded by the same three conditions, applied cross-pipeline (D7-SA7.5-F7.5.6): (1) **Advisory lock-note (best-effort, NOT atomic — D7-27)** — before Phase 1, the orchestrator writes/reads `.hatch3r/.lock` (JSON: `pid`, `command`, `branch`, `correlation_id`, `started_at`); clear on completion/abort; treat a note older than 6h as stale. This is an advisory coordination note, not a mutual-exclusion primitive: there is no `hatch3r` lock verb and no atomic acquire in `src/` (the only cross-process lock that ships is `acquireWriteLock` in `src/merge/safeWrite.ts`, scoped to single-file atomic writes, not top-level pipelines), so the LLM orchestrator's read-then-write is TOCTOU by construction and `src/cli/commands/status.ts` already labels this file "advisory". It exists to surface a likely collision, never to guarantee exclusion. (2) **Detect-then-warn** — if a live note names the same branch or an open `.hatch3r/hatch.json` board transaction, WARN and ASK (proceed on a separate branch / wait / abort); never silently co-mutate shared state. (3) **True isolation via worktree, not the lock-note (D7-27)** — when concurrency must actually be conflict-free rather than merely warned (the parallel-implementer path), route each pipeline through `hatch3r worktree-setup <name>` — the isolation primitive hatch3r already ships (`src/cli/commands/worktreeSetup.ts`; `commands/board/pickup-delegation-multi.md` already uses it per implementer) — so the pipelines write disjoint working trees and integrate back on completion, satisfying the disjoint-writes safety condition without relying on the non-atomic note. Worktree isolation guarantees FILE-level disjointness only — it cannot detect two lanes mutating the same shared contract (a persisted collection name, an exported store symbol, a wire field) from different files; contract disjointness is established before dispatch and re-checked at lane exit per `rules/hatch3r-contract-census.md`. (4) **Cache-sharing** — `.hatch3r/learnings/` is read-many, write-once-at-completion with timestamp-ordered conflict resolution. Cross-task context sharing is bounded by the no-shared-mutable-state condition: learnings consolidate at pipeline completion; mid-pipeline writes are out of scope to preserve parallel-safety determinism (D7-SA7.5-F7.5.8). Each command's Guardrails cite this subsection.
 
 ## Cross-Phase Error Propagation
 
-On a non-SUCCESS status, the orchestrator MUST propagate error context downstream, never silently drop it. **Phase 1 PARTIAL:** include `researchGaps` in the implementer prompt and set confidence expectations accordingly. **Phase 2 PARTIAL:** include `reason` + unimplemented acceptance criteria in the reviewer prompt (reviewer distinguishes "not done yet" from "done incorrectly"). **Phase 3 UNRESOLVED:** include the unresolved findings in Phase 4 specialist prompts (specialists must not conflict with known issues). **Phase 4 specialist FAILED:** include the failure reason when surfacing — never report "Phase 4 failed" without naming which specialist and why.
+On a non-SUCCESS status, the orchestrator MUST propagate error context downstream, never silently drop it. **Phase 1 PARTIAL:** include `researchGaps` in the implementer prompt and set confidence expectations accordingly. **Phase 2 PARTIAL:** include `reason` + unimplemented acceptance criteria in the reviewer prompt (reviewer distinguishes "not done yet" from "done incorrectly"). **Phase 3 UNRESOLVED:** include the unresolved findings in Phase 4 specialist prompts (specialists must not conflict with known issues). Unresolved findings persist as ledger rows (`rules/hatch3r-findings-ledger.md`); Phase 4 prompts carry the open `finding_id` list. **Phase 4 specialist FAILED:** include the failure reason when surfacing — never report "Phase 4 failed" without naming which specialist and why.
 
-**Sub-agent-failure handling (shared clause; all commands cite this — never inline).** When any spawned implementer/fixer/specialist sub-agent FAILS or returns no usable output: (1) retry once with the same prompt; (2) if the retry fails, re-spawn `hatch3r-fixer` (Phase 3) with the failure reason + partial output as failure context — `hatch3r-fixer` is the code-mutation path, so the work stays delegated; (3) if the re-spawn also fails, emit `BLOCKED_OTHER` with a one-sentence reason and ASK the user (fix-manually vs adjust-scope vs accept-risk). The orchestrator MUST NOT fall back to inline implementation — that is the issue #73 bypass mode (see Mandatory Delegation Directive). The sole exception is `hatch3r-quick-change`, whose Tier-1 carve-out permits inline retry per its declared scope.
+**Sub-agent-failure handling (shared clause; all commands cite this — never inline).** When any spawned implementer/fixer/specialist sub-agent FAILS or returns no usable output: (1) retry once with the same prompt; (2) if the retry fails, re-spawn `hatch3r-fixer` (Phase 3) with the failure reason + partial output as failure context — `hatch3r-fixer` is the code-mutation path, so the work stays delegated; (3) if the re-spawn also fails, emit `BLOCKED_OTHER` with a one-sentence reason and ASK the user (fix-manually vs adjust-scope vs accept-risk). Flush in-flight ledger rows per W4 — before the retry and again before the BLOCKED_OTHER ASK (`rules/hatch3r-findings-ledger.md`). The orchestrator MUST NOT fall back to inline implementation — that is the issue #73 bypass mode (see Mandatory Delegation Directive). The sole exception is `hatch3r-quick-change`, whose Tier-1 carve-out permits inline retry per its declared scope.
 
 ## Correlation ID
 
@@ -247,4 +247,4 @@ All `scope: always` rules apply to every task including sub-agent work; include
 
 - **Tier 1 — always include (every sub-agent):** `hatch3r-security-patterns`, `hatch3r-code-standards`.
 - **Tier 2 — by phase:** `hatch3r-testing` (testability/implementer/reviewer); `hatch3r-accessibility-standards` (ui, UI reviewer); `hatch3r-git-conventions` (orchestrator git ops); `hatch3r-ci-cd` (ci-watcher/devops); `hatch3r-dependency-management` (security CQ3 supply-chain slice).
-- **Tier 3 — on-demand by role + scope:** `hatch3r-api-design`, `hatch3r-secrets-management`, `hatch3r-data-classification`, `hatch3r-performance-budgets`, `hatch3r-browser-verification`, `hatch3r-component-conventions`, `hatch3r-i18n`, `hatch3r-theming`, `hatch3r-migrations`, `hatch3r-feature-flags`, `hatch3r-observability-logging`, `hatch3r-observability-metrics`, `hatch3r-observability-tracing`.
+- **Tier 3 — on-demand by role + scope:** `hatch3r-api-design`, `hatch3r-secrets-management`, `hatch3r-data-classification`, `hatch3r-performance-budgets`, `hatch3r-browser-verification`, `hatch3r-component-conventions`, `hatch3r-i18n`, `hatch3r-theming`, `hatch3r-migrations`, `hatch3r-feature-flags`, `hatch3r-observability-logging`, `hatch3r-observability-metrics`, `hatch3r-observability-tracing`, `hatch3r-dynamic-stack-verification`.

package/dist/content/rules/hatch3r-agent-orchestration.mdc

@@ -34,7 +34,7 @@ Score task complexity per the `hatch3r-deep-context` rule before Phase 1. Apply
 - **Tier 2 hard gate (B1).** Before Phase 2, run `hatch3r-researcher` with `requirements-elicitation:quick` mode to detect ambiguity. The researcher sub-agent emits the elicited questions as a structured list in its result — it does NOT call the platform-native question tool (a spawned sub-agent has no interactive surface). The **orchestrator** renders that list to the user via the platform-native question tool (per `agents/shared/user-question-protocol.md`), mirroring `commands/hatch3r-workflow.md` Phase 1 "Present the `requirements-elicitation` questions inline and await answers". Orchestrator awaits answers and integrates them into the Phase 1 brief; do not begin Phase 2 with unresolved questions. Tier 1 is exempt only when scope is single-file, single-concern, and acceptance is testable from the user message alone.
 - **Tier 3 (Deep):** Present Pre-Implementation Summary and ASK for confirmation. Do NOT proceed until all unresolved questions are answered.
 
-**Tier-to-Phase-4 specialist depth mapping** (Finding D7-M9 / D7-SA7.4-1). Deep-context tier drives Phase 1 researcher depth; the same tier drives Phase 4 specialist depth so quality coverage scales with task risk: Tier 1 → run only the always-mode floor (`hatch3r-security` + `hatch3r-testability`) at `quick` depth — UI/perf/maintainability/etc. specialists are skipped per Phase Skip Criteria. Tier 2 → always-mode floor at `standard` depth + every triggered conditional specialist at `quick` depth. Tier 3 → every applicable specialist at `deep` depth — full WCAG AA / OWASP ASI / CWV / mandate-map sweep with N=3 sampling on always-mode specialists when `floor:security` items are touched (per `agents/shared/quality-charter.md` -> Non-Determinism Budget). The depth signal rides on the specialist prompt as the explicit field `depth: quick | standard | deep`; specialists read it via the shared `agents/shared/quality-specialist-frame.md`. Tier also drives **model class** as a first-order effort lever (Tier 1 → economy, Tier 2 → default, Tier 3 → strongest), resolved per-adapter against `models.default` / `src/models/resolve.ts` and ignored where an adapter has no model-routing surface — see `hatch3r-deep-context` -> Tier Assignment.
+**Tier-to-Phase-4 specialist depth mapping** (Finding D7-M9 / D7-SA7.4-1). Deep-context tier drives Phase 1 researcher depth; the same tier drives Phase 4 specialist depth so quality coverage scales with task risk: Tier 1 → run only the always-mode floor (`hatch3r-security` + `hatch3r-testability`) at `quick` depth — UI/perf/maintainability/etc. specialists (mandatory-on-match included) are skipped per Phase Skip Criteria. Tier 2 → always-mode floor at `standard` depth + every triggered conditional specialist at `quick` depth, and a triggered mandatory-on-match specialist (`hatch3r-ui` CQ1 / `hatch3r-ux` CQ2) MUST spawn as its own dedicated sub-agent instance (never merged into one spawn) — skipping a triggered one at Tier 2/3 is a gate failure. Tier 3 → every applicable specialist at `deep` depth (same mandatory-on-match mandate) — full WCAG AA / OWASP ASI / CWV / mandate-map sweep with N=3 sampling on always-mode specialists when `floor:security` items are touched (per `agents/shared/quality-charter.md` -> Non-Determinism Budget). The depth signal rides on the specialist prompt as the explicit field `depth: quick | standard | deep`; specialists read it via the shared `agents/shared/quality-specialist-frame.md`. Tier also drives **model class** as a first-order effort lever (Tier 1 → economy, Tier 2 → default, Tier 3 → strongest), resolved per-adapter against `models.default` / `src/models/resolve.ts` and ignored where an adapter has no model-routing surface — see `hatch3r-deep-context` -> Tier Assignment.
 
 ## Mandatory Delegation Directives
 
@@ -44,7 +44,7 @@ Spawn `hatch3r-researcher` before implementing any task (skip only for trivial s
 
 ### Research Completeness Checklist
 
-Before Phase 1 to Phase 2 handoff, verify all four: (1) **affected files identified** (create/modify/delete listed); (2) **blast radius assessed** (downstream consumers + integration points documented); (3) **existing tests located** (or absence noted); (4) **dependencies mapped** (internal + external). If any item is unconfirmed, re-run researcher with additional modes or surface to user.
+Before Phase 1 to Phase 2 handoff, verify all five: (1) **affected files identified** (create/modify/delete listed); (2) **blast radius assessed** (downstream consumers + integration points documented); (3) **existing tests located** (or absence noted); (4) **dependencies mapped** (internal + external); (5) **shared contracts inventoried** (every contract the change will mutate — exported symbol, persisted name, wire field, event schema, shared constant — listed with a first-pass consumer grep count per `rules/hatch3r-contract-census.md` → Shared-Contract Taxonomy). If any item is unconfirmed, re-run researcher with additional modes or surface to user.
 
 ### Implementation Delegation
 
@@ -100,20 +100,20 @@ For multi-sub-task implementations, the implementer performs a lightweight mini-
 **Phase 3 — Review Loop:**
 
 1. Spawn `hatch3r-reviewer` with diff, acceptance criteria, and blast-radius summary.
-2. Critical/Warning findings: spawn `hatch3r-fixer` with full reviewer output, then re-review. Repeat until 0 Critical + 0 Warning, or max 4 iterations (matches `DEFAULT_MAX_REVIEW_ITERATIONS` in `src/pipeline/reviewLoop.ts`, raised 3→4 in Cycle 7.5 W2B2 H26 to reach the oscillation detector in default config; kept in sync by `src/__tests__/pipeline/reviewLoop.test.ts`, CI-enforced).
+2. Critical/Warning findings: spawn `hatch3r-fixer` with full reviewer output, then re-review. Repeat until 0 Critical + 0 Warning, or max 4 iterations (matches `DEFAULT_MAX_REVIEW_ITERATIONS` in `src/pipeline/reviewLoop.ts`, raised 3→4 in Cycle 7.5 W2B2 H26 to reach the oscillation detector in default config; kept in sync by `src/__tests__/pipeline/reviewLoop.test.ts`, CI-enforced). Before each fixer dispatch, append the W1 write-ahead rows; after each re-review, append the W2 disposition rows (`rules/hatch3r-findings-ledger.md` → Write Points).
    - **Re-run honor-rule (anti-self-approval, F15.2-H2 / D13-SA13.2-F6 / D15-SA15.2-F3).** The fixer's `Reviewer re-run required` boolean is advisory; the orchestrator derives the authoritative value `reRunRequired = (fixer Files changed list is non-empty)`. When `true`, another `hatch3r-reviewer` pass is MANDATORY before the loop may exit clean — a fixer `Status: SUCCESS` with a non-empty `Files changed` can never self-approve to a clean exit. A `Reviewer re-run required: false` printed alongside a non-empty `Files changed` is overridden to `true` and noted as a self-declared protocol violation. The `Files changed` list is SSOT-bound: it is attested by the same fixer `delegation_proof_id` the orchestrator quotes in its End-of-Turn Delegation Attestation, so the signal cannot be forged without spawning the fixer.
 3. **Confirmation pass** after clean review: lightweight re-review checking only (a) no new test failures vs Phase 2 baseline, (b) no type errors introduced, (c) acceptance criteria still met. Does not re-run the full checklist.
    - **Runtime calibration second pass (orchestrator-owned).** At this would-be-clean exit, the orchestrator — not the stateless reviewer — evaluates the `rules/hatch3r-reviewer-calibration.md` trigger: read the cross-run `consecutive_clean_pass_count` from `.hatch3r/calibration-state.json`, increment it for this clean run, and fire a second-pass review (different model class, else same-class re-roll) when the post-increment count is a multiple of `N` (default 5) OR — for a high-risk diff (`floor:security` / auth / CQ3-security-dispatch files) — on the first clean PASS. A divergent second pass reverts the exit to step 2 (`REQUEST CHANGES`) and resets the count to 0; persist the count (atomic write via `src/merge/safeWrite.ts`) and append the calibration-log record. A REQUEST CHANGES or DESIGN_OBJECTION at any iteration also resets the persisted count to 0.
-4. Max iterations reached: surface to user with a structured summary (iteration count, remaining Critical findings with file:line, remaining Warnings, fix-manually-vs-accept-risk recommendation). Never present raw reviewer output unsummarized.
+4. Max iterations reached: surface to user with a structured summary (iteration count, remaining Critical findings with file:line, remaining Warnings, fix-manually-vs-accept-risk recommendation). Never present raw reviewer output unsummarized. Reconcile the ledger to the run-exit invariant (W3) before surfacing; the structured summary lists the open `finding_id`s.
 5. **Review gate confidence signal:** on a clean verdict, record the iteration count in `PipelineContext.reviewResult.iterations`. Clean-on-first-pass signals higher confidence than clean-after-multiple-iterations; Phase 4 and the orchestrator factor this into risk assessment.
 
 **Phase 4 — Final Quality** (after review loop is clean):
 
-Launch Phase 4 specialists in parallel, bounded by an orchestrator-honored fan-out width `max_phase4_parallel` (default `8` — covers the empirical maximum of applicable specialists per the trigger table, so a typical Tier 3 change fans out in at most 2 batches). This bound is LLM-honored orchestrator guidance, not a code-enforced cap: the host Task tool is the actual dispatcher and applies no platform fan-out limit, so no hatch3r module reads an env var or clamps the count — the orchestrator self-limits per this prose. The bound exists for upstream provider rate-limit headroom (RPM/TPM) — a true dependency edge — NOT per-orchestrator context cost; token cost never serializes independent work (P8 dominates P7). A non-rate-limited orchestrator MAY raise the width up to the full applicable-specialist set. **Rate-limit back-off (orchestrator-LLM guidance):** when the orchestrator observes ≥3 consecutive rate-limit-class transient failures, reduce the active fan-out width by 1 for the next batch and record the back-off in the Iteration Summary; never silently cap a healthy run. When applicable specialists exceed the bound, batch by severity-descending priority `CRITICAL → HIGH → MEDIUM → LOW` (severity is the worst-case finding class the specialist surfaces: always-on testability (CQ5) / security (CQ3) → CRITICAL, conditional CQ1/CQ4/CQ7 (ui/reliability/performance) → HIGH, docs/lint → MEDIUM, low-impact → LOW); within a bucket, dispatch in trigger-table order. Each batch runs to completion before the next starts; the validation pass runs once after the final batch. The applicable specialists and their trigger conditions are listed in the Phase 4 Specialist Trigger Table below.
+Launch Phase 4 specialists in parallel, bounded by an orchestrator-honored fan-out width `max_phase4_parallel` (default `8` — covers the empirical maximum of applicable specialists per the trigger table, so a typical Tier 3 change fans out in at most 2 batches). This bound is LLM-honored orchestrator guidance, not a code-enforced cap: the host Task tool is the actual dispatcher and applies no platform fan-out limit, so no hatch3r module reads an env var or clamps the count — the orchestrator self-limits per this prose. The bound exists for upstream provider rate-limit headroom (RPM/TPM) — a true dependency edge — NOT per-orchestrator context cost; token cost never serializes independent work (P8 dominates P7). A non-rate-limited orchestrator MAY raise the width up to the full applicable-specialist set. **Rate-limit back-off (orchestrator-LLM guidance):** when the orchestrator observes ≥3 consecutive rate-limit-class transient failures, reduce the active fan-out width by 1 for the next batch and record the back-off in the Iteration Summary; never silently cap a healthy run. When applicable specialists exceed the bound, batch by severity-descending priority `CRITICAL → HIGH → MEDIUM → LOW` (severity is the worst-case finding class the specialist surfaces: always-on testability (CQ5) / security (CQ3) → CRITICAL, mandatory-on-match CQ1/CQ2 (ui/ux) + conditional CQ4/CQ7 (reliability/performance) → HIGH, docs/lint → MEDIUM, low-impact → LOW); within a bucket, dispatch in trigger-table order. Each batch runs to completion before the next starts; the validation pass runs once after the final batch. The applicable specialists and their trigger conditions are listed in the Phase 4 Specialist Trigger Table below. The width bound and the rate-limit back-off never drop a triggered mandatory-on-match specialist — `hatch3r-ui`/`hatch3r-ux` dispatch as dedicated instances in the first HIGH batch.
 
 **Specialist Prompt Enrichment:** When spawning Phase 4 specialists, include the Phase 2 `filesChanged` list (focus on affected code), the Phase 3 review verdict summary (avoid re-flagging reviewed issues), and `researchFindings.blastRadius` (assess downstream impact).
 
-**Runtime trigger evaluation (D6-M11):** the orchestrator harness calls `shouldTriggerSpecialist(specialist, changedFiles, projectType)` from `src/pipeline/pipelineContext.ts` to evaluate whether each specialist applies to the current change set. The function returns `{ triggered: boolean, reasons: string[] }` and consumes the same `SPECIALIST_TRIGGER_TABLE` that the prose table below mirrors. Treat the prose as a quick reference; treat the TS predicate as the authoritative gate. `npm run validate:specialist-roster` enforces parity.
+**Runtime trigger evaluation (D6-M11):** the orchestrator harness calls `shouldTriggerSpecialist(specialist, changedFiles, projectType)` from `src/pipeline/pipelineContext.ts` to evaluate whether each specialist applies to the current change set. The function returns `{ triggered: boolean, reasons: string[], mandatory?: boolean }` and consumes the same `SPECIALIST_TRIGGER_TABLE` that the prose table below mirrors. Treat the prose as a quick reference; treat the TS predicate as the authoritative gate. `npm run validate:specialist-roster` enforces parity. The orchestrator treats `mandatory: true` as non-skippable at Tier 2/3.
 
 **Phase 4 Specialist Trigger Table:**
 
@@ -123,8 +123,8 @@ Launch Phase 4 specialists in parallel, bounded by an orchestrator-honored fan-o
 | `hatch3r-lint-fixer` | Conditional | Lint/type errors present |
 | `hatch3r-architect` | Conditional | Architectural decisions, new modules/services |
 | `hatch3r-devops` | Conditional | CI/CD or infrastructure changes |
-| `hatch3r-ui` (CQ1) | Conditional | UI component / theme / token files modified (`*.{tsx,jsx,vue,svelte}`, `tailwind.config.*`, design-token registries) |
-| `hatch3r-ux` (CQ2) | Conditional | Flow / modal / route-transition / error-state files modified; microcopy or i18n strings changed |
+| `hatch3r-ui` (CQ1) | Mandatory-on-match | UI component / theme / token files modified (`*.{tsx,jsx,vue,svelte}`, `tailwind.config.*`, design-token registries) |
+| `hatch3r-ux` (CQ2) | Mandatory-on-match | Flow / modal / route-transition / error-state files modified; microcopy or i18n strings changed |
 | `hatch3r-security` (CQ3) | Always | Any code change (always-mode floor — absorbs legacy security-auditor scope); auth / JWT / OAuth / WebAuthn code modified; release workflow modified; cookie or session handling modified; dependency files modified |
 | `hatch3r-reliability` (CQ4) | Conditional | Service handler / OTel / SLO / retry / circuit-breaker code modified; Kubernetes probe manifests modified |
 | `hatch3r-testability` (CQ5) | Always | Any code change (always-mode floor — absorbs legacy test-writer scope); test code added or modified; mandate-map feature class introduced; coverage threshold or runner config modified |
@@ -168,9 +168,9 @@ Default is **linear per task** (Phase 1 → 2 → 3 → 4 serially) — `Pipelin
 
 ### Three Conditions to Parallelize
 
-ALL three must hold: (1) **read-only or disjoint writes** (no conflict zone); (2) **deterministic aggregation** (outputs merge without orchestrator intervention — tests pass-if-all-pass, findings union); (3) **no shared mutable state** (agents that mutate `PipelineContext.state`/`featureFlags`/`metadata` serialize; parallel agents only READ).
+ALL three must hold: (1) **read-only or disjoint writes, at file AND contract granularity** (no conflict zone: two lanes are NOT disjoint when their file-disjoint diffs mutate the same shared contract — exported symbol, persisted collection/field name, wire field, event schema, shared constant; on contract overlap, assign one seam owner or serialize per `rules/hatch3r-contract-census.md` → Seam-Owner Protocol); (2) **deterministic aggregation** (outputs merge without orchestrator intervention — tests pass-if-all-pass, findings union); (3) **no shared mutable state** (agents that mutate `PipelineContext.state`/`featureFlags`/`metadata` serialize; parallel agents only READ).
 
-**Parallel-safe:** Phase 4 specialists; intra-Phase-1 researcher modes on a self-contained task; per-module Phase 2 fan-out on disjoint `affectedFiles` (merged post-Phase-2); Tier 2/3 elicitation researchers (outputs tagged with confidence + perspective). **NOT parallel-safe:** cross-phase execution (each phase depends on the prior's output); Phase 3 review-loop iterations (reviewer → fixer → re-reviewer serial); overlapping-file implementers (serialize or use a merge-conflict gate); Phase 4 validation re-review.
+**Parallel-safe:** Phase 4 specialists; intra-Phase-1 researcher modes on a self-contained task; per-module Phase 2 fan-out on disjoint `affectedFiles` AND disjoint contract sets (merged post-Phase-2); Tier 2/3 elicitation researchers (outputs tagged with confidence + perspective). **NOT parallel-safe:** cross-phase execution (each phase depends on the prior's output); Phase 3 review-loop iterations (reviewer → fixer → re-reviewer serial); overlapping-file implementers (serialize or use a merge-conflict gate); contract-overlapping implementers even when file-disjoint (one seam owner mutates, peers consume — `rules/hatch3r-contract-census.md`); Phase 4 validation re-review.
 
 **Cost-Dominance Principle.** Token cost of sub-agent invocation never justifies serialization of independent work. The three safety conditions govern WHEN parallelism is safe; cost does not govern WHETHER to parallelize. When in doubt, fan out. Serialization is only valid on true dependency edges.
 
@@ -178,13 +178,13 @@ ALL three must hold: (1) **read-only or disjoint writes** (no conflict zone); (2
 
 ### Concurrent Invocation Handling
 
-Two top-level pipelines running at once (e.g. `hatch3r-workflow` in one shell, `hatch3r-board-pickup` in another) are bounded by the same three conditions, applied cross-pipeline (D7-SA7.5-F7.5.6): (1) **Advisory lock-note (best-effort, NOT atomic — D7-27)** — before Phase 1, the orchestrator writes/reads `.hatch3r/.lock` (JSON: `pid`, `command`, `branch`, `correlation_id`, `started_at`); clear on completion/abort; treat a note older than 6h as stale. This is an advisory coordination note, not a mutual-exclusion primitive: there is no `hatch3r` lock verb and no atomic acquire in `src/` (the only cross-process lock that ships is `acquireWriteLock` in `src/merge/safeWrite.ts`, scoped to single-file atomic writes, not top-level pipelines), so the LLM orchestrator's read-then-write is TOCTOU by construction and `src/cli/commands/status.ts` already labels this file "advisory". It exists to surface a likely collision, never to guarantee exclusion. (2) **Detect-then-warn** — if a live note names the same branch or an open `.hatch3r/hatch.json` board transaction, WARN and ASK (proceed on a separate branch / wait / abort); never silently co-mutate shared state. (3) **True isolation via worktree, not the lock-note (D7-27)** — when concurrency must actually be conflict-free rather than merely warned (the parallel-implementer path), route each pipeline through `hatch3r worktree-setup <name>` — the isolation primitive hatch3r already ships (`src/cli/commands/worktreeSetup.ts`; `commands/board/pickup-delegation-multi.md` already uses it per implementer) — so the pipelines write disjoint working trees and integrate back on completion, satisfying the disjoint-writes safety condition without relying on the non-atomic note. (4) **Cache-sharing** — `.hatch3r/learnings/` is read-many, write-once-at-completion with timestamp-ordered conflict resolution. Cross-task context sharing is bounded by the no-shared-mutable-state condition: learnings consolidate at pipeline completion; mid-pipeline writes are out of scope to preserve parallel-safety determinism (D7-SA7.5-F7.5.8). Each command's Guardrails cite this subsection.
+Two top-level pipelines running at once (e.g. `hatch3r-workflow` in one shell, `hatch3r-board-pickup` in another) are bounded by the same three conditions, applied cross-pipeline (D7-SA7.5-F7.5.6): (1) **Advisory lock-note (best-effort, NOT atomic — D7-27)** — before Phase 1, the orchestrator writes/reads `.hatch3r/.lock` (JSON: `pid`, `command`, `branch`, `correlation_id`, `started_at`); clear on completion/abort; treat a note older than 6h as stale. This is an advisory coordination note, not a mutual-exclusion primitive: there is no `hatch3r` lock verb and no atomic acquire in `src/` (the only cross-process lock that ships is `acquireWriteLock` in `src/merge/safeWrite.ts`, scoped to single-file atomic writes, not top-level pipelines), so the LLM orchestrator's read-then-write is TOCTOU by construction and `src/cli/commands/status.ts` already labels this file "advisory". It exists to surface a likely collision, never to guarantee exclusion. (2) **Detect-then-warn** — if a live note names the same branch or an open `.hatch3r/hatch.json` board transaction, WARN and ASK (proceed on a separate branch / wait / abort); never silently co-mutate shared state. (3) **True isolation via worktree, not the lock-note (D7-27)** — when concurrency must actually be conflict-free rather than merely warned (the parallel-implementer path), route each pipeline through `hatch3r worktree-setup <name>` — the isolation primitive hatch3r already ships (`src/cli/commands/worktreeSetup.ts`; `commands/board/pickup-delegation-multi.md` already uses it per implementer) — so the pipelines write disjoint working trees and integrate back on completion, satisfying the disjoint-writes safety condition without relying on the non-atomic note. Worktree isolation guarantees FILE-level disjointness only — it cannot detect two lanes mutating the same shared contract (a persisted collection name, an exported store symbol, a wire field) from different files; contract disjointness is established before dispatch and re-checked at lane exit per `rules/hatch3r-contract-census.md`. (4) **Cache-sharing** — `.hatch3r/learnings/` is read-many, write-once-at-completion with timestamp-ordered conflict resolution. Cross-task context sharing is bounded by the no-shared-mutable-state condition: learnings consolidate at pipeline completion; mid-pipeline writes are out of scope to preserve parallel-safety determinism (D7-SA7.5-F7.5.8). Each command's Guardrails cite this subsection.
 
 ## Cross-Phase Error Propagation
 
-On a non-SUCCESS status, the orchestrator MUST propagate error context downstream, never silently drop it. **Phase 1 PARTIAL:** include `researchGaps` in the implementer prompt and set confidence expectations accordingly. **Phase 2 PARTIAL:** include `reason` + unimplemented acceptance criteria in the reviewer prompt (reviewer distinguishes "not done yet" from "done incorrectly"). **Phase 3 UNRESOLVED:** include the unresolved findings in Phase 4 specialist prompts (specialists must not conflict with known issues). **Phase 4 specialist FAILED:** include the failure reason when surfacing — never report "Phase 4 failed" without naming which specialist and why.
+On a non-SUCCESS status, the orchestrator MUST propagate error context downstream, never silently drop it. **Phase 1 PARTIAL:** include `researchGaps` in the implementer prompt and set confidence expectations accordingly. **Phase 2 PARTIAL:** include `reason` + unimplemented acceptance criteria in the reviewer prompt (reviewer distinguishes "not done yet" from "done incorrectly"). **Phase 3 UNRESOLVED:** include the unresolved findings in Phase 4 specialist prompts (specialists must not conflict with known issues). Unresolved findings persist as ledger rows (`rules/hatch3r-findings-ledger.md`); Phase 4 prompts carry the open `finding_id` list. **Phase 4 specialist FAILED:** include the failure reason when surfacing — never report "Phase 4 failed" without naming which specialist and why.
 
-**Sub-agent-failure handling (shared clause; all commands cite this — never inline).** When any spawned implementer/fixer/specialist sub-agent FAILS or returns no usable output: (1) retry once with the same prompt; (2) if the retry fails, re-spawn `hatch3r-fixer` (Phase 3) with the failure reason + partial output as failure context — `hatch3r-fixer` is the code-mutation path, so the work stays delegated; (3) if the re-spawn also fails, emit `BLOCKED_OTHER` with a one-sentence reason and ASK the user (fix-manually vs adjust-scope vs accept-risk). The orchestrator MUST NOT fall back to inline implementation — that is the issue #73 bypass mode (see Mandatory Delegation Directive). The sole exception is `hatch3r-quick-change`, whose Tier-1 carve-out permits inline retry per its declared scope.
+**Sub-agent-failure handling (shared clause; all commands cite this — never inline).** When any spawned implementer/fixer/specialist sub-agent FAILS or returns no usable output: (1) retry once with the same prompt; (2) if the retry fails, re-spawn `hatch3r-fixer` (Phase 3) with the failure reason + partial output as failure context — `hatch3r-fixer` is the code-mutation path, so the work stays delegated; (3) if the re-spawn also fails, emit `BLOCKED_OTHER` with a one-sentence reason and ASK the user (fix-manually vs adjust-scope vs accept-risk). Flush in-flight ledger rows per W4 — before the retry and again before the BLOCKED_OTHER ASK (`rules/hatch3r-findings-ledger.md`). The orchestrator MUST NOT fall back to inline implementation — that is the issue #73 bypass mode (see Mandatory Delegation Directive). The sole exception is `hatch3r-quick-change`, whose Tier-1 carve-out permits inline retry per its declared scope.
 
 ## Correlation ID
 
@@ -242,4 +242,4 @@ All `scope: always` rules apply to every task including sub-agent work; include
 
 - **Tier 1 — always include (every sub-agent):** `hatch3r-security-patterns`, `hatch3r-code-standards`.
 - **Tier 2 — by phase:** `hatch3r-testing` (testability/implementer/reviewer); `hatch3r-accessibility-standards` (ui, UI reviewer); `hatch3r-git-conventions` (orchestrator git ops); `hatch3r-ci-cd` (ci-watcher/devops); `hatch3r-dependency-management` (security CQ3 supply-chain slice).
-- **Tier 3 — on-demand by role + scope:** `hatch3r-api-design`, `hatch3r-secrets-management`, `hatch3r-data-classification`, `hatch3r-performance-budgets`, `hatch3r-browser-verification`, `hatch3r-component-conventions`, `hatch3r-i18n`, `hatch3r-theming`, `hatch3r-migrations`, `hatch3r-feature-flags`, `hatch3r-observability-logging`, `hatch3r-observability-metrics`, `hatch3r-observability-tracing`.
+- **Tier 3 — on-demand by role + scope:** `hatch3r-api-design`, `hatch3r-secrets-management`, `hatch3r-data-classification`, `hatch3r-performance-budgets`, `hatch3r-browser-verification`, `hatch3r-component-conventions`, `hatch3r-i18n`, `hatch3r-theming`, `hatch3r-migrations`, `hatch3r-feature-flags`, `hatch3r-observability-logging`, `hatch3r-observability-metrics`, `hatch3r-observability-tracing`, `hatch3r-dynamic-stack-verification`.

package/dist/content/rules/hatch3r-anti-duplication.md

@@ -37,6 +37,15 @@ After every implementation:
 
 Tools accepted as equivalent: `jscpd` (Node), `PMD CPD` (Java/multi-lang), `simian` (multi-lang), `sonar-scanner` duplication module. Pick by toolchain; the threshold below is tool-agnostic.
 
+## Value-Drift Census (Shared Constants)
+
+A rate, default, threshold, or enum value consumed by ≥2 features has exactly one owning module; every other reader imports it. On touching such a constant, run the census: grep the constant NAME and the literal VALUE repo-wide; every independently-defined copy is a drift candidate.
+
+- **Damage ranking — silent-wrong beats loud-broken.** A deleted import fails at build; a stale copy computes wrong values in production with no signal.
+- **Remedy per copy:** import from the owner, or document intentional divergence with an inline ADR comment.
+- **Outside the jscpd scan's reach:** one-line literals sit below clone-scan thresholds (≥30-line blocks) — the census is name+value grep, not block matching.
+- **Taxonomy owner:** `rules/hatch3r-contract-census.md` → Value-Drift Census; this section is the anti-duplication-side enforcement hook.
+
 ## Tunable Thresholds Per Maturity Tier
 
 Per `hatch3r config maturity=<tier>`:
@@ -69,7 +78,7 @@ discovery_scan:
 
 `overlap_assessment: high` without `decision: extend-existing | refactor-shared | adr-distinct` is a P4 violation — author is creating a duplicate.
 
-At the user surface, the Iteration Summary carries the `Duplication:` exception line only on a non-clean or skipped scan — `Duplication: <n> match(es), closest <path>, overlap <none|partial|high>` when matches were found, or `Duplication: scan skipped (<reason>)` when skipped; a clean scan emits no line.
+At the user surface, the Iteration Summary carries the `Duplication:` exception line only on a non-clean or skipped scan — `Duplication: <n> match(es), closest <path>, overlap <none|partial|high>` when matches were found, or `Duplication: scan skipped (<reason>)` when skipped; a clean scan emits no line. Value-drift census hits reuse the same channel and silent-failure contract with the variant `Duplication: value-drift — <n> independent definition(s) of <constant>, owner <path>`; a clean census emits no line.
 
 ## Worked Example
 

package/dist/content/rules/hatch3r-anti-duplication.mdc

@@ -37,6 +37,15 @@ After every implementation:
 
 Tools accepted as equivalent: `jscpd` (Node), `PMD CPD` (Java/multi-lang), `simian` (multi-lang), `sonar-scanner` duplication module. Pick by toolchain; the threshold below is tool-agnostic.
 
+## Value-Drift Census (Shared Constants)
+
+A rate, default, threshold, or enum value consumed by ≥2 features has exactly one owning module; every other reader imports it. On touching such a constant, run the census: grep the constant NAME and the literal VALUE repo-wide; every independently-defined copy is a drift candidate.
+
+- **Damage ranking — silent-wrong beats loud-broken.** A deleted import fails at build; a stale copy computes wrong values in production with no signal.
+- **Remedy per copy:** import from the owner, or document intentional divergence with an inline ADR comment.
+- **Outside the jscpd scan's reach:** one-line literals sit below clone-scan thresholds (≥30-line blocks) — the census is name+value grep, not block matching.
+- **Taxonomy owner:** `rules/hatch3r-contract-census.md` → Value-Drift Census; this section is the anti-duplication-side enforcement hook.
+
 ## Tunable Thresholds Per Maturity Tier
 
 Per `hatch3r config maturity=<tier>`:
@@ -69,7 +78,7 @@ discovery_scan:
 
 `overlap_assessment: high` without `decision: extend-existing | refactor-shared | adr-distinct` is a P4 violation — author is creating a duplicate.
 
-At the user surface, the Iteration Summary carries the `Duplication:` exception line only on a non-clean or skipped scan — `Duplication: <n> match(es), closest <path>, overlap <none|partial|high>` when matches were found, or `Duplication: scan skipped (<reason>)` when skipped; a clean scan emits no line.
+At the user surface, the Iteration Summary carries the `Duplication:` exception line only on a non-clean or skipped scan — `Duplication: <n> match(es), closest <path>, overlap <none|partial|high>` when matches were found, or `Duplication: scan skipped (<reason>)` when skipped; a clean scan emits no line. Value-drift census hits reuse the same channel and silent-failure contract with the variant `Duplication: value-drift — <n> independent definition(s) of <constant>, owner <path>`; a clean census emits no line.
 
 ## Worked Example
 

package/dist/content/rules/hatch3r-clarification-default.md

@@ -1,7 +1,7 @@
 ---
 id: hatch3r-clarification-default
 type: rule
-description: "P8 B1 floor: every hatch3r-invoked agentic workflow detects and resolves ambiguity via the platform-native question tool BEFORE executing — default behavior, not exception-driven. Names the 4-trigger set and mandates a §0 ambiguity gate on every mutating agent, command, and skill."
+description: "P8 B1 floor: every hatch3r-invoked agentic workflow detects and resolves ambiguity via the platform-native question tool BEFORE executing — default behavior, not exception-driven. Names the 5-trigger set and mandates a §0 ambiguity gate on every mutating agent, command, and skill."
 tags: [orchestration, floor:protocol]
 scope: always
 precedence: high
@@ -20,7 +20,7 @@ Canonical reference for the *how* of asking: `agents/shared/user-question-protoc
 
 Default-path, not exception: a workflow that proceeds without resolving a live trigger below has violated B1, even if it later succeeds. Asking is the baseline; silent assumption is the deviation that must be justified.
 
-## Four-trigger set
+## Five-trigger set
 
 Apply the protocol before any write-tool invocation when ANY of these hold:
 
@@ -28,14 +28,15 @@ Apply the protocol before any write-tool invocation when ANY of these hold:
 2. **Multiple valid interpretations** — two or more viable approaches with materially different cost, scope, or risk.
 3. **Irreversible action** — deleting an artifact, renaming a public artifact id, dropping a frontmatter field, force-pushing a branch.
 4. **Missing acceptance criteria** — no testable definition of done for the requested change.
+5. **Unattested product decision** — the change destroys or transforms user data, or alters user-visible behavior beyond stated acceptance criteria, with no user statement authorizing it; an agent-authored comment or PR sentence is self-certification, not authorization.
 
-If none of the four hold and the safer default is obvious and reversible, proceed and note the default — do not manufacture a question (anti-pattern per `agents/shared/user-question-protocol.md` "Echo-as-question").
+If none of the five hold and the safer default is obvious and reversible, proceed and note the default — do not manufacture a question (anti-pattern per `agents/shared/user-question-protocol.md` "Echo-as-question").
 
 ## §0 ambiguity gate (every mutating artifact)
 
 Every artifact under `agents/`, `commands/`, and `skills/` that can mutate files MUST carry a §0 (or "Step 0 — Ambiguity gate") block as its first procedural step. The block:
 
-- scans the request against the four-trigger set above before any write;
+- scans the request against the five-trigger set above before any write;
 - on a live trigger, asks via the platform-native question tool per `agents/shared/user-question-protocol.md` and awaits the answer before proceeding;
 - declares the default-if-no-response option so the workflow never deadlocks;
 - on a non-response, wires the central deadlock-break: apply the declared default AND log it in the Iteration Summary via the `Default applied:` exception line (orchestrator path), OR return Status `BLOCKED_AMBIGUITY` when no default line was emitted (sub-agent / authoring-bug path) — never silent-pick, per `agents/shared/user-question-protocol.md` → Operationalising Default-if-no-Response.
@@ -54,7 +55,7 @@ Binds every hatch3r-invoked workflow that mutates artifacts in the end-user repo
 
 ## Confidence-floor calibration (D13-SA13.3-F13.3.3)
 
-The four-trigger set above is the floor for *scope/intent* ambiguity. Orthogonally, the `--confidence-floor=any|medium|high` flag (and the persisted `hatch3r config confidence_floor=...` default) calibrates a *result-confidence* ASK surface in the core orchestrators (`hatch3r-workflow`, `hatch3r-board-pickup`, `hatch3r-quick-change`, `hatch3r-revision`). At floor `high`, the orchestrator ASKs the user on every low-confidence finding regardless of severity — an additional, user-selected ASK trigger layered on top of the always-on four-trigger set. The floor never relaxes the four triggers; it only adds ASK pressure on uncertain results. Per P1 maturity tier (Decision 16): solo defaults `any`, enterprise defaults `high`.
+The five-trigger set above is the floor for *scope/intent* ambiguity. Orthogonally, the `--confidence-floor=any|medium|high` flag (and the persisted `hatch3r config confidence_floor=...` default) calibrates a *result-confidence* ASK surface in the core orchestrators (`hatch3r-workflow`, `hatch3r-board-pickup`, `hatch3r-quick-change`, `hatch3r-revision`). At floor `high`, the orchestrator ASKs the user on every low-confidence finding regardless of severity — an additional, user-selected ASK trigger layered on top of the always-on five-trigger set. The floor never relaxes the five triggers; it only adds ASK pressure on uncertain results. Per P1 maturity tier (Decision 16): solo defaults `any`, enterprise defaults `high`.
 
 ## Exemptions (D5-M5)
 

package/dist/content/rules/hatch3r-clarification-default.mdc

@@ -1,7 +1,7 @@
 ---
 id: hatch3r-clarification-default
 type: rule
-description: "P8 B1 floor: every hatch3r-invoked agentic workflow detects and resolves ambiguity via the platform-native question tool BEFORE executing — default behavior, not exception-driven. Names the 4-trigger set and mandates a §0 ambiguity gate on every mutating agent, command, and skill."
+description: "P8 B1 floor: every hatch3r-invoked agentic workflow detects and resolves ambiguity via the platform-native question tool BEFORE executing — default behavior, not exception-driven. Names the 5-trigger set and mandates a §0 ambiguity gate on every mutating agent, command, and skill."
 tags: [orchestration, floor:protocol]
 alwaysApply: true
 precedence: high
@@ -20,7 +20,7 @@ Canonical reference for the *how* of asking: `agents/shared/user-question-protoc
 
 Default-path, not exception: a workflow that proceeds without resolving a live trigger below has violated B1, even if it later succeeds. Asking is the baseline; silent assumption is the deviation that must be justified.
 
-## Four-trigger set
+## Five-trigger set
 
 Apply the protocol before any write-tool invocation when ANY of these hold:
 
@@ -28,14 +28,15 @@ Apply the protocol before any write-tool invocation when ANY of these hold:
 2. **Multiple valid interpretations** — two or more viable approaches with materially different cost, scope, or risk.
 3. **Irreversible action** — deleting an artifact, renaming a public artifact id, dropping a frontmatter field, force-pushing a branch.
 4. **Missing acceptance criteria** — no testable definition of done for the requested change.
+5. **Unattested product decision** — the change destroys or transforms user data, or alters user-visible behavior beyond stated acceptance criteria, with no user statement authorizing it; an agent-authored comment or PR sentence is self-certification, not authorization.
 
-If none of the four hold and the safer default is obvious and reversible, proceed and note the default — do not manufacture a question (anti-pattern per `agents/shared/user-question-protocol.md` "Echo-as-question").
+If none of the five hold and the safer default is obvious and reversible, proceed and note the default — do not manufacture a question (anti-pattern per `agents/shared/user-question-protocol.md` "Echo-as-question").
 
 ## §0 ambiguity gate (every mutating artifact)
 
 Every artifact under `agents/`, `commands/`, and `skills/` that can mutate files MUST carry a §0 (or "Step 0 — Ambiguity gate") block as its first procedural step. The block:
 
-- scans the request against the four-trigger set above before any write;
+- scans the request against the five-trigger set above before any write;
 - on a live trigger, asks via the platform-native question tool per `agents/shared/user-question-protocol.md` and awaits the answer before proceeding;
 - declares the default-if-no-response option so the workflow never deadlocks;
 - on a non-response, wires the central deadlock-break: apply the declared default AND log it in the Iteration Summary via the `Default applied:` exception line (orchestrator path), OR return Status `BLOCKED_AMBIGUITY` when no default line was emitted (sub-agent / authoring-bug path) — never silent-pick, per `agents/shared/user-question-protocol.md` → Operationalising Default-if-no-Response.
@@ -54,7 +55,7 @@ Binds every hatch3r-invoked workflow that mutates artifacts in the end-user repo
 
 ## Confidence-floor calibration (D13-SA13.3-F13.3.3)
 
-The four-trigger set above is the floor for *scope/intent* ambiguity. Orthogonally, the `--confidence-floor=any|medium|high` flag (and the persisted `hatch3r config confidence_floor=...` default) calibrates a *result-confidence* ASK surface in the core orchestrators (`hatch3r-workflow`, `hatch3r-board-pickup`, `hatch3r-quick-change`, `hatch3r-revision`). At floor `high`, the orchestrator ASKs the user on every low-confidence finding regardless of severity — an additional, user-selected ASK trigger layered on top of the always-on four-trigger set. The floor never relaxes the four triggers; it only adds ASK pressure on uncertain results. Per P1 maturity tier (Decision 16): solo defaults `any`, enterprise defaults `high`.
+The five-trigger set above is the floor for *scope/intent* ambiguity. Orthogonally, the `--confidence-floor=any|medium|high` flag (and the persisted `hatch3r config confidence_floor=...` default) calibrates a *result-confidence* ASK surface in the core orchestrators (`hatch3r-workflow`, `hatch3r-board-pickup`, `hatch3r-quick-change`, `hatch3r-revision`). At floor `high`, the orchestrator ASKs the user on every low-confidence finding regardless of severity — an additional, user-selected ASK trigger layered on top of the always-on five-trigger set. The floor never relaxes the five triggers; it only adds ASK pressure on uncertain results. Per P1 maturity tier (Decision 16): solo defaults `any`, enterprise defaults `high`.
 
 ## Exemptions (D5-M5)
 

package/dist/content/rules/hatch3r-code-standards.md

@@ -93,7 +93,7 @@ When working in a monorepo (multiple packages or apps in a single repository):
 
 - Remove unused imports, variables, functions, and type definitions immediately. Do not comment them out "for later."
 - Use the compiler's unused-symbol diagnostics (e.g., TypeScript `noUnusedLocals`/`noUnusedParameters`, Go `go vet`, Rust `dead_code`) and the linter (`no-unused-vars` or equivalent) to catch dead code automatically.
-- After removing a feature or completing a refactor, search for all references to the removed code. Delete orphaned tests, fixtures, and documentation.
+- After removing a feature or completing a refactor, search for all references to the removed code. Delete orphaned tests, fixtures, and documentation. Exception — façade contract-hold: while a dropped or renamed shared-contract field is inside its compatibility window (`rules/hatch3r-contract-census.md` → Façade Contract-Hold), the nulled field behind the façade and its guarded consumer reads are NOT dead code; run this reference sweep at the contract phase, when the held field is deleted.
 - Feature-flagged code that has been fully rolled out (flag removed) must have the flag-off branch deleted in the same PR.
 - Commented-out code is never acceptable in committed code. Use version control history to retrieve old implementations.
 

package/dist/content/rules/hatch3r-code-standards.mdc

@@ -88,7 +88,7 @@ When working in a monorepo (multiple packages or apps in a single repository):
 
 - Remove unused imports, variables, functions, and type definitions immediately. Do not comment them out "for later."
 - Use the compiler's unused-symbol diagnostics (e.g., TypeScript `noUnusedLocals`/`noUnusedParameters`, Go `go vet`, Rust `dead_code`) and the linter (`no-unused-vars` or equivalent) to catch dead code automatically.
-- After removing a feature or completing a refactor, search for all references to the removed code. Delete orphaned tests, fixtures, and documentation.
+- After removing a feature or completing a refactor, search for all references to the removed code. Delete orphaned tests, fixtures, and documentation. Exception — façade contract-hold: while a dropped or renamed shared-contract field is inside its compatibility window (`rules/hatch3r-contract-census.md` → Façade Contract-Hold), the nulled field behind the façade and its guarded consumer reads are NOT dead code; run this reference sweep at the contract phase, when the held field is deleted.
 - Feature-flagged code that has been fully rolled out (flag removed) must have the flag-off branch deleted in the same PR.
 - Commented-out code is never acceptable in committed code. Use version control history to retrieve old implementations.
 

package/dist/content/rules/hatch3r-contract-census.md

@@ -0,0 +1,160 @@
+---
+id: hatch3r-contract-census
+type: rule
+description: Shared-contract discipline for brownfield changes — contract taxonomy, repo-wide consumer census as lane-exit gate, seam-owner protocol for parallel lanes, façade contract-hold on field drop/rename, value-drift census for shared constants.
+tags: [implementation, review, orchestration, ctx:brownfield-only]
+precedence: high
+scope: always
+---
+# hatch3r Contract Census
+
+**Pillars:** P2 (Scientific & Practical Quality), P8 (Clarification & Fan-out Discipline), CQ8 (Maintainability Quality)
+
+## Binding Condition
+
+This rule binds any diff that mutates a contract in the Shared-Contract Taxonomy below, in a repo where that contract has existing consumers. A change touching no taxonomy class exits with `Consumer census: N/A (no shared-contract change)`. A greenfield-new contract with zero consumers exits trivially — the census grep returns nothing and reports `clean`.
+
+The failure class this rule closes: parallel file-disjoint lanes silently breaking a shared runtime contract — a renamed persisted collection, a changed store symbol, a dropped wire field — that no quality gate and, in a no-static-types stack, no compiler checks. File disjointness is not contract disjointness: two lanes whose diffs share zero files still collide when both touch the same contract.
+
+## Shared-Contract Taxonomy
+
+Seven contract classes. Each line states why file-disjointness does not protect it.
+
+- **(a) Exported symbol** — function, class, or store: its name, signature, or subpath. Importers live in other files by definition. (researcher: `api_signature` for signature changes, `public_interface` for name/visibility/subpath changes)
+- **(b) Persisted collection/table/field name** — string-typed: writer and reader each spell the string independently; no compiler links them. (researcher: `data_migration`)
+- **(c) Client↔server wire field** — a JSON key crossing the network: serializer and deserializer sit in different trees, often different languages. (researcher: `type_shape`)
+- **(d) Event name + payload schema** — emitter and subscribers are decoupled by design; the bus hides the seam. (researcher: `event_schema`)
+- **(e) Shared constant** — rate, default, threshold, or enum value: copies compile independently, and a stale copy computes wrong values with no error. (researcher: `type_shape` for enum members, `cli_contract` for surfaced defaults; value-only drift sits below the researcher's radar → §Value-Drift Census)
+- **(f) CLI flag / config / env key** — the producer is code; the consumers are shell scripts, CI YAML, and dotfiles no type-checker loads. (researcher: `cli_contract`)
+- **(g) Type or schema shape where no checker is authoritative** — plain JS, dynamic access, JSON blobs: shape agreement exists only by convention, so classes (a)-(f) all degrade to grep.
+
+The parenthesized categories map classes (a)-(f) onto the researcher's breaking-change vocabulary — `api_signature / type_shape / event_schema / public_interface / data_migration / cli_contract` (`agents/hatch3r-researcher.md` → Full-Mode Breaking-Change Detection) — so a researcher Breaking Change Candidates row converts directly into a taxonomy class and back.
+
+## Consumer Census
+
+**Trigger:** any taxonomy mutation — rename, removal, re-signature, revalue, key add/drop.
+
+**Procedure:**
+
+1. Repo-wide grep of the OLD identifier AND the NEW identifier. Never limit the grep to the lane's `affectedFiles` — consumers live outside the lane's file set by definition.
+2. Per-class patterns:
+   - **String contracts** (classes b, d, f) — grep the quoted forms: `'name'` and `"name"` (plus template-literal and heredoc forms where the language has them).
+   - **Wire fields** (class c) — grep BOTH the client and server trees, plus fixtures, mocks, and recorded payloads: a test fixture still asserting the old key is an unreconciled consumer.
+   - **Constants** (class e) — grep the NAME and the literal VALUE; an independent copy defines the value without the name.
+   - **Infra/config** (class f) — include non-code files: CI YAML, Dockerfiles, `.env*`, deploy manifests.
+3. Narrowing tactics for short identifiers that over-match: qualified member access (`\.field\b`), collection-prefixed forms (`users.field`), and quoted-literal-only matches.
+
+**Reconciled** — a consumer counts as reconciled when it is updated to the new shape in this diff, OR reads through a guard added in this diff (façade-hold null-guard), OR is justified by name. Exactly three justifications are valid:
+
+1. Consumer owned by another lane's seam — name the lane/issue.
+2. Dead code with a linked deletion.
+3. Dynamic/reflective access grep cannot resolve — name the mechanism, add a runtime guard.
+
+**Output grammar** — the census field emits exactly one of:
+
+```
+Consumer census: clean | reconciled(N) | N unreconciled — justification | N/A (no shared-contract change)
+```
+
+- `clean` — the grep found zero consumers of the old shape outside this diff.
+- `reconciled(N)` — N consumers found; every one updated, guarded, or justified by name.
+- `N unreconciled — justification` — N consumers not yet updated or guarded; each carries one of the three justifications above. `unreconciled` WITHOUT a named justification caps the producing agent's Status at PARTIAL — done is declared from grep evidence, not assumed (the large-scale-change discipline of *Software Engineering at Google* ch. 22).
+- `N/A (no shared-contract change)` — the diff touches no taxonomy class.
+
+**Spec-time seed:** when a brownfield spec exists, its Integration-Surface consumer inventory (`agents/hatch3r-brownfield-spec.md` → Integration-Surface Analysis) seeds the grep list. The lane still re-runs the grep at lane exit: parallel lanes may have added consumers since spec time.
+
+**Where it runs:** `agents/hatch3r-implementer.md` Step 5d and `agents/hatch3r-fixer.md` Step 5b carry this census as a lane-exit gate — `Status: SUCCESS` requires census ∈ {`clean`, `reconciled(N)`, `N/A`}.
+
+## Consumer-Scoped Review
+
+The reviewer re-derives the census itself — a self-run grep of every changed contract's OLD and NEW identifier. The implementer's census field and the Phase-1 blast radius, when present, seed the list; neither substitutes for the reviewer's own grep.
+
+The reviewer then OPENS and READS each consumer at its use site, both sides of every seam:
+
+- serializer AND deserializer for a wire field;
+- exporter AND importers for a store symbol;
+- writer AND readers for a persisted name.
+
+The verdict cites the captured grep output per the reviewer's Grounding rule. Severity: a consumer left reading the old shape is **Critical**; an implementer census of `unreconciled` without a named justification is **Critical**; a taxonomy diff with no census field at all is a **Warning** (protocol violation). Mirrors `agents/hatch3r-reviewer.md` item 11 (contract preservation, consumer-scoped, two-lens).
+
+## Seam-Owner Protocol
+
+A contract touched by ≥2 parallel lanes in the same batch gets exactly ONE owning lane: the lane whose acceptance criteria REQUIRE the mutation. The owner lands the emitter change AND all consumer reconciliation in one diff. Every peer lane's dispatch prompt gains the line:
+
+```
+Seam constraint: contract <X> is owned by issue #<N> this batch — consume the current shape; do not mutate it
+```
+
+Two lanes that BOTH require mutating the same contract are re-leveled sequential — the later lane consumes the owner's landed shape.
+
+**Detection — two passes:**
+
+1. First-pass textual scan of issue bodies and linked specs at batch triage: endpoint paths, collection/table names, event names, exported symbols, shared constants named in ≥2 issues (`commands/hatch3r-board-pickup.md` Step 3 item 5).
+2. Authoritative cross-check before dispatch: union the per-issue researcher Breaking Change Candidates tables; any contract appearing in ≥2 issues of the same level is a collision (`commands/board/pickup-delegation-multi.md` Step 6c.2 item 4).
+
+**Violation surface:** a post-batch semantic conflict on an owned contract names the violating peer lane. Reconciliation routes to the seam owner — one diff owns the contract; the conflict is never patched in the merge.
+
+**Normative amendment to parallel safety:** the parallel-safety condition "disjoint writes" holds at file AND contract granularity (`rules/hatch3r-agent-orchestration.md` → parallel-safety conditions point here). Two lanes are NOT disjoint when their file-disjoint diffs mutate the same shared contract.
+
+## Façade Contract-Hold
+
+Never delete a field out from under consumers. On drop or rename of a shared output field:
+
+1. **Hold the key-set** — the façade keeps emitting every key it emitted before, through the compatibility window.
+2. **Hard-null the dropped field** — key present, value `null`. Absence is a silent shape change a consumer misreads without error; an explicit null fails loudly at the value, at the consumer, on the first read.
+3. **Reconcile every consumer to guarded reads** — null-tolerant access, census-tracked: each guarded read counts as reconciled.
+4. **Delete the key only at the contract phase** — when the census shows zero unguarded readers. This is Fowler's ParallelChange termination criterion: contract begins only "once all usages have been migrated"; skipping contract leaves you worse than you started (martinfowler.com/bliki/ParallelChange.html, accessed 2026-07-08).
+
+**Renames** are a drop plus an add: the new key is added, the old key is held nulled or aliased through the window, and consumers migrate key-by-key under the census.
+
+**Lane-exit question, answered verbatim before returning:**
+
+**"Did you delete the field, or null it behind the façade?"**
+
+"Deleted" during the compatibility window is a gate failure — revert to the hold pattern before returning.
+
+**Precedence carve-out:** a held nulled field and its guarded reads are NOT dead code until the contract phase. The `rules/hatch3r-code-standards.md` dead-code reference sweep runs at the contract phase, when the held field is deleted — that rule carries the matching exception.
+
+**Boundary:** this section is the in-code analog of expand-contract for the database (`rules/hatch3r-migrations.md`) and of the API field-removal lifecycle for public endpoints (`rules/hatch3r-api-versioning.md`). Those rules own their boundaries; this section owns every contract they do not load for — internal wire fields, store symbols, event payloads inside the app. Through the window the system builds and runs at all times, the BranchByAbstraction invariant.
+
+## Value-Drift Census
+
+A shared constant — rate, default, threshold, enum value — consumed by ≥2 features has exactly one owning module; every reader imports it. On touching such a constant:
+
+1. Grep the constant NAME and the literal VALUE repo-wide.
+2. Each independently-defined copy is a drift candidate.
+3. Remedy per copy: import from the owner, or an inline ADR comment documenting intentional divergence.
+
+**Damage ranking — silent-wrong beats loud-broken.** A deleted import fails at build; a stale copy computes wrong values in production with no signal.
+
+One-line literals sit below clone-scan thresholds (≥30-line block matching) — this census is name+value grep, not block matching. `rules/hatch3r-anti-duplication.md` → Value-Drift Census carries the anti-duplication-side enforcement hook.
+
+## Ownership Boundaries
+
+| Surface | Owning artifact |
+|---------|-----------------|
+| Database schema (expand/migrate/contract, online DDL, backfill) | `rules/hatch3r-migrations.md` |
+| Public API lifecycle (versioning, deprecation, `Sunset`) | `rules/hatch3r-api-versioning.md` |
+| Event registry compatibility (BACKWARD/FORWARD/FULL) | `rules/hatch3r-event-schema-evolution.md` |
+| Spec-time consumer inventory | `agents/hatch3r-brownfield-spec.md` → Integration-Surface Analysis |
+| Code clones / block duplication | `rules/hatch3r-anti-duplication.md` |
+
+This rule owns: internal cross-lane seams, the census output grammar, seam-owner assignment, the façade contract-hold, and value drift on shared constants.
+
+## Enforcement Wiring
+
+- `agents/hatch3r-implementer.md` Step 5d — consumer census as lane-exit gate.
+- `agents/hatch3r-fixer.md` Step 5b — consumer census on shared-contract fixes.
+- `agents/hatch3r-reviewer.md` item 11 — consumer-scoped review with self-run grep.
+- `commands/hatch3r-board-pickup.md` Step 3 item 5 — first-pass contract-overlap scan at batch triage.
+- `commands/board/pickup-delegation-multi.md` Step 6c.2 item 4 — seam-owner assignment; Step 6c.4 — per-lane census verification + cross-lane old-identifier grep.
+- `agents/shared/quality-charter.md` → Data integrity quality — shared-contract census row + verification gate.
+- `checks/code-quality.md` → Shared Contracts and Constants.
+
+## References
+
+- Fowler, M. "ParallelChange." martinfowler.com/bliki/ParallelChange.html (accessed 2026-07-08, T3 canonical, 2014). Expand/migrate/contract for any interface change; contract begins only "once all usages have been migrated" — skipping contract leaves you worse than you started.
+- Fowler, M. "BranchByAbstraction." martinfowler.com/bliki/BranchByAbstraction.html (accessed 2026-07-08, T3 canonical, 2014). The system builds and runs at all times during the hold; the abstraction layer outlives the old supplier.
+- Winters, T., Manshreck, T., Wright, H. *Software Engineering at Google*, ch. 22 "Large-Scale Changes." abseil.io/resources/swe-book/html/ch22.html (accessed 2026-07-08, T2, 2020). Tool-driven consumer discovery sharded by ownership; reintroduction guards; done is declared from evidence, not assumed.
+- Sourcegraph. "Batch Changes." sourcegraph.com/docs/batch-changes (accessed 2026-07-08, T1, current). Text-search census as the consumer-discovery primitive; burndown-to-zero completion tracking.
+- Pact. "can-i-deploy." docs.pact.io/pact_broker/can_i_deploy (accessed 2026-07-08, T1, current). Verification is recorded evidence per consumer; all known consumers gate the deploy.