Drop the pending probe without judging it (pause, native-DPR change, disable). The caller keeps the current DPR as-is so the manager and renderer never diverge; no floor is learned.
Mark the in-flight probe as spanning a scene-content change (no-op when none is pending). Used when the caller decides to KEEP the probe across the change — a loop too slow to run a replacement experiment prefers a contaminated verdict to no verdict at all — so that the settle can be applied with the reduced confidence it deserves.
Evaluate the pending probe at timestamp against currentFPS.
Returns null when no probe is in flight. A non-pending verdict
clears the probe; the caller applies its consequences.
Quality gating: past the probe window, the verdict is only settled from a CLEAN sample — enough frames, enough span, no data-loading suppression. Judging a probe on 2 janky post-stall frames is how wrong floors get learned. While unclean, the probe keeps waiting up to 2× the window; past that it is voided as inconclusive (DPR stays, nothing learned).
Whether a probe is currently in flight.