DIVERGE-TIER-01 Adoption — Divergence Cost Tiers (2026-07-07)

New §11.8 "Divergence cost tiers (DEFAULT, DIVERGE-TIER-01)" added to

skills/dev-pabcd/references/loop-engineering.md (full text) with a summary

bullet in the skills/dev-pabcd/SKILL.md §11 router. Router-resident §11.6/§11.7

got stub headers in the reference file so the "canonical full text" claim stays

honest.

Owner interview (codexclaw, 2026-07-07): with near-unlimited subagents the old

implicit assumption "subagents are expensive, dispatch sparingly" breaks, but the

binding constraint becomes wall-clock + collapse-owner triage attention. The

owner rejected a 3-round agent-roundtable design as cost-inflated; the adopted

shape is conceptual-first divergence with a rare, criteria-gated implementation

spike. Distilled from the Fable field-guide unknown-discovery practices

(conceptual prototypes, blindspot passes) without importing model-specific

behavior.

devlog/_plan/260707_harness_divergence_interview/000_interview_record.md,

010_plan.md, 011_audit_synthesis.md (5 reviewer findings accepted; the

Tier-1 gate was tightened to include per-candidate provenance so §11.8 never

relaxes §11.7).

Fable research: codexclaw devlog/_plan/260707_thariq_fable_youtube_plugin_research/

and 260707_codex_rs_native_tooling_research/200_fable_loop_adoption.md.

Same rule id, runtime-adapted wording, shipped in the same pass:

collapse owner, dispatch employees = explorers).

clause in the loop-spec header divergence bullet.

Emergence/Divergence layer (explorer subagents author Tier-1 docs; minds stay

interview-time contradiction lenses; front-matter lives in the candidate doc,

the divergence CLI records the archive row alongside).