plan: adaptive learning dynamics — 19 tasks across 5 phases
Implementation plan for unified LearningHealth system (all gems/pearls/novels):
- Phase A (A1-A4): ISV extension, LearningHealth module, training loop wiring,
HEALTH_DIAG logging
- Phase B (B1-B4): G2 uncertainty-gated CQL, G3 health-coupled tau,
G4 temp-continuous Expected SARSA, G5 adaptive gradient budget
- Phase C (C1-C4): P1 health-weighted PER priorities, P2/P3 subsumed by A3,
P4 temporal-coupled gamma
- Phase D (D1-D8): N1 temporal self-distillation, N2 Q-gap barrier,
N3 health-triggered plasticity, N4 CF curriculum, N5 information bottleneck,
N6 ensemble oracle, N7 contrarian override, N8 meta-Q collapse predictor
- Phase E (E1-E2): collapse-recovery smoke test + L40S deployment verification
Spec: docs/superpowers/specs/2026-04-20-adaptive-learning-dynamics-design.md
Target: WinRate >55% by epoch 20, Q-gap stays above 0.1 (no collapse).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
ab7567875a