[JoyboyBrian]

The repository-root binding repair in #5102 was archived before its test-fixture follow-up landed. One standalone cp1252 overlap fixture copied find_overlap.py without its new shared helper. Seven historical source-role assertions were instead stopped by the unrelated changed checker input. Copy the standalone helper closure and restore the exact manifest-pinned checker blob only inside those disposable historical-role fixtures, so they exercise their intended predicates again.

The original eight hosted test failures were reproduced. All37 affected methods pass under normal and optimized Python; the same37 optimized methods were rerun on canonical20a1e31 after this follow-up was isolated. The planner succeeds. No production code, frozen package, source pin, scientific receipt or replay policy changes; no historical scientific batches are rerun locally.

The actual frozen isolated-descendant-bytecode package still rejects working source drift: scripts/check_research_commit.py introduced by #5102. This PR documents that remaining incompatibility rather than repinning or suppressing it. Correcting the test fixture is not evidence of a successful historical package replay. Native Windows remains unrun.

Prepared by JoyboyBrian with substantial OpenAI Codex assistance; Apache-2.0. Retains the original runner, source-role and onboarding test authors. This is the focused follow-up to the same investigation and does not count as a newly claimed research task.

Hosted source-only diagnosis: plan PASS; overlap-path-index rejects an inherited base-to-recorded-blob mismatch in scripts/find_overlap.py, and research-package-pin-batching rejects an onboarding input already drifted at exact parent20a1. Both errors reproduce against parent bytes before any verifier executes. These remain failed scientific replays; no frozen pin is changed.