iGPU-gated — the Floor reading NFR §1 asserts

This is question 1 above, finally with an instrument. Extracted from Plan 0207 at that plan’s close (2026-09-23) so the plan could close on its two completed dev phases without waiting on a box that is not in hand, exactly as every item above was. The machine it needs is §9’s “Older Windows PC (iGPU)”, which is not the dev box and not the Arch laptop.

What Plan 0207 changed is that the walk now has a number rather than an impression. Phase 2 gave shot --report an advisory per-preset frame cost in ms/frame at 1920x1080, taken fully driven, with a header naming the adapter and the build profile and a ! on any reading past the 60 fps budget (ADR-0232, Capturing). It is headless — no window, no present, no compositor, no audio thread — so it ranks presets and predicts nothing about the app, which is exactly why the window’s F3 overlay is the cross-check here and not an afterthought.

• Name the machine in NFR §9. Make, GPU and OS build, replacing the class “Older Windows PC (iGPU)”. Until a row names a configuration, no reading taken on it can be compared with a later one (ADR-0071). This half needs the box only for its identity, not for a measurement.
• Walk the shipped set at Floor, 1080p, with the report as the instrument. A release build, because a debug profile moves the CPU side of every reading by an order of magnitude: cargo run -p standalone --release --example shot -- --presets presets --report. Record the machine header and every preset the block marks !. A clean run with no ! is a result worth recording too.
• Cross-check the flagged presets in the window. ritmolux.exe --tier floor at 1080p with audio playing, F3 on, and read fps and frame_ms p99 for each preset the report flagged, plus two it did not. Record both numbers side by side — the headless cost and what the screen actually did — with a scale column beside them: the number F3 prints after the tier (FLOOR 1.00). An integrated GPU resolves auto from its own row of the grid-scale table, so a reading without the scale cannot be compared with one taken on another class of GPU or at another pin (Configuration). Divergence between the two numbers is itself the finding, because it is the gap the report’s own caveat predicts and nothing in this repository has ever measured.
• Write the reading into NFR §1’s Floor line, dated and naming the machine and the grid scale, beside the newer integrated part’s reading already there. A miss is a result, not a failure: if a shipped preset cannot hold ≥ 60 fps at Floor on baseline hardware, that is the first real evidence this project has had on the question. It belongs here and in a backlog entry — never in a silent retune, and never in a hard gate written before the evidence it would be designed against. (Plan 0207 Phase 3, extracted at that plan’s close 2026-09-23.)