mark-spec-fram-feat-fork

When to use

• A latest-sample debug run shows strong factor signal but raw pre-Bayes labels are clearly mismatched with multi-timeframe evidence.
• Example pattern: latest signal is bullish with strong HTF alignment, but raw build_frame_features labels still come out bear/hostile and force observe_only.
• Use especially in ~/projects-ict-engine/ict-engine for structure_ict / expansion_manipulation work.

Workflow

1. Prove the issue is upstream labeling, not factor weakness.

   • Run factor-pipeline-debug first.
   • Check latest_signal, raw_pre_bayes_labels, filtered_pre_bayes_labels, gating_status, and raw_label_trace.
   • If factor signal is still strong but raw regime/liquidity are wrong, treat build_frame_features as the target surface instead of blindly mutating factor params.

2. Add a narrow market-aware fork at the frame-feature layer.

   • Extend FrameFeatures with:
     • market: Option
   • Keep build_frame_features(candles) as the baseline generic builder.
   • Add:
     • infer_market_from_symbol(symbol: &str) -> String
     • build_frame_features_for_market(candles: &[Candle], market: Option<&str>) -> Result
   • In the market-aware wrapper, call the generic builder first, then overwrite only the labels that need market-specific handling.