Preserve six input roles in Windows selected-output comparisons

[AAAmirinu]

PR2101's complete selected-output comparator rejects a correctly bound native Windows receipt at its POSIX-root validator. This opt-in adapter authenticates the six exact native input-role keys and their same-side hashes, translates only those keys, and delegates all 105 outputs to the unchanged source-pinned comparator. Literal path-like strings elsewhere and the original invoice/log/timing checks stay covered.

The bounded interface accepts canonical POSIX and ordinary absolute Windows drive roots. It rejects ambiguous spelling and case-colliding roles; UNC/device/extended-length paths remain unsupported. No supplier code or selected frontier is modified. The package is research/selected-windows-role-adapter/ only.

Validation on native Windows / CPython3.12.11: metadata/source pins passed; registered normal and optimized replays both passed 6 positive and 176 negative controls (63.81seconds total). Controls include all fifteen role swaps, six-role key/digest failures, all105 output omissions, unsupported roots, a real six-key/five-casefold-identity collision, scientific/raw-link changes and immutable predecessor corruption. A separate Codex review verified all five vendor files against PR2101's exact commit and found a test-coverage gap, fixed before final replay. No independent human or hosted-CI pass is claimed.

Base current canonical main 0fdd7d82de03e43c9e5a1000dfbcd380d1b687d2, priorityrevision4. Refreshed before submission; final overlap covered552 open PRs and1063 Discussion records with complete pagination/no warnings. Claim: #1472 (comment) . The original Windows supplier handoff was already claimed in #2333, so this scoped complementary comparator work avoids duplicating it.

Limit: these are constructed path relocations of authenticated reviewed outputs, not a fresh supplier run. Concurrent #2336 reports a native Windows successor replay and CRLF-only inner logs. This adapter retains PR2101's byte-exact log policy; it does not claim to admit that complete Windows run. A separately reviewed log policy/composed caller remains open, as do cold ancestry and all inherited mathematical obligations. No new kappa.

Credit Adam Gall (#2101 at a247ef876e50bf3a7d890c7c1f9b409576674d7d, #2059), commdata2338 (#2040), Josie Elliston (#1824), josecarlosgonzalezroque-design (#2316), Maxime Fleury/Antoine Olivier, and Rohan Arun, Jeremy Kothe, Hao Yan, Douglas Colkitt and retained supplier lineage. Predecessor files/notices are unchanged. Prepared for AAAmirinu with substantial OpenAI Codex assistance in design, implementation, tests and writing. Apache-2.0; no merge, closure or frontier change requested.

python3.12 scripts/research_package.py --package research/selected-windows-role-adapter

[ywz65]

DONE: submitted PR2477 from ywz65. Native Windows CPython3.12.7 registered replay passed all11 stages,105 outputs,30 original raw references per side,6 positives and203 rejection controls. Separate AI source/proof review found no blocker. Scope remains final-stage execution with authenticated archived inputs; no cold ancestry or changed selected result.

Original claim retained:

CLAIM — composed native Windows caller and authenticated LF/CRLF log policy (ywz65, additional batch round 8).

I am taking the explicit PR2344/Discussion2333 handoff: compose the unchanged six-role adapter and PR2101 complete comparator with a narrow log representation policy, then run PR2336's pinned successor on native Windows. All 30 same-side raw reference edges must authenticate before normalization. Only the eleven named inner logs may change CRLF to LF, with the corresponding 22 receipt links rebound after authentication; the complete outer log remains tied to the same-side times. Bare CR and any other byte changes remain rejected. No generic hash exemption is proposed.

Base main 0fdd7d8, priority revision 4. Sources: PR2336 head 3747d54 and PR2344 head b599a78, retaining PR2101/2059/2040 provenance. The original Windows inventory failure was reproduced; it is credited, not counted as a new finding. Initial successor run reached the inherited bit controls but encountered sandbox TEMP permissions; a workspace TEMP rerun is in progress.

Expected artifact: one opt-in composed caller, actual final-stage replay, complete 105-file comparison, positive representation controls, and stale/repaired log-reference, role, output-omission and semantic failure controls. No cold ancestry, mathematical admission, selected-source change or new kappa. Evidence remains LOCAL-ONLY until a checked submission. Claim expires 2026-10-15 14:00 UTC. Substantial OpenAI Codex assistance disclosed.