[MeNass89]

November 13 final Field check and resolution selector

• Added the BAMA AppKit output picker: Switch 2 (2048×1152), 2K (2560×1440), and aspect-correct physical-panel native output. Mac build, installed self-test/config check, four inspected light/dark/FI/native window captures, code signature verification; existing host and user settings preserved. New pure-C panel-fit fixture is part of the host test command.
• Added presets/2k60-interpolated.env: native30 + actual MetalFX-generated30, spatial output2K, HDRoff. The present implementation still requires a display detected as at least120Hz. 90FPS remains unimplemented.
• Verified Hyrule Field, not CastleTown, from the loaded scene. At native60+FI60 the static window measured 111.44 physical FPS at2K and79.96 at4K, with uneven8.33/16.67ms spacing. Some angles hold120; no locked120 qualification.
• Found the camera-turn stall: one new pipeline took144.4ms. Warm replay at native30+FI30 removed it. Each resolution's three windows total19s, with570native +570generated positive display callbacks: physical60FPS, max gap16.667626ms(2K) /16.667335ms(4K), zero gaps>25ms. Both bounded runs exit0 / no reported errors. Screenshots were before measurement in this replay;0newPSOs from60–91s. New compatible rows/archive retained privately; no game cache published.
• Same M4Pro host, HDRoff, silent sink, existing gpukeep policy active.4K is rendered workload on the built-in panel, not a physical4K display certification. Whole-game FI quality, long-session worst cases, independent-object motion and clean upstream integration remain open. PR remains draft.

Detailed methods and limits: docs/oot-m4-pro-fork.md. This update supersedes earlier pending resolution-control / untested Field notes below; earlier entries remain historical evidence.

Actual 30-native + 30-generated gameplay check

Two matched 92-second runs from APFS-cloned player saves/archives, muted, same menu inputs, Link movement and camera turn. Both exited 0 with no RED. The inspected save loads Castle Town's entrance street, not Hyrule Field.

Across three selected gameplay windows totaling 19 seconds, the 30+30 arm produced 570 native frames, 570 actual GPU generations, 570 positive generated-image presentations and 570 positive native presentations. Each window measured 30.0 native FPS and approximately 59.9998 physical FPS; physical intervals p99 16.667 ms, maximum 16.676 ms, no gap over 25 ms. Host consumed-time ratio was 0.989 versus 0.990 for the native-60 control, not half speed. This is a bounded gameplay cadence/clock check, not proof of every physics consumer, whole-game FI quality, Hyrule Field or 120 FPS. The current user profile was left unchanged.

Capture limitation: the control file is append-only; later screenshot requests overwrote it, so only the initial gameplay screenshot exists per arm. No after-movement coordinate/image-equivalence claim is made. Timing and movement input records remain valid.

November 13 follow-up: selector and introduction speed

Commit 8e52e1ff adds an installed external-build AppKit settings selector (independent MetalFX upscale/FI and 60/120 output targets; 90 disabled), validated with BAMA's existing-app flow, functional/grid tests and actual light/dark window snapshots.

The player localized slow motion to the introduction. A matched same-host control confirms the 60-FPS AV1 movie was being consumed at the 30-FPS gameplay cap: 5.950 seconds of media in 12.905 seconds. Timestamp-based detection now temporarily holds OOT at 60 native while consuming that video: 5.950 seconds in 6.130 seconds, and later 14.950 in 15.130. Repeated property reads do not prolong the exception; it expires after 500 ms. Thermal limits remain authoritative. The new cadence test passes through the actual Makefile under GCC; the Mac host builds and was installed with signature verification and backups. In-game 30-native/FI speed and quality remain unqualified.

Shared preparation lists installed and reloaded

The current player list was merged with PR #2324 and #2466 using this branch's production key parser, with Clang/arm64 verification and shader-stage validation. Seed + user list: 20,068 unique entries; combined list: 22,042. The public lists contribute 1,974 additional entries (169 task/mesh, 255 render, 24 compute, 1,526 shader loads), after deduplication; no parser/stage-map rejection.

Actual Mac prewarm finished in 30.54 seconds with no RED errors, including 168 newly built/captured task/mesh pipelines. The profile/archive was activated reversibly without replacing the shader library or save. A new process reloaded the 22,042 entries in 0.88 seconds: 350 mesh pipelines came from the archive, two were built again; three bases were also built again. This is a substantial prepared-coverage improvement, not a zero-rebuild or complete-game cache claim. The evolving user list is still loaded on subsequent starts.

Summary

• Share a player-confirmed OOT/M4 Pro checkpoint with a portable 2560×1440, 60-native-FPS cap, MetalFX spatial, FI off, HDR off preset and a public build recipe.
• Make the underlying correctness, resource-lifetime, MetalFX interpolation and timing work available for review and community contributions, with focused fixtures and explicit validation boundaries.
• Triage the work into general mechanisms, mechanisms needing engine/platform adaptation, and OOT-specific contracts; identify fixes already present in main instead of proposing duplicate implementations.

Why this is a draft integration PR

This preserves the engine checkpoint actually tested, rather than rebasing a large renderer change and applying old results to new bytes. It is a contribution map and a reproducible fork entry point, not a request to merge the whole old branch over current main. Current main has advanced substantially and already contains several mechanisms listed below. Selected extractions should be rebased, reviewed and tested independently.

The initial shared engine pin was 09a2572523bea861d0c7c7006826e57801d16de5; its common ancestor with upstream is f02d0474cf1ed1de32ccee1313e1448d4f7da401. Before the sharing documentation, the actual contribution range was 112 files, +6,613/−374 lines. An endpoint-tree comparison wrongly presents later upstream additions as removals; use the PR/merge-base diff.

Public entry points:

• docs/oot-m4-pro-fork.md: build, settings, provenance, tests and remaining limits.
• dm/games/0400486024DEC000/presets/2k60-native.env: portable settings only; no private paths or game assets.
• docs/oot-upstream-triage.md: extraction map, prior art and contribution priorities.

What the player actually tested

The user reported normal speed and very smooth 2K play on an M4 Pro after we switched to 60 native FPS with interpolation disabled. Live config confirms this. The user later reported Hyrule Field still imperfect, so this is not a locked-60 claim for every area.

An attempted 30-native / 60-displayed profile produced user-observed slow motion. Returning to native-60/FI-off restored normal speed. This initial observation changed two variables; the follow-up above subsequently isolates and corrects the introduction movie cadence. Gameplay at native 30 remains a separate validation item. The host's game time … 1.000 of real time counter does not prove every simulation or animation consumer is correct.

The requested final product is 30→60, 45→90 and 60→120, with Switch 2, 2K and Mac-native output choices. Those nine combinations are not implemented/qualified, and this PR does not silently enable them. 45-native timing needs an actual fractional clock contract; the present integer 60-Hz stepper does not supply one. Evenly displayed 90 FPS also needs a suitable display mode or validated VRR behavior.

Triage: general mechanisms

These are semantically general; that does not mean the combined branch has passed other games.

• Uniform re-recording and short-buffer correctness: preserve command-record byte spans and valid uniform content. Public-API/synthetic regressions in t_nvn.c and t_uniform_padding.m; checkpoint 7ca292f9.
• Bounded CPU texture-upload batching: retain targets and respect the staging budget inside open submissions; 2d1003fc, CPU-texture/upload-budget/upload-slot/direct-upload fixtures.
• Finalized texture/view lifetimes and dirty registration epochs: e9678f60, 09a2572, ghost/view-cache and dirty-reuse fixtures. Main already has corresponding owner/ghost-view and born-epoch mechanisms, with a newer owner layout; credit them as carried fixes rather than merging a duplicate.
• In-flight visibility/ring ownership: 5b77e8da, exercised past the former 64/128 limits by t_wait_event.m. Main already contains SUB_FLIGHT_MAX and the corresponding rings.
• Residency commit outside the hot mutex: a5863042, t_res_commit.m. Main already has its separate commit mutex and deferred changes.
• Pipeline prediction on a genuine new miss: be568428, t_pso_pending.m. Main already has pso_predict_due and g_pp_admit.
• Bounded deferred diagnostics: 189f313a, 92b0d024, exact TSV output/overflow/concurrent drains and flow traces. These remove intrusive timing writes; they are not a renderer-FPS claim.

Triage: reusable after adaptation

• Immutable FI snapshots and async resource ownership: source/history validation, generation and display reader timelines, retained interpolators, paired presentation. Another engine still needs correct color/depth/motion surfaces and reset boundaries.
• Legal MetalFX spatial output: private outputs, conditional final copy/render, and checks for format, size, orientation, usage and storage mode (cd966bef, a3c568e3). Not every drawable is a legal scaler destination.
• Profile-driven spatial replacement with native fallback: color encoding, grid and binding contract (149fa1ea, 7dda971c). A replacement must also preserve scene-discovery notifications; skipping those originally disabled FI silently.
• Camera/depth reprojection: the geometric method and validation are reusable; observed agl offsets and object-motion limitations are not universal.
• Native Mac panel sizing: distinguish physical pixels from logical desktop points (61411451). External screens, scaling and refresh changes need platform validation.
• HDR/EDR transport research: present in history/fixtures, disabled in the delivered preset and outside the current release promise.

Triage: OOT-specific today

• 0400486024DEC000 v131072, the observed camera layout and frame-ratio path.
• The guarded fa3033d05c0a41d6_b square-root-linear recipe, 1536×864 scene, 2048×1152 intermediate and exact bindings/grid.
• Pre-UI/final scene discovery, matching depth, history-reset boundaries, prewarm rows and routes.
• Build-gated native-material-refresh research. It is not installed or enabled by the normal preset.

OOT's executed four-stage NDSR chain is spatial; it does not supply observed temporal motion/depth/history data to translate directly. FI currently reconstructs camera motion using depth. Independent Link/Navi motion, translucent ocarina tutorial overlap, dialogue sharpness transitions and scene cuts remain quality work.

Test plan and evidence

Run for this publication

• python3 -m unittest dm.tools.tests.test_docs dm.tools.tests.test_game_profile dm.tools.tests.test_nsp_chain: 32 passed. Docs were rerun after the latest player-report update.
• Actual pure-C fixtures compiled under Linux/GCC: t_agl_camera, t_pace (10 checks), t_fstep (78 checks), t_tsv_stream, t_flow_trace (20,000 concurrent intervals): pass. These are not described as Mac/Metal passes. The initial Makefile attempt found no Linux clang; the actual fixtures were subsequently compiled with GCC, not a disguised compiler shim.
• Preset checked for unique keys, absence of local paths, correct 2K/native-60/MetalFX settings and FI-off state; git diff --check: clean.
• Actual Mac impact command, with no game run: python3 scripts/regress.py impact f02d0474..09a2572. Completed successfully. Breath of the Wild, Pokémon Scarlet and Tears of the Kingdom are affected, one additional profile only through broad API rules, 152 profiles unknown. These are reachability results, not gameplay passes.
• Live read-only verification: installed host hash, effective profile, shader map identical to the committed map, and formerly dirty hle_vi.c, nvn_temporal_async.inc, front_mac.m in the build mirror equal their committed 5b77e8d versions. The user's running installation was not rebuilt or restarted for publication.

Existing engine/runtime receipts

• 09a2572's four code/test files exactly match the final candidate source hashes. Sixteen focused runtime fixtures passed across the final receipts on M4 Pro.
• t_vc_ghost: 19 checks enabled / 17 disabled, zero failures. Old dirty implementation fails the three new recycled-registration checks; the candidate passes them.
• t_tbuf_align: 80 checks, zero failures with its Makefile's normal environment. An earlier extra-validation harness abort was corrected; intermediate failures are not relabeled passes.
• In-flight visibility fixture: 101 checks per encoding mode, including 160 retained submissions and old ring boundaries.
• Pre-signature host SHA256: 44d7ac2ec80898c17ffec98013c37b3a12b9761f3648ff4a4ca43bf74ca2fc1d; installed ad-hoc-signed host: d6273954e9798d2e4c842f52e90950daf897cd8e0ee07eadd1a0310d05b3ba4e.

Still open before merge/release qualification

• Clean public packing/reproduction with the new portable env file.
• Complete current graphics suite, lint ratchets and repository regression on impact-selected games. Only OOT has actual player/targeted evidence for this combined branch; the other games are unchecked.
• Integration conflict resolution onto current main, with fresh tests for the resulting code.
• In-game 30→60 speed/quality validation and 45→90 clock/display contract (introduction cadence corrected in the follow-up above).
• FI object/transparency quality, stable tail latency in heavy areas, latency and energy measurements.

Community work with the best immediate payoff

1. Reproduce the native-60 preset and report chain, game update, hardware, scene and helper policy.
2. Deduplicate compatible prewarm additions from Ocarina of Time's prewarm list gets 171 more task / mesh pipelines (data only) #2324, 0400486024DEC000's prewarm list: 84 task / mesh pipelines and 1,806 rows from 18 areas #2466 and game: Ocarina of Time prewarms 89 more task / mesh pipelines #2550 against the effective manifest plus user list. No complete shared Mac cache is certified here; preparation can reduce first-visit compilation stalls without changing rendering quality.
3. Diagnose 30-native FI off/on in one matched moving-scene capture, separating simulation from presentation.
4. Extract still-relevant generic mechanisms individually onto main; keep prior upstream implementations and their credits.
5. Address heavy-scene GPU budget and independent-motion/transparency quality using real evidence rather than larger benchmark campaigns. Neutral experiments are not included as proven gains.

Front-ends: the new portable 2K60 preset is native-60/FI-off. No launcher claim that 30→60, 45→90 or sustained 60→120 is ready. Game files, caches, saves and machine-specific installers are excluded.