Sundog Balance
A cart-pole workbench where the hidden body angle is denied. Phase 10 confirms shadow-derived control inside the mapped lighting and delay envelope, with overhead-light and high-delay degradation kept visible.
What This Slice Tests
The hidden target is the pole angle from upright. The exposed signal is the cast shadow: endpoint, length, residual from the upright-shadow target, and recent motion. The naive lane uses the shadow residual directly; the Sundog lane gates residual and residual-velocity control by shadow confidence; the oracle lane uses privileged true angle feedback.
The immediate question is diagnostic, not promotional: does the shadow projection move coherently enough with force and pole state to justify a proper SCAN/SEEK/TRACK controller? The first Sundog prototype and replay harness make that question reproducible; the recovery trace separates post-impulse recovery from merely delaying a fall. Phase 9 exposes where the shadow should not be trusted; Phase 10 confirmed the bounded operating envelope under repaired verdict criteria.