Aegis
Adaptive position-management subsystem — per-position state, Layer-0 invariant fires, and the Layer-1 detector audit log.
Aegis is the brain wrapping every open position. Layer 0 is the immovable safety floor (force-close on -2R, stop failure, sector crash, daily-DD breach); Layer 1 is the existing 9 detectors logged for later learning. Higher layers — adaptive parameters, the learned exit classifier, weekly LLM reflection, and detector bandit allocation — build on the data accumulating here. Aegis can ONLY tighten exits, never relax them.
Adaptive parameters
Per-(scanner_family × regime × parameter) discrete-grid posteriors. A grid value is PROMOTED once it has ≥30 samples AND its lower 90% CI on the head-to-head win-rate exceeds 0.5.
Decisions
Last 7 days. Layer-0 invariants represent hard force-closes; Layer-1 detectors are advisory.