docs(research): complete the PR2101 Linux replay evidence

[petalas]

Summary

Complete the executable Linux/WSL2 audit left pending for Adam Gall's PR #2101, exact head a247ef876e50bf3a7d890c7c1f9b409576674d7d.

research/selected-final-closure-linux-audit/
  README.md, TASK.json, SOURCES.json   scope, reproduction and exact pins
  EXECUTION.json, *-stdout.txt        commands, environment and execution evidence
  COMPARISON.json, OUTPUTS.json       every output comparison and raw hash
  OUTPUT-DELTA.tar.gz                 six actual changed outputs, 26 KB
  REPLAY-RESULT.json                  complete unchanged control receipt

No caller, comparator, frozen source, shared review disposition or selected record changes. Credit Adam Gall, commdata2338, Josie Elliston, Maxime Fleury, Antoine Olivier and the retained supplier/compiler lineage in the README. Coordination: Discussion1472.

Evidence

• Before: the dated review completed the supplier stage but exceeded its batch budget before finishing PR2101's composed comparison and controls.
• After: CPython 3.12.15 on Linux x86_64 under WSL2 completed both an isolated direct run and the registered replay, in 430.221 and 421.852 seconds. Each checked all 105 outputs, the coupled 13-line outer log, five positive controls and 287 rejection controls. The complete deterministic receipt is byte-identical to the predecessor.
• Actual output has 99 raw-byte matches and six differences, including the invoice and both authenticated raw-hash references. Under the unchanged declared policy the full report has 103 bytes classifications and two declared-timing differences. The README distinguishes normalized classifications from raw-byte equality.
• The six-file delta reconstructs the entire first output tree from the pinned reviewed archive. A separate round-trip check matched all 105 sizes/hashes, the complete comparison report and the outer log.
• All 144 test_research_*.py workflow tests and the selected baseline binding check passed. Source-pin validation passed before execution. Before submission all 42 scientific input hashes still matched canonical main f0cf35964766b21a67c5e3b52e4a7daa4ca92a13, priority revision 5. Fresh overlap covered 489 open PRs and 1,605 discussion/comment/reply records.

The README gives a copyable exact-head worktree/export/replay recipe. This evidence-only package intentionally adds no new verifier or registered hook. The 287 controls were independently executed, not independently reimplemented.

Merge Danger

Door: two-way. Only a scoped evidence package is added.

Blast Radius: documentation.

This is not make frontier1094-verify, supplier admission or a new kappa. Ancestors were reused from authenticated archives, not rebuilt. Maintainer source/policy review, full cold ancestry and inherited all-size, physical, hosting, precision/tape and analytic obligations remain. No hosted-CI result is claimed.