Experiment 2 detail

Regime Detection, Round 2

Refinement pass on the phase-aware detector family, still fully separated from the operating branch.

Top Step 2 detector

phase_aware_v2_struct

Combined score -12.43

Manual transition reference

Low churn target

Step 24 regime source

simple_live_v2

Still untouched

Branch decision

Research only

Ready for dry validation, not promotion

Refinement logic

What changed

Core evidence

Detector	Agreement	Macro recall	Transitions	Avg segment days	Short segments <7d
simple_live_v2	47.44%	40.05%	1164	1.36	94.51%
phase_aware_v1	29.70%	33.71%	153	10.32	40.26%
phase_aware_v2_struct	51.26%	46.10%	47	33.10	2.08%
phase_aware_v2_active	52.34%	50.28%	93	16.90	22.34%

Current snapshot

Detector	Latest regime	Bar time	Close	7d return	Dist from 200MA
simple_live_v2	sideways	2026-03-15T15:00:00Z	2093.59	7.73%	2.54%
phase_aware_v1	bull	2026-03-15T15:00:00Z	2093.59	7.73%	2.54%
phase_aware_v2_struct	sideways	2026-03-15T15:00:00Z	2093.59	7.73%	2.54%
phase_aware_v2_active	sideways	2026-03-15T15:00:00Z	2093.59	7.73%	2.54%

Verdict

Best Step 2 candidate: phase_aware_v2_struct.
Main message: both Step 2 variants beat Step 1 structurally.
Decision: justify a bounded dry-trading validation branch, not a handoff into Step 24.
Generated: 2026-06-05T18:10:37Z