[Arkex-ai]

Summary

Adds 59 pipeline states absent from current main, recorded on stock main in daytime Kokiri Forest on an RTX 5070 Ti. Refs #2034.

Technical changes

• Add one version 2 list and its provenance note through the official nx2/pipeline_states.py add command. These remain the only two changed files; both blobs are unchanged by the rebase.
• The list contains 78 unique records: 59 new to main and 19 also in the independently replaced tour. There are no duplicates in main's permanent contribution lists, and the refreshed complete open-PR file inventory found no competing new list; build(release): the RTX 5090 fork list's note says what its build changed (#2143 follow-up) #2703 only clarifies an existing RTX 5090 list's provenance.
• Preserve the actual recording build, settings and scripted route. Rebasing does not make this a new recording or a current-main gameplay test. The list contains hashes, formats and draw state, without game code or data.

Validation

Current tested contribution head: 102e2608db73aeab3c08f0779aeae8908faf57de, rebased onto main 51da0464b4cf0c7f56b81ebd6bb309325835e175. The recording remains from ef81aee767f848255ca08171bba33b8c40c79347.

• All 12 focused pipeline-state and affected upstream checks pass on Windows and Linux with zero failures/skips, including build-key, Windows emit, team-chat/test-request and V-Sync regressions. Both contribution blobs are unchanged.
• Earlier local full suites at b20f55ed on main 46780503: Linux 468 passed / 0 failed / 24 skipped; Windows 454 / 3 / 35. All three Windows failures reproduce on that exact clean base (test_bug_report.py, test_input_background.py, nx2/test_nup_paths.py). These full suites predate the latest rebase; they are not current-head full-suite results.
• Official list check at b20f55ed: 41,898 unique records; neither upstream nor this rebase changed the lists or checker. Current-head source checks and Windows Builder pass. The AppImage job failed before compilation in System packages (timeout exit124). A maintainer rerun is requested; the exact-head CI is not fully green.

Original recording and native-consumption evidence

The fresh stock recording at ef81aee7, with precompile disabled and the original renderer source, contains 837 unique version 2 states from 572 programs. All 78 contributed states occur in it. Older recordings were excluded; deterministic compression and the original independent review passed.

On the recorded stock build, Vulkan preparation noted all 78 states and 61 programs, with zero stages absent from the seed and zero missing program variants; normal exit 0. One general seed stage outside this contribution was untranslatable. Driver caches were warm: this is list-consumption evidence, not a cold-start benchmark or a new-head game run.

Only title/file select and the child-Link balcony camera route were checked. No stutter/frame-time gain, cold preparation time, late-game coverage or other-GPU result is claimed. The draft will be reconsidered after the new exact-head checks complete.

[Arkex-ai]

At 102e2608, source checks and the Windows Builder pass. The AppImage run stops before compilation in System packages, ending with timeout exit124. The other rebased PR, #2649, also fails in this step and logs a Launchpad HTTP504.

This PR does not change the workflow or its package setup. Could a maintainer rerun the failed AppImage job? The description now records the failure; no full-green result is claimed.