Independent SDL fixture integration, 2026-10-24

Accepted candidate 3509517e9d3eb999dc2190ccc0168f96a8bffe2e as a scoped test-fixture repair and retained historical internal SDL consumer evidence. No production implementation changes.

The historical GetProcAddress allocates a fresh eight-byte thunk per lookup. The repaired fixture accepts distinct records only after checking exact waveOutGetErrorTextA, API 4590, record range/alignment and name bounds. All original consumer bytes, output/NUL/canary, caller argument, return sentinel, ESP/EIP/yield checks and the negative output oracle remain byte-exact against producer ce17debea05b1e40c2359a25cf1d3d2c41266649, apart from the identity check and diagnostic field.

Independent verification matched every candidate path to an actual Git blob; verified the original 35 seals, repaired 37 seals and closeout 63 seals in their documented relative scopes; verified all 41 historical closure files (2,512,870 bytes), API pin, source excerpts, recovery recipe, SDL/module identities and baseline/negative receipts. Relative seal paths were normalized for Git object access; the first review attempt rejected an unnormalized ../../../test/ object path before this accounting correction. No acceptance oracle changed. Historical failed baseline and capacity denials remain preserved.

Reused corrected baseline exit 0 and unchanged --reject-output exit 1. The actual negative stderr fails specifically at original SDL consumer output oracle for supported./SUPPORTED.; both commands bind the same fixture/module/SDL/source, and argv differs only by the negative flag. Neither native run was replayed. Cleanup receipts establish reaped native groups; resource receipts establish admission for those historical runs.

Fresh cheap checks passed with Node heap 256 MiB, CPU/wall 60s and fresh MemAvailable >=805,306,368: actual helper regression, 11 additional rejecting controls plus exact-end positive boundary, candidate syntax, and complete test-tier validation using the exact final Git test-tree inventory. The unchanged tier implementation classifies the helper test as unit and native candidate as e2e; metadata repair is unnecessary. The inventory adapter changes only directory enumeration for the explicitly sparse checkout and is retained for review. This is a membership check, not a full test-manifest timeout scan or full build.

Adjacent evidence records exact Git blobs, hashes, source binding, checks and task resource receipts. Integration reconciles actual remote main under the exclusive shared lock and preserves unrelated main files. Main/task commit and verified remote refs are recorded in the task-root publication receipt and append-only messageboard.

Historical source b41a7c5f4b752d1f0bdab1141a11d4fed4dc6f9a, module SHA256 8fac39936b6df2418ad86cab118cef19a65a727f649fbc14c412b37818904686 (291 imports, layout 9a3bc9149d58209c), SDL SHA256 55eced17cf8a9b7acc671cc307503de847a3c36fdafafe14cee7917433cfe410. This does not qualify current-main runtime semantics, public SDL_OpenAudio, a real waveOutOpen failure, gameplay, audio listening or parity. No build, native instantiation/replay, guest, acquisition, browser, deploy or box operation. No box assigned; lifecycle remains with coordinator.