KB: peer coupled-motion architecture -- completion plan + session landings (task #50)

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

context/open-questions.md

@@ -610,3 +610,22 @@ register.  ⚠ The audit also flags the damage-economy item as SELF-CONTRADICTOR
 ## Key Relationships
 - Feeds from: every subsystem/render topic (their deferral notes collect here).
 - Gating master: [[source-completeness]] (the missing BT source).
+
+## Peer coupled-motion architecture — COMPLETION PENDING (task #50 tail, 2026-07-14)
+The user's speed-change peer glitch is the last visible MP motion artifact. Root
+cause is ARCHITECTURAL and decomp-confirmed: the original peer (FUN_004ab430 ->
+FUN_004ab1c8) drove position FROM the clip's root travel between records (feet
+locked to ground by construction) + pose-sync offset decay; our port dead-reckons
+position from velocity while legs run on commanded speed — they disagree exactly
+during accel/decel. The PEER half is implemented env-gated (BT_ROOT_POS=1,
+mech4.cpp replicant branch) but measured worse alone: with sparse records the
+master's VELOCITY mirror under-sends for a gait-driven peer (11u snaps); with
+dense records the two authorities tug-of-war (path/net 1.86). COMPLETION NEEDS
+THE MASTER HALF: (1) master send-gate linear mirror driven by the same gait
+travel (its own channel-B IntegrateMotion projection, as the binary's), so
+records model the peer; (2) then sparse authentic deadband cadence + offset
+decay. All fixed-and-committed groundwork: exact quaternion integration
+(a8eb8a4), incremental peer heading (d78e77b), clock guard (93456be), commanded-
+speed cadence feed (6374efc), Abs-macro fix (7615ecd). Speed model decomp-settled
+(2c6db6a): analog throttle, '5 speeds' false, throttleState@0x4a4 is actually the
+fall-surface material (RENAME PENDING).