Ran 2026-09-06 — the first frame-time row that names the discrete GPU

Done, and recorded here because the checklist is where the dev box’s own GPU-facing runs land. Every windowed frame-time figure this project had published was an integrated-GPU figure — not because the discrete part was unreachable, but because an unflagged run resolves AdapterChoice::Default and Plan 0144 deliberately left it there so no existing number would move underneath a CLI change.

Two 3-minute runs, one build, minutes apart, 1920x1080 windowed, Rich tier, rotation on at a 20-40 s dwell. The pair and its numbers are in Non-functional requirements — added beside the existing table, which was not edited. In short: the discrete part holds 165.0 fps median with a worst p99 sample of 8.936 ms; the unflagged integrated part reads 112.8 median and spends 21 of 171 samples under the 60 fps floor. Neither drops a frame.

The console’s dual-GPU degrade path did not fire, and it remains unexercised. A window pinned to the adapter that does not drive the display is the first configuration in which console surface unavailable on this adapter could execute, and it did not: the console opened normally on the RTX 3080 (Mailbox, frame latency 1) and presented 9,834 times with 0 skips. That is a finding, not a failure — on this single-display Optimus laptop the discrete adapter can present to a window the integrated part composites, so the path has still never run anywhere. It needs a genuinely multi-adapter display topology, which this box does not have; the item stays open rather than being marked covered.

(Plan 0147 Phase 6. The console figure is also a second-adapter witness for that plan’s Phase 4: the same preset that runs 53 fps on the integrated part holds 165.0 with the console open here.)