refactor(acm): supreme dao-shu guidance — #10+#11 synthesis

[KorenKrita]

Summary

Supreme always-on ACM guidance: unconstrained synthesis of #10 (Grok) and #11 (Composer). Optimizes for both teachability and corrective action (~9607 chars).

From #11 (teaching)

Addition	Why
How judgment layers fit together	smells → moments → criteria → failure shapes wiring
Causal opening	Process detail accumulates naturally + ACM keeps that growth recoverable
Cold start remains hard	explicit pressure paragraph line
Boundary decides whether folding is valid	clearer fold-criteria framing
without rereading archived summaries	actionable post-rebase instruction

From #10 (execution)

Retained	Why
Corrective failure shapes	each shape names the next corrective lean
Rebase candidate compare line	boundary placement and cold start—not proximity
Split skill disclosure	rebase-inline pointer only; end pointer without ambiguous-base duplicate
Fold target: boundary-only	cold start scoped to rebase, not every fold
Explicit lightweight reply triple	no evidence lookup, no side effects, no multi-step reasoning

Unchanged good choices

• Single Recognizable moments table (no Judgment duplicate)
• 4 smells; NEXT stalls only in Failure shapes
• Fold/Rebase criteria (soft naming)
• Why-annotated examples + travel self-check
• Nudge ↔ CORE vocabulary chain

vs #10 / #11

	#10	#11	Supreme
CORE chars	~9088	~9304	~9607
Judgment layers integrator	—	✅	✅
Corrective failure shapes	✅	—	✅
Rebase candidate compare	✅	—	✅
Split skill disclosure	✅	—	✅
without rereading archived summaries	—	✅	✅

Recommend as the merge target for unconstrained always-on injection. For strict budget, still prefer #9.

Test plan

• bun run verify:acm
• Live: read layers once → smell → moment → criteria self-check → travel → failure-shape recovery

Supersedes

Closes the #10 vs #11 debate. Optional follow-up: cherry-pick judgment-layers paragraph into lean #9 if desired.

[KorenKrita]

Superseded by #12 (chosen merge target for ACM dao-shu guidance). Closing this alternative.