The dispatcher (master perf @0x4aa505-0x4aa588, from the raw disasm -- the
function Ghidra never decompiled), re-hosted at the gait seam in Simulate:
From Standing, arm the trn clip when the mech is TURNING (the binary's
operand is |angularVelocity| > 1e-4; at a standstill the turn rate is
exactly walkingTurnRate, so the test is turnDemand * walkingTurnRate),
the speed demand sits in the FULL sub-walk band [0, standSpeed], the
model carries a trn clip (turnCapable, real since 5.3.103), and the
wind-down debounce is clear.
BOTH channels arm on the same frame -- the lockstep weld. Arming only
the leg lets the body enter its next cycle frames apart and the two walk
cycles run permanently out of phase (the donor's rhythmic-gait-skip
finding). Each trn clip plays one figure and drops back to Standing via
the case-4 group, so a held turn re-arms clip after clip.
The debounce is the master perf's own shape: legResetLatch cleared at the
top of every frame (@0x4a9bff), set by the advancer's wind-down, honoured
here -- a walk that wound down THIS frame cannot re-enter as a turn until
the next.
LIVE (new pod_render_trn.conf: throttle 0, turn 0.5, exterior clips):
pos PINNED at (-641.043, 0, -243.722) across the whole run
yaw sweeping the full circle
spd oscillating +/-0.001347
That last number is the proof of authenticity: it is the trn clip's authored
root translation -- the foot shuffle. The mech turns by STEPPING, not by
having its yaw slid underneath it, and the stride-driven speed model shows
the steps.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
CydandClaude Fable 5
6b9a54089c