Arm Phase 10 (§13.32) + ai-laziness-traps project rule

[artyhoo]

Summary

Lands scoping for Phase 10 (project foundations audit & re-evaluation) and introduces a project-level rule formalising AI-laziness countermeasures. Documentation-only commit; no executable behavior changes.

What changed

• docs/meta-factory/open-questions.md:486-500 — §13.32 armed (Phase 10 umbrella). Six audit streams A1-A6 + 5-class scope-reduction discipline + AIF-chain risk mitigation. Full detail in gitignored local scaffold at .claude/orchestrator-prompts/phase-10-foundations-audit/kickoff.md (per Wave 9 kickoff precedent).
• .claude/rules/ai-laziness-traps.md — new project-level rule. 16 canonical traps T1-T16, including T11 (designing without prior-art), T12 (skipping literature sweep), T13 (ADOPTED zero-work assumption), T15 (self-application skipped), T16 (#pattern-matching-on-name). Self-defending mechanism §3 requires kickoff authors to cite + enumerate active T-numbers + add ≥1 domain-specific trap.

Stat: 2 files, +174/-0.

Why now

Wave 9 (§13.31, Project-wide discipline-theatre audit) is currently armed. During its scoping, the maintainer surfaced (2026-05-12) that the audit needs to go beyond behavioral compliance into foundational adequacy: every load-bearing decision in the project was made with internal reasoning + context7 only — external literature, academic research, related-project comparisons were never part of methodology. This is a project-wide gap from day one and likely the structural origin of #discipline-theatre.

Phase 10 is the response. This PR is the armed-trigger entry + supporting rule. Phase 10 fires AFTER Wave 9 closes AND maintainer commits to scope (including possibility of foundational refactor).

How to verify

• wc -l docs/meta-factory/open-questions.md — ≤500 lines (currently 500 exactly, at hook boundary).
• npm run test:principles — 56/56 pass (no changes to executable disciplines).
• Read .claude/rules/ai-laziness-traps.md §3 — verify self-defending mechanism is implementable (kickoff authors can cite + enumerate + extend).
• Read open-questions.md §13.32 — verify Phase 10 scope is coherent and triggers are observable.

§1.7 Forward-check applied

This PR introduces a new discipline-bearing rule (ai-laziness-traps.md) and arms a Phase 10 trigger. Forward-validation:

• .claude/rules/ai-laziness-traps.md:88-95 — §3 obligations on kickoff authors are operational, not aspirational: «cite + enumerate T-numbers + add ≥1 domain-specific trap». Three concrete actions, all verifiable at PR review.
• .claude/rules/ai-laziness-traps.md:97-102 — §4 anti-patterns include #trap-catalogue-blanket-reference — the failure mode the rule itself can produce («cite without enumerate»). Self-protection against the most obvious laziness path.
• docs/meta-factory/open-questions.md:496-498 — §13.32 trigger is observable (Wave 9 closes + maintainer commits to scope), not «when we feel like it». Resists #scope-not-formal-trigger anti-pattern.

§1.7 Backward-check applied

Backward-validates resistance to known failure modes in prior wave scoping:

• docs/meta-factory/open-questions.md:488-492 — §13.32 references existing SSOT entries docs: Phase 8 entry research (Step 0 capability matrix + GO verdict) #8, docs(research-patches): Wave 6 AI-doc cold audit (§13.26) #27-32 in prior-art-evaluations.md for the AIF re-evaluation stream (A4). Anchors new work in existing decision-record discipline; resists #own-stack-blind-spot.
• .claude/rules/ai-laziness-traps.md:11-14 — rule's origin block names the source (Wave 9 kickoff §6 inline → hoisted to project rule); preserves audit trail. Resists #decision-without-origin.
• .claude/rules/ai-laziness-traps.md:131-137 — §5 promotion/retirement triggers prevent catalogue bloat (#trap-list-grew-without-pruning anti-pattern). Without retirement policy, every wave would add 3-5 new traps and the list would become noise within 6 months.

Recursive self-application audit

This PR is itself a discipline-bearing artefact. Its claims (16 canonical traps, 5-class classification, ~120 decision points, ~35-40% scope reduction) have NOT been externally validated — they are the orchestrator's best honest estimate based on Wave 9 scoping context. Phase 10 R-phase, when it fires, MUST re-verify §13.32's classification accuracy and surface this PR's drift as a Wave 10.0 finding if §13.32 is wrong.

The rule applies to its own kickoff: anyone running Phase 10 must cite .claude/rules/ai-laziness-traps.md §2, enumerate active T-numbers (mandatory: T11, T12, T13, T15, T16), and add ≥1 domain-specific trap per §3 of the rule. If they don't — they fall into #trap-catalogue-blanket-reference and the orchestrator (or a reviewer) must reject the R-phase output.

§1.7 Bootstrap: this commit introduces ai-laziness-traps.md as a new discipline-bearing rule. The rule's own enforcement mechanism applies prospectively to Phase 10 and later kickoffs; this introducing commit pre-dates that enforcement and uses the Bootstrap escape-hatch pattern established in Wave 8.1 and §9 substance calibration window (active through 2026-06-10).

Follow-ups (NOT in this PR)

• Wave 8.5 (wave-8.5/retroactive-sweep-and-cleanup) — currently in flight in parallel worktree; closes §13.29.
• Wave 9 kickoff §6 will get a backport pointer to ai-laziness-traps.md in a separate small commit (Wave 9 kickoff is gitignored, so not in this PR).
• §13.30 (Wave 8 mutation-coverage residue) + §13.31 (Wave 9 umbrella) — added uncommitted in parallel session; will be committed separately by whoever owns those scopes, OR absorbed into the Wave 9 R-phase output.