Ran 2026-09-11 — the studio beside the player, Windows half (Plan 0167 Phase 8)
Half the phase. The macOS arm is unrun — its done-when names both machines, and one machine’s readings are not the comparison it asks for. Plan 0167 closed on 2026-09-11 with this arm still unrun, by the owner’s decision, so this entry is where the macOS reading is owed now; the plans index carries it under Standing (not a plan).
Two runs, same build (0.117.0), same machine (RTX 3080 Laptop, DX12), same preset held (Whorl),
--input loopback, Rich tier pinned, rotation held. The preset must match across the pair —
a first attempt compared Whorl against whatever the studio happened to load and read as a 76 %
collapse, nearly all of which was preset cost rather than preview cost.
| player alone | studio attached | delta | |
|---|---|---|---|
| samples | 257 | 92 | |
| fps mean | 140.1 | 125.8 | −10.2 % |
| fps min / max | 71.3 / 165.0 | 102.3 / 146.6 | |
frame_ms_avg | 7.31 | 8.04 | +0.73 ms |
frame_ms_p99 | 14.56 | 15.30 | +0.74 ms |
| player frames dropped | — | 0 of 17,077 | |
| preview | — | 2,915 painted / 23 dropped | 0.78 % |
The comparison the phase exists for. Plan 0159 Phase 4 recorded 42 → 35.4 fps, −15.7 %; this pair is −10.2 %, so the gap narrowed. Read the percentage, not the pair: the unattached baselines differ more than threefold (140.1 against 42), so the absolute figures are not two measurements of the same thing.
Fullscreen was toggled three times during the run and the preview survived every one — the gesture that used to end it. Frames kept painting and the strip kept reporting.
bgra8 reproduced here, which discharges one of the plan’s own risks. Its risk list allows
that the development adapter might negotiate RGBA, leaving Phase 4’s swizzle exercised only by a
synthetic test. It negotiated 640x360 @ 165 bgra8 in windowed mode and the colours are correct,
so the swizzle is confirmed by observation rather than by proxy. The windowless path negotiated
30 rgba8 in the same session, so both formats are covered on this box.
One defect came out of the run, and it is filed rather than recorded here:
design-backlog 0205 — in windowless mode the player reports 0.0 fps and
writes no diagnostics.log rows at all while rendering normally.
(Plan 0167 Phase 8, Windows half. The player’s own figures are read from diagnostics.log, whose
third column is frame_ms_avg rather than a p50 — the health event’s frame_ms_p50 is a
different statistic and the two are not interchangeable.)
Built from 13c7582 at version 0.158.0. This site tracks main and is not versioned per release.