Native GL stencil/depth fixture observation — 2026-11-01

test/test-opengl-stencil.js passed unchanged on native Node 24.18.1, exit 0 in 7.990 seconds, under a sequential 120-second timeout. The complete 8,719-byte fixture was read before execution; SHA-256: 411833ed93bd2fc12caeb806d7a9e85a88dea30709dbafa41f4841efb6395f59. No matching accepted complete-input test result was found, so this assignment executed the fixture once. No production, test, helper, or configuration edits.

Fixture contract Observed outcome
Default stencil queries; masked clears; one-pixel scissor clear PASS, including neighboring pixel preservation
Point stencil replacement and software color PASS, stencil 7 and red 0xffff0000
Alpha rejection before stencil/depth PASS, rejected alpha leaves stencil 7
Stencil-fail, depth-fail, depth-pass operations PASS, increment/decrement/invert paths
All eight stencil comparisons PASS, reference compared against stored stencil
Stencil rejection cannot write nearer depth and reject a farther fragment PASS, subsequent farther fragment writes stencil 7
Saturating increment/decrement, ZERO, KEEP, read/write masks PASS
Triangle and line stencil spans PASS, stencil 12 then 13
Reference clamping, invalid enums, begin/end state protection PASS
Attribute masks, context isolation/deletion, display-list state PASS
Unsupported stencil-index readback Expected WebAssembly.RuntimeError caught by the fixture
Invalid clear bits and invalid display-list stencil function Expected GL errors asserted
Stencil attachment allocation failure PASS, ensure returns 0 and GL error 0x505
Attachment resize to 8×8 and back to 16×16 PASS, pitch and all 64 cleared bytes asserted
Two queued draws with immutable stencil snapshots PASS, manual native worker-draw replay produces stencil 4
Reset releases the owned stencil attachment PASS, pointer becomes 0

The raw log's heap refusal for 4294836225 bytes (0xfffe0001) is the intentional allocation-failure case, followed by the fixture's final PASS. There was no unexpected assertion failure and no repair or weakened assertion.

Gate, build and actual module

Evidence: scratch/runs/20261101-gl-stencil-depth/ (ignored). Predecessor Incoming01a196a6-6ab0-7e43-9802-e5be6bb9f096 supplied fresh launcher final.txt and result.json and an explicit runtime/source/build RELEASE at 10:47:38 UTC. The gate watcher observed both at 10:47:47 UTC; source/test/build access followed that gate. Actual main was c834033c1adfb181a2e3d372be618fd739c71b06, advanced from the assigned 558e59dcf36b92471184154015cb46ac7b563950. Runtime allowance ended 11:02:47 UTC, within the overall 13:15 deadline.

The accepted canonical locked-build evidence from 20261101-dark-reign2-native-mip-implementation was reused: all 4,392 accepted input hashes and all 141 recorded build reads matched, as did the canonical module and generated region mirror. No canonical rebuild was needed. All 4,444 currently pinned inputs matched before/after execution.

Canonical module SHA-256: dd43e4a6624d2449d7133b731d687c32530f7fcf0041a69ba71a09f78c09aaf4. Actually instantiated fixture module SHA-256: 98741a0c50f0f7713bd306d2a6ec56eefc26200b34907bc454f7cdcb3731fd0e (2,027,097 bytes). It is not byte-identical to the canonical module: all 12,982 production function bodies are exact matches and all production exports remain, while the unchanged fixture adds three functions/exports: stencil_pixels, stencil_pitch, and stencil_allocation_failure. Noncustom sections 3, 7 and 10 differ accordingly; other noncustom sections match. The canonical 37-byte custom section is absent from the test module. verification.json retains section hashes and all three added function bodies.

The read observer recorded 201 unique file/hash records, including 130 src/ files; all still matched at verification. Raw compressed output, Node exit metadata, actual module bytes, source snapshots, build-reuse proof, fixture copy and verification are retained in the run directory.

Observer and limits

The exact accepted observer and offline fixture from 20261101-gl-software-projective were reused with matching hashes and input proof. Retained raw compressed output, exit 0 and validation show actual synchronous and asynchronous Wasm execution, before/after hashing of the whole 512 MB shared memory, unchanged host/native clock identities, and emitted late retention through event 10001 (last 64 retained). The observer records readFileSync, complete bounded descriptor reads, Wasm compilation and instantiation; it does not establish exhaustive asynchronous filesystem, streaming, preloaded-module, or guest-memory diagnostic coverage.

This fixture executes native WAT/software rasterization and queue contracts. Its host override intercepts GL commands and captures queued draws; replay calls the native software worker export directly. This does not establish GPU execution, original-game pixels, rendering parity, real worker-thread scheduling, full27 membership, or general gameplay/crash freedom.

Disk samples stayed above 2,000,000,000 bytes: wait minimum 2,030,944,256; test minimum 2,028,408,832. No separate startup or RAM gate, foreign cleanup, browser, GPU, boat, download, deployment, worker launch or source repair. Local native box ID/state/expiry are N/A. Own timeout process group and driver exited; no test-created temporary path remained. Publication watcher cleanup and strict-host-key remote verification are retained with the final result.

Next: use this as the bounded native stencil/depth result. GPU, original-game pixels and parity require separate evidence outside this assignment.