Function step_response 

pub fn step_response(
    rise: &[CaptureImage],
    fall: &[CaptureImage],
) -> StepResponse

Measure a step response from two captured segments: rise starting at the last frame before the step up, fall starting at the last frame before the step down. Each segment’s own last frame is taken as its settled state.

Both segments should be the same length, because each is normalized against its own final frame: a segment that has not fully settled underestimates the total change and so settles early.

Equal windows do not make that bias cancel (Plan 0038 Phase 8 corrected the reverse claim, which had been written here). Cancellation would need both directions to be truncated by the same fraction, which is exactly what an { attack, release } pair is built not to do: at attack = 0.02 against a release = 0.5 the rise finishes in 80 τ and carries no bias at all, so the fall’s has nothing to cancel against and passes straight into StepResponse::ratio. That is not hypothetical — it is how this repo’s own asymmetric probe reported a fall of 61 frames where the settled answer is 69.

Equal windows remain the right default. They are just not a guarantee: gate on segment_settled before trusting either number.