The move helped (>= improvement) — keep the new DPR.
The move did not help — revert to previousDPR and floor probedDPR.
No clean sample arrived within the extended window (few frames, short span, or data-loading jank the whole time) — the experiment is inconclusive. The DPR stays as-is and NOTHING is learned: no revert, no floor.
Probe window still open — keep gathering samples, decide nothing.