Train Controlled No-Context RAPM

This is the controlled ablation for HIPSTER PM's recursive contextual state. It retains the same annual data, published lambda schedule, possession-weighted prior centering, value-conditioned aging prior, and exposure-gated cold-start prior as Value-Conditioned Aging HPM. It disables only the prior-season lineup context correction:

\[ C_{t-1}(U,V)=0. \]

Each season is still fit recursively, so the resulting player state is the proper no-context counterfactual rather than an after-the-fact translation of an older RAPM artifact. The frozen 2025-26 evaluation also uses zero context.

uv run nba-train-forward-centered-value-conditioned-aging-no-context-rapm \
  --through-season 2025-26

Outputs

artifacts/models/forward_centered_value_conditioned_aging_no_context_rapm/2025-26/

The immutable artifact uses the usual frozen possession, game-margin, team net-rating, and Pythagorean-win evaluation tables. Its metadata.json marks context_enabled: false, and season_context_models.joblib is intentionally an empty state.