feature_schema.json

{
  "bar_feature_names": [
    "open_rel",
    "high_rel",
    "low_rel",
    "close_rel",
    "volume_z_30m",
    "amount_z_30m"
  ],
  "feature_names": [
    "ret_5m",
    "ret_15m",
    "ret_30m",
    "rsi_5m",
    "rsi_15m",
    "rsi_30m",
    "vol_std_5m",
    "vol_std_15m",
    "vol_std_30m",
    "ma_dev_5m",
    "ma_dev_15m",
    "ma_dev_30m",
    "boll_z_5m",
    "boll_z_15m",
    "boll_z_30m",
    "macd_5m_15m",
    "macd_15m_30m",
    "cross_section_ret_rank_5m",
    "cross_section_vol_rank_5m",
    "cross_section_ret_rank_15m",
    "cross_section_vol_rank_15m",
    "cross_section_ret_rank_30m",
    "cross_section_vol_rank_30m",
    "market_ret_5m",
    "market_vol_5m",
    "market_ret_15m",
    "market_vol_15m",
    "market_ret_30m",
    "market_vol_30m",
    "hour_sin",
    "hour_cos",
    "dayofweek_sin",
    "dayofweek_cos"
  ],
  "feature_windows_minutes": [
    5,
    15,
    30
  ],
  "bar_size": "1m",
  "normalization": {
    "stats_file": "normalization_stats.npz",
    "bar": {
      "mean_key": "bar_mean",
      "std_key": "bar_std"
    },
    "features": {
      "mean_key": "feature_mean",
      "std_key": "feature_std"
    },
    "fit_scope": "train split only"
  },
  "leakage_constraints": [
    "Use only bars with timestamp <= decision timestamp t.",
    "All rolling, EMA and cross-section rank features must be computed causally.",
    "Do not refit normalization stats online or on validation/test/live data.",
    "Do not use future volume, future slippage, centered rolling windows or future universe membership."
  ]
}