The damaged-copy twist path (@004b65f8 + @004b6510), reconstructed from the
raw decomp rather than the BT411 donor -- which drifted twice: its ease aged
by lastUpdate (+0x14) where the binary reads lastPerformance (+0x10), and it
calls the joint-write helper from the copy sim where 1995 does not (port-era
addition, dropped). The extrapolator predicts targetTwist = twistAtUpdate +
twistRate * elapsed on two time bases (settled = this frame's window, slewing
= the command window), and the copy sim snaps when settled or eases by the
remaining-time form dt/((stamp - now) + dt) -- landing exactly when the clock
reaches the command stamp. The decomp's DAT_0052140c divide is just
Time::operator- inlined; no hand conversion existed.
Members carved from the dynamicsState reserve (16->12): targetTwist @0x218,
twistAtUpdate @0x21C, twistRate @0x238, lastUpdateTime @0x254. The ctor now
registers the copy Performance on replicants (PTR @00510c1c); [T2] the gate
mirrors the binary's simulationFlags 0xC/0x100 pair with the instance test,
same as the whole watcher family, until the stream builders that seed those
bits are reconstructed. Feeders (Read/WriteUpdateRecord) stay staged --
zero-init eases a copy to centre, correct-by-vacuity in single-pod.
Stubs: 16 across 10 files. Regression: clean mission soak on the rig
(no faults, wheel turning, pacing steady).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
CydandClaude Fable 5
ea99e8987f