Train Blended Prior RAPM

This model chooses a convex combination of frozen lagged-RAPM and aging-model prior means, jointly with the usual prior-centered RAPM penalty.

uv run nba-train-blended-prior-rapm \
  --season 2025-26 \
  --aging-run-id aging-2025-26-20260801T220356Z-4de5f001 \
  --lagged-run-id forward-lagged-rapm-2025-26-20260801T012439Z-3fe31ed6

The default lagged-weight grid is 0, 0.25, 0.5, 0.75, 1. A weight of 1 uses only lagged RAPM; 0 uses only the aging forecast. Selection pools chronological validation MSE across the target-season folds. The final regular holdout and playoff evaluation use the frozen first-1,044-game coefficient state.

Runs are written to:

artifacts/models/blended_prior_rapm/{season}/{run_id}/

blend_selection.parquet records every weight/lambda/fold result; blended_player_priors.parquet records the selected player means; and the frozen coefficient, holdout, playoff, metadata, and manifest artifacts follow the age-informed prior RAPM contract.