Cap on the share of the window's total rise a SINGLE frame may carry. Feedback is
exponential in amplitude ⇒ linear in dB ⇒ its rise is distributed roughly evenly across
the window; a musical attack / re-articulation puts almost the whole rise in one hop
(the exact shape behind every one of issue #106's 72 live false positives). One half is
generous to real ramps (an even N-frame ramp puts 1/(N−1) per hop) and fatal to steps.
A side effect worth knowing: a 2-sample window can never pass (one hop IS the whole
rise), so confirmation always rests on at least three frames of actual ramp evidence.
Cap on the share of the window's total rise a SINGLE frame may carry. Feedback is exponential in amplitude ⇒ linear in dB ⇒ its rise is distributed roughly evenly across the window; a musical attack / re-articulation puts almost the whole rise in one hop (the exact shape behind every one of issue #106's 72 live false positives). One half is generous to real ramps (an even N-frame ramp puts 1/(N−1) per hop) and fatal to steps. A side effect worth knowing: a 2-sample window can never pass (one hop IS the whole rise), so confirmation always rests on at least three frames of actual ramp evidence.