Train Forward Context-Reattributed HPM
This recursive candidate carries a controlled fraction of each completed season's player-projectable context into the next-season player prior, while leaving the untransferred context as the portable composition-plus-matchup offset:
uv run nba-train-forward-context-reattributed-hpm \
--through-season 2025-26 \
--context-reattribution-weight 0.5
For a handoff weight \(\rho\), the season-to-season contract is
\[
\mu_{t+1}=\mu^{\mathrm{aging/cold}}_{t+1}+\rho\gamma_t,
\qquad
C^{\mathrm{residual}}_t=C_t-\rho X\gamma_t.
\]
The same \(\rho\) is added to the player prior and removed from the frozen context offset. This prevents double counting. The initial exemplar uses 0.5; the next validation pass will select \(\rho\) using expanding historical seasons only.