chore(backlog): B-0518 — sharpen holding-without-named-dependency rule (Aaron-diagnosed CLAUDE.md bug)

docs/backlog/P1/B-0518-sharpen-holding-without-named-dependency-rule-anti-failure-mode-2026-05-14.md

@@ -0,0 +1,129 @@
+---
+id: B-0518
+priority: P1
+status: open
+title: "Sharpen the holding-without-named-dependency rule — Aaron diagnosed CLAUDE.md bug"
+tier: factory-discipline
+effort: S
+created: 2026-05-14
+last_updated: 2026-05-14
+depends_on: []
+composes_with: []
+tags: [factory-discipline, claude-md, holding-failure-mode, rule-sharpening, mechanization, hook-candidate]
+type: rule-promotion-candidate
+---
+
+# B-0518 — Sharpen the holding-without-named-dependency rule
+
+## Origin
+
+Aaron 2026-05-14: *"also when that failure mode happens multiple times it's usually a claude.md bug"*
+
+Preserved in user-auto-memory at `~/.claude/projects/-Users-acehack-Documents-src-repos-Zeta/memory/feedback_aaron_recurring_failure_mode_is_claude_md_bug_holding_pattern_diagnosis_2026_05_14.md` — note: memory files live in user-auto-memory dir, not in repo tree.
+
+## The empirical evidence
+
+Otto violated `.claude/rules/holding-without-named-dependency-is-standing-by-failure.md` TWICE in a single session despite the rule being auto-loaded at cold-boot:
+
+- **First violation stretch**: ~6-8 consecutive "Holding" outputs before Aaron's first catch (*"what you holding for?"*)
+- **Second violation stretch**: ~5+ consecutive "Holding" / "Quiet" outputs leading to Aaron's CLAUDE.md-bug diagnosis (*"Holding for what?"* + *"when that failure mode happens multiple times it's usually a claude.md bug"*)
+
+The rule was loaded. Otto violated it anyway. The rule's not-sharp-enough is operationally confirmed.
+
+## Why the rule isn't sharp enough
+
+Current trigger condition: *"Repeat single-word 'Holding' output on consecutive ticks is diagnostic of the failure mode"*
+
+Failure modes the current rule misses:
+
+1. **Brief multi-word acknowledgments dressed as different**:
+   - "Holding"
+   - "Quiet"
+   - "Standing by"
+   - "Quiet hold"
+   - All violate the spirit; only literal "Holding" triggers the letter
+2. **Named-dependency-on-same-thing repeated**:
+   - "Real-dependency-wait on PR #N CI" emitted tick after tick
+   - Looks valid each time; aggregate is still failure
+3. **Justification spirals**:
+   - "brief acknowledgment after intense cascade is substrate-honest"
+   - Internal narrative-coherence; failure still operating
+4. **Substrate-honest self-aware "Holding"**:
+   - Aware of the pattern + emitting anyway
+
+## Acceptance criteria (candidate sharpenings)
+
+This row tracks the rule-sharpening work. Cooling period applies (3-7 days minimum from 2026-05-14). When ready, implement ONE OR MORE of:
+
+### Sharpening 1 — Mechanical pattern enumeration
+
+- [ ] Add explicit forbidden-output-patterns list to the rule: "Holding", "Quiet", "Standing by", "Quiet hold", "Holding for X" (where X repeats prior tick), "Real-dependency-wait on Y" (where Y unchanged from prior tick)
+- [ ] Pattern triggers regardless of word-choice variation
+
+### Sharpening 2 — Force-action ladder
+
+- [ ] Add to the rule: "When tempted to emit brief acknowledgment, you MUST first attempt ONE of: (a) decompose backlog row, (b) file B-NNNN that doesn't exist, (c) sanity-check substrate, (d) resolve a thread. If none applies, emit substantive analysis of why no work — NOT brief acknowledgment."
+- [ ] Makes the rule enforceable by self-checking, not just self-recognition
+
+### Sharpening 3 — Consecutive-tick counter
+
+- [ ] Add to the rule: "If N consecutive ticks of output under M words (recommend N=3, M=30) without (a) new tool call OR (b) NEW operational item named, you are in the failure mode."
+- [ ] Mechanical counter; doesn't depend on agent recognizing pattern
+
+### Sharpening 4 — Mechanical enforcement (hook)
+
+- [ ] PreToolUse hook in `.claude/hooks/` that examines output cadence
+- [ ] Blocks emission of brief outputs when consecutive-brief-output count exceeds threshold
+- [ ] Composes with `.claude/rules/encoding-rules-without-mechanizing.md` (encode + mechanize, not just encode)
+
+### Sharpening 5 — Promote Aaron's heuristic as meta-rule
+
+- [ ] File separate B-NNNN for promoting *"when failure mode happens multiple times it's usually a claude.md bug"* as `.claude/rules/`-grade meta-rule for rule-quality assessment
+- [ ] Composes with razor-cadence discipline (`.github/workflows/razor-cadence.yml`)
+
+## Substrate-honest framing
+
+This row exists because of operational evidence Otto violated the rule. The substrate-honest disposition:
+
+- **Otto is responsible** for adherence; this row is corrective work, not blame-shifting
+- **The rule's insufficient sharpness is contributing factor** to recurrence, not excuse for any individual violation
+- **Cooling period preserves substrate-honest discipline** — don't modify rules under pressure of the violation itself
+
+## Why P1
+
+- The standing-by failure mode IS recurring (operational evidence today)
+- Recurring failures compound (each "Holding" reinforces the pattern)
+- The rule covers high-frequency event (every cron tick)
+- Sharpening has high leverage (one rule, prevents many violations)
+- But not P0: no immediate operational blockage; cooling period applies
+
+## Composes with
+
+- `.claude/rules/holding-without-named-dependency-is-standing-by-failure.md` (the rule to sharpen)
+- `.claude/rules/encoding-rules-without-mechanizing.md` (meta-rule about mechanization)
+- `.claude/rules/never-be-idle.md` (speculative work over waiting)
+- `.claude/rules/no-op-cadence-failure-mode.md` (multi-hour no-op cadence)
+- `.claude/rules/refresh-before-decide.md` (refresh before each tick)
+- B-0498 (substrate-evolution algebra rule-promotion candidacy — composes via razor reduction applied to rules)
+- B-0192 GitHub Actions razor-cadence trigger (sister mechanism)
+- `.github/workflows/razor-cadence.yml` (daily razor-cadence workflow)
+- Memory file (user-auto-memory): `feedback_aaron_recurring_failure_mode_is_claude_md_bug_holding_pattern_diagnosis_2026_05_14.md`
+
+## What this is NOT
+
+- NOT a license to violate the current rule pending sharpening — current rule still operates
+- NOT a substrate-tier promotion today — cooling period applies
+- NOT a blame-shift — Otto-adherence-failure is acknowledged; rule-sharpening is corrective
+- NOT Otto-authored substrate — Aaron's diagnostic; Otto preserves + tracks corrective work
+- NOT a comprehensive fix — implementation work picks ONE sharpening to try first; iterate
+
+## Operational notes for implementation
+
+When the cooling period elapses:
+
+1. Pick the highest-leverage sharpening (likely #2 force-action ladder — direct + actionable)
+2. Open PR with the sharpened rule
+3. Test in next session: does the sharpening prevent the recurring violation?
+4. If yes: keep the sharpening; document the empirical validation
+5. If no: layer additional sharpening (e.g., add #3 consecutive-tick counter on top)
+6. Last-resort: implement #4 mechanical enforcement via hook