SirStone 0d35646dc9 feat(PPO_Bot): deterministic eval + fix logStd warm-start ...

- actorForward: deterministic param, uses mean-only when PPOB_EVAL_ONLY=1
  (eval was adding unit Gaussian noise to every action — unreliable scores)
- warm_start.py: log_std initialized to -1.0 (std≈0.37) instead of copying
  snapshot values (were 2.27-4.68 → std 9-108, completely drowning signal)
- training.env: LOG_STD_CEILING 0.0→-0.5 (cap exploration at std≈0.6)

2026-08-20 15:18:07 +02:00

..

curriculum_retrain.sh

feat(PPO_Bot): bot-relative bullets + scan staleness (STATE_DIM=57)

2026-08-20 14:27:28 +02:00

FIRE_CAMPAIGN_RESULTS.md

PPO_Bot Fire campaign: 3787/3789 wins (99.95%); frozen eval 500/500

2026-08-19 04:00:47 +02:00

generalist_train.sh

tune(training): 20-round chunks, 50 eval rounds, 30k total rounds

2026-08-20 14:38:09 +02:00

mixed_train.sh

feat(PPO_Bot): bot-relative bullets + scan staleness (STATE_DIM=57)

2026-08-20 14:27:28 +02:00

run.sh

cap terminal reward scale (bounded score bonus) and restore single-battle campaigns

2026-08-19 03:50:14 +02:00

RunTraining.java

fix(PPO_Bot): SIGSEGV crash fixes + static buffers for thread safety

2026-08-20 14:23:08 +02:00

training.env

feat(PPO_Bot): deterministic eval + fix logStd warm-start

2026-08-20 15:18:07 +02:00