Link branch-only Android harness, raw emulator/Retroid results and screenshots. Record eight synthetic checks per target and explicit real-game/performance limits. Mainline engine, app sources and licensing remain unchanged; no replacement APK released.

Link 9e758ab evidence: candidate 18/18 on Retroid/emulator, unchanged native baseline 10/10 on Retroid, and 17 frame comparisons. Record unresolved baseline emulator failures and synthetic-only acceptance boundaries. No app source, license or release change.

Link the six host checks, three exact game variants and eight captures on d91321b. Keep public release, experimental source and local research APKs distinct; record DiRT auto-profile failure, lifecycle artifact boundaries and unmeasured performance. Main application code and licenses are unchanged.

Link the branch-only measured artifacts, same-scenario Retroid Asphalt comparison and emulator lifecycle captures. Record higher candidate paint cost, visible rendering differences and rejected emulator timing separately from replacement acceptance. Longer fixed-sample comparisons remain pending; no release or license change.

Point the public status page to 86d7906 with 17 host/raster checks, before/after evidence and Deep/DiRT captures. Distinguish rejected surface timing, unavailable paint cost and pause-menu return from complete acceptance. Keep current mainline engine, APK channels and licensing unchanged.

Link research revision 044d9a2 with reproduced material/fog failures, 32 measured-host checks and 18 standalone checks. Document Deep timing and resumed gameplay, unresolved Asphalt shadow and observation limits. Main production engine, licensing and APK availability remain unchanged.

Record conditional maintainer authorization for a ready beta, the newly identified EPL tone dependency and candidate-only replacement, verified combined-host checks, and remaining production/licensing/upgrade work. Keep main and public release availability distinct from experimental source.

Keep research, Java-only verification and previous combined-build evidence distinct. Document platform math replacement and the next bounded GPU investigation; do not announce production parity, license transition or a beta release.

Update the product handoff for website and post consumers: new native raster code, independent texture wrapping, 63 checks per target, exact-pixel same-APK speedup and observed Retroid race. Keep experimental source, production and public releases distinct.

Record four-layer and fog implementation, the zero-density tint fix, current 66-check device results and observed Deep gameplay. Distinguish same-APK synthetic speedups, residual simple-scene overhead and outstanding replacement/release work.

Record b624a71 implementation, mip sampling/cache fixes, 71 checks on each device and exact 18-scene comparisons. Keep synthetic timings, Retroid smoke evidence, remaining filtering/performance gaps and unreleased production status separate. This commit contains documentation only.

Record 610302b implementation, corrected LOD, independent color checks and 74 passing checks per device. Distinguish improved image quality and its higher cost from Java/native speed comparisons, game acceptance and the remaining production-release gates. Documentation only.

Record source checkpoint 197f995, 81-check device results, unchanged scene pixels, same-APK cached versus uncached timings, and the short Asphalt race check. Distinguish this experiment from production/release readiness and identify GPU rasterization as the next proposed implementation rather than promising parity from CPU optimizations.

Reference source checkpoint 5b777ae, 87-check device results, hardware execution counters, same-APK timings, precision differences and comparison images. Keep GPU off-by-default research separate from main, published APKs and real-game acceptance; identify resident GPU batching as the next implementation target.

Describe checkpoint 88d4808, 92 checks on each device, identical residency images and bounded same-APK timings. Keep current production, licensing, private builds and public releases distinct; record outstanding GPU effects and game acceptance.

Update the website/post source of truth with the research-only checkpoint, 98 checks on each device, bounded texture-workload gains and GPU-confirmed Asphalt/Deep captures. Keep main, release and full-game acceptance boundaries explicit.

Update the website/post source of truth with the separate research checkpoint, signed-depth regression repair, 104 checks on each target and fresh Asphalt evidence. Preserve performance, audio and release limitations.

Separate research support from production availability and standalone benchmark cost from game FPS. Link the 285d26c implementation, 112-check device results and Deep screenshot; retain default-off and release boundaries.

Link source checkpoint 52ec72d and the final 119-check Retroid/Phone evidence. Record the fixed allocation regression, unchanged benchmark images and measured limits. Keep research, main application code, licensing and public release availability distinct; name GPU submission batching as the next implementation step.