Date: 2026-08-29 Related: issue #57, merged PRs #48, #68, #98, #104, and #106
Complete issue #57 professionally while coordinating it with other parallel issue work. Work from current main, avoid competing ownership, preserve the repository’s merge-safety rules, and provide reproducible evidence.
The merged Phase-A controller already provided deterministic replicator-mutator updates, a non-zero exploration floor, explicit lineage-only benefit, carrying-capacity suppression, zero-action outcomes, a hard one-write budget, and a read-only workflow.
One acceptance gap remained. The controller rejected repeated lineage identities, but a repeated descendant artifact could be supplied under a different caller-chosen identity. That could multiply the reproduction numerator and policy fitness even though the lineage protocol defines the normalized descendant artifact as the deduplication key.
The reproduction calculation also counted a verified receipt whose parent existed but was not both matured and verified. That was inconsistent with the independent eligible-parent denominator.
Phase A now fails closed when two lineage receipts reference the same normalized descendant artifact, regardless of their lineage IDs. The reproduction numerator includes only unique verified descendant receipts whose parent is in the eligible matured-and-verified parent set.
A deterministic fixture matrix now covers:
Each case is evaluated twice and must produce the same result. All cases keep both activation and actuation disabled, so they emit no modeled public action.
This change remains an offline controller hardening pass. It does not call GitHub, create cohorts, modify issues, merge code, send notifications, change governance, or enable Phase B. The public-write invariant remains at most one modeled action per generation, and the checked-in fixtures remain shadow-only.
Live evidence at audit time showed that main protection is now enabled, but Bootstrap Cohort Observatory #109 still reported zero external participants and zero verified descendant PRs. Therefore the real-descendant activation gate remains unsatisfied. This audit does not reinterpret owner activity or raw repository activity as descendant evidence.
Artifact-keyed deduplication makes the controller easier for reviewers to reason about and prevents repeated metadata from looking like community growth. The three named fixtures give future contributors small, reproducible examples of the intended controller regimes without requiring access to private context.
This bounded audit and implementation were prepared with Codex using the repository files and live public GitHub metadata. The deterministic Python tests and repository CI are the verification boundary; no claim of real descendant evidence was generated by the tool.