Train Additive Profile-Prior RAPM

Run the controlled no-context experiment through the frozen target season:

uv run nba-train-forward-additive-profile-prior-rapm --through-season 2025-26

The command builds the same lagged player profiles used by the HPM x3 additive context terms, moves them into a residual player-prior ridge model, and fits no lineup context state. It writes the usual forward RAPM artifacts plus annual box-score residual models and regularization selections under:

artifacts/models/forward_additive_profile_prior_rapm/2025-26/

Evaluate the immutable 2023-24 through 2025-26 states without any replay-season refit:

uv run nba-evaluate-additive-profile-prior-rapm

The focused report is published under artifacts/models/analysis/additive_profile_prior_frozen/ and does not replace the shared leaderboard's latest-report pointer.

Run 10,000 paired, season-stratified game-block bootstrap draws against both the complete no-context player-prior control and full linear HPM x3 context:

uv run nba-bootstrap-additive-profile-prior-rapm