[mihir-s-05]

The current selected-source inventory can make research_baseline_sparse.py reject a valid Windows sparse checkout. At canonical main cef4b52f2a59009865bf8eabde1be3f159063703, 503 omitted inputs produced a 38,351-character Git command and WinError 206. The helper then incorrectly reported the first genuine sparse omission as an invalid file.

Batch selected-input checks within a conservative 16,000 UTF-16-unit quoted-path budget, reusing the existing sparse/HEAD validator. Complete omission coverage, schema routing, ordinary-deletion and staged-drift rejection, and failure propagation are preserved. Five new real-Git regression cases exercise 600 paths, including spaces, literal brackets and supplementary Unicode, an actual CLI/materialization round trip, late invalid entries and a later Git failure.

Validation at 9a06fb2983202e951e294774a56a15c5a169867e:

• All five new cases failed before the fix on Windows, including the unmocked CLI.
• All 14 sparse-baseline tests pass normally and under -O on native Windows / CPython 3.12.0; the same 14 pass on Ubuntu/WSL / CPython 3.12.3.
• After selected inputs were materialized, the preflight reports no omissions and the unchanged successor reports PASS_RESEARCH_BASELINE_BINDINGS_ONLY.
• All 6,136 top-level SOURCES/VERIFY manifests were checked against their committed blobs; none pins or declares the changed preflight. The exact committed-change planner recognizes all three changed paths and selects zero scientific packages.
• git diff --check passes.

The broader Windows helper run executed 362 tests: 322 passed, 22 skipped, and 18 encountered setup errors. Twelve use newline-containing filenames unsupported on Windows; six initially lacked a sparse fixture. After fixture/source hydration, three source-role methods still failed on native Windows at alternate-store historical-source resolution. All 18 affected methods passed on Ubuntu/WSL. These unchanged fixture limitations are documented; a clean full native Windows suite is not claimed. The full mathematical cold replay was not run.

This refines the portable-replay card (priority revision 7; nonce 1568f02fbdb6dc3102dbdeab4f3c4e6e). Canonical claim and coordination. Final overlap covered 308 open PRs, 4,007 Discussion records and all 6,601 tracked local notes; it found this claim and the distinct fixture-maintenance predecessor. Canonical main was fetched again before publication and remained unchanged.

Evidence, reproduction and remaining scope. The emitted directory-materialization command and other callers of the generic validator remain separate potential large-inventory boundaries. Selected kappa, frozen scientific inputs, the pinned chooser and baseline implementations are unchanged.

Credit Douglas Colkitt and the credited Claude/Mister-Meeseeks preflight (#3725), Arham-Tahir64's sparse contract (#3102), lee-phys's schema repair (#4631), Clear1nsight's fixture cleanup (#4857), and the retained onboarding contributors. Prepared by Mihir Srivastava with substantial OpenAI Codex assistance; no subagents or independent human review. Apache-2.0.

Hosted validation subsequently passed for this exact PR head: run 39891861128, job plan. Its Linux helper suite ran 362 tests in 142.993 seconds with 11 expected platform skips; the changed-package planner emitted {"include": []}, and scientific replay was correctly skipped. Local platform limitations above remain recorded. All three published files also passed a raw working/commit byte comparison, and the pushed branch resolves to the stated commit.