public class AdaptivePerformanceStableFrameModeProvider : IAdaptivePerformanceModeProvider

On mode start the provider reads the AdaptiveFramerate scaler's bounds and pins Time.maximumDeltaTime to — a physics-stability clamp so a hitch can never stretch a single physics step past the budget.

The provider then forces vsync off (QualitySettings.vSyncCount to 0) and takes ownership of framerate: vsync's snap-to-refresh-divisor behavior turns a missed deadline into a large frame-time spike, exactly the variance this mode fights. It force-enables the AdaptiveFramerate scaler, starts Application.targetFrameRate at MaxBound, and lets the scaler participate in the indexer's cost-based adjustment so the framerate seeks the highest stably-reachable rate (MinBound is the floor). The original vsync count, framerate, and maximum delta time are all restored on mode end.

Each frame it produces a performance action and adopts whichever signal asks for the most aggressive change. A headroom guard runs on every frame regardless of vsync: it asks for a decrease when CPU/GPU work climbs toward the presented frame interval, which is one load bump from a dropped frame. Alongside it a variance signal runs, whose source depends on whether presentation is vsync-locked. When it is (mobile always is — its compositor paces frames at the hardware level regardless of QualitySettings.vSyncCount), wall-clock frame time quantizes to the display cadence and its variance goes blind until a frame drops, so the provider measures variance on CPU/GPU work time. When it isn't, wall-clock frame time is authoritative and its variance is measured directly. Both paths also honor the MinBound floor via AdaptivePerformanceStableFrameModeProvider.TargetFrameTimeVarianceMs's companion lower-bound check.