wassname 9c886428bf proj: measure_only kwarg + train.py always-on cos_in diagnostic ...

Vanilla arm now reports cos_in per step too (cosine of accumulated
Dr.GRPO grad with v_hack), as long as v_hack file is on disk. The
projection action only mutates the gradient when arm=projected;
vanilla just measures.

This makes Phase 2 (pilot scale) directly inform Phase 3: vanilla
cos_in trajectory says whether v_hack is even aligned with the GRPO
direction, before we burn 65h on the full sweep.

2026-05-25 11:50:41 +00:00

..

projected_grpo

proj: measure_only kwarg + train.py always-on cos_in diagnostic

2026-05-25 11:50:41 +00:00