[bdbarnett]

Three moves that still stepped now walk over 20 ms: Fuzz's Mix to 0 and back, DeEsser's Range to 0 and back, and Tremolo's Depth to 0 and back.

Fuzz cut to its source at Mix 0 and cut back in. It now walks to the dry before the port switches and walks back in from it. The germanium chain has no mixer, so it walks on its stages' own mix; at an off-centre Bias it still comes back at once, because a walk would let the shapers' answer to silence through (6847 LSB out of silence at patch 2, which TheBypassComesBackAsBuilt catches). DeEsser's Range 0 cut to its input adapter and played a 256-frame lump of the old graph on the way back; it now walks the same way, with its Splitters dropping what they held and its voices primed afresh. Tremolo needed no change of its own: AUDIODSP_PIN moves to ec82463, where audiomath.Multiply walks a mix move (audiodsp#411). The pin's own commit records what else moved beneath the effects.

How I checked: renders of a 110 Hz triangle through each move, plotted against an instance that never moved. The steepest step after the move, over the triangle's own, went from 20.95 to 1.07 (Tremolo Depth to 0, 48 kHz) and from 31.6 to 1.0 (22.05 kHz). Fuzz's and DeEsser's plots show a 20 ms walk where there was a cut; the step ratio there is dominated by the fuzz's own edges, so the plots are the evidence. In the lifecycle matrix, Fuzz's 36 Mix 0 step cells, DeEsser's two Range 0 steps at 22.05 kHz and RingMod's 176 Mix step cells are green. Fuzz's walk-in differs from a fresh instance while its output pole settles (a few hundred LSB just after the walk, under 20 within 100 ms), and DeEsser's by one LSB in the block after the walk; both are declared in lifecycle_known.py. tools/scoped_tests.py class:Fuzz and class:DeEsser, tests/test_level_ramp.py, validate_api.py and flake8 pass, and the lifecycle matrix for the four classes prints the same lines on CPython, MicroPython and CircuitPython. That last check caught a plain mix assignment that landed on the Python-level Waveshaper as an attribute, so on CPython the shapers stayed wet through the walk; the walk now goes through set() there too. The full suite is left to CI.

What still steps is in the roadmap bullet: filter corners, an effect built at Mix 0 coming in for the first time (it cuts in, as every effect that walks to its bypass does today), and crossfaded patch changes. docs/how-it-works/control-smoothing.md says the same.