Render adapter: A6 frames; stability: display view fixed
Render adapter: A6 host pass objects; stability: early-freed view
Render adapter: approved specification Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Mirror current visual issue status Describe the save-reload blood report and keep the older flash and colour-noise observations on the watch list.
Mark lighting fix in 0.8.2 Mirror the runtime stability table after the texture collector fix is included in 0.8.2.
Document character lighting collector fix Mirror the runtime stability and Metal renderer conclusions after the native 1440p copied-save checks.
Mirror Cathedral Ward stability caveat Note that the separate GXRenderThread null read remains unconfirmed after the GX worker fix.
Mirror runtime stability and MetalFX notes Update the wiki with the GX worker and shader cache fixes, native renderer status, MetalFX measurements and contributor credit.
Document Metal recorder lifetime fix Mirror the runtime stability and native renderer findings after the shader lifetime fix.
Record native 1080p crash scope
Mirror the docs (aspect ratios) Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Mirror the docs: one page per doc, generated from the repository The wiki now mirrors the repository's docs (rewritten 2026-10-20): the same pages and words, with links between them as wiki links. Pages without a doc any more are removed. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Stability: GPU device loss with motion vectors fixed on main
Black sky and popping fixed with the stretching in 0.7.0
Release 0.7.0: stretched-geometry fix, trophies, motion-vector crash known issue
Document per-command capture and reported route recurrence Record bounded binding metadata, diagnostic output validation, and the missing onset capture after the maintainer closed the game. Rendering investigation is paused. Co-Authored-By: OpenAI Codex <codex@openai.com>
Record negative FaceGen-disable comparison The maintainer reported recurrence in the verified disabled repeat at hold sample 301. Preserve capture limits and exclude the first early-quit trial; no workaround is accepted. Co-Authored-By: OpenAI Codex <codex@openai.com>
Record incomplete FaceGen trial and retry Separate an unexplained early SDL quit from crash and geometry acceptance. Record owned-process runner checks and the pending repeat. Co-Authored-By: OpenAI Codex <codex@openai.com>
Record FaceGen control and isolated patch comparison Credit Kyo and pin the public patch source. The normal control reproduced; disabled result remains pending, with two external writes verified. Co-Authored-By: OpenAI Codex <codex@openai.com>
Record synchronization experiments and loading hangs Separate the diagnostic pipeline-cache cycle and teleport hang from process crashes. Keep the backing-hash repair and third retry marked pending, with no accepted geometry fix. Co-Authored-By: OpenAI Codex <codex@openai.com>
Record geometry reproductions and unified-memory diagnostics Separate coarse overlaps from exact declared binding ranges; keep the refined route result pending and exclude the post-exit capture. Co-Authored-By: OpenAI Codex <codex@openai.com>
Document local trophies and edited-save checks Credit Ambrose and Supermedo/bloodborne_pc 79d44f0. Document the protocol, local routing in online mode, synthetic tests and clean game smoke check. Keep real trophy earning and online support limitations explicit. Co-Authored-By: OpenAI Codex <codex@openai.com>
Document motion-pipeline preload guard and smoke checks Credit João Pedro Valadares and Supermedo/bloodborne_pc f33e7617. Mirror runtime credits, renderer port decisions and the two clean isolated game checks, including the motion-off validation limit. Co-Authored-By: OpenAI Codex <codex@openai.com>
Record counted GPU trace comparison and reader usage Mirror the complete-frame methodology, off/on GPU work and encoder counts, post-trace holds and limits. Keep baseline capture opt-in because the larger earlier untraced difference is unresolved. Document the frame-reader option and six synthetic tests; no game trace records are included. Co-Authored-By: OpenAI Codex <codex@openai.com>
Record hash cost diagnosis and capped CPU comparisons Mirror the aggregate diagnostics, four fault-free 45-second holds and measured 0.05-0.06 ms/frame hash-path cost. Keep capture opt-in because the earlier uncapped difference remains unresolved; record that no new visual or video acceptance was performed. Co-Authored-By: OpenAI Codex <codex@openai.com>
Document the opt-in texture baseline port and negative comparisons Credit Supermedo/bloodborne_pc 7e8bd6c1 (devilismyfriend, ayozetr) and 136a72d0 (Supermedo). Mirror normal-flow and sanitizer tests, the GPU-buffer fast-path guard, four isolated holds and the decision to keep baseline capture opt-in. Record that character creation and precise-readback gameplay were not tested and exclude the shutdown capture. Co-Authored-By: OpenAI Codex <codex@openai.com>
Document the Windows-fork stable storage port Credit Supermedo/bloodborne_pc 3a7cde8 by João Pedro Valadares, following shadPS4, and mirror the implementation, normal-flow and sanitizer tests, Linux CI coverage and isolated gameplay smoke result. No performance gain or historical video failure cause is claimed. Co-Authored-By: OpenAI Codex <codex@openai.com>
Record published 0.6.0 and release verification Mirror the approved tag/source revisions and passing release CI. Preserve the accepted stability checks and known limits. Co-Authored-By: OpenAI Codex <codex@openai.com>
Record native launcher release acceptance Mirror the packaged 0.6.0 first-profile/Home check, synthetic database verification and remaining clean-machine UI limits. Co-Authored-By: OpenAI Codex <codex@openai.com>
Publish pre-release stability measurements Mirror observed opening-video checks, self-contained CI-package verification and the 90-second gameplay hold. Record the excluded mixed-driver trial, cold project-cache limits and unchanged historical assertion counts. Co-Authored-By: OpenAI Codex <codex@openai.com>