chess-sim: RL pilot (ReinFlow-style PPO on SmolVLA), stopped after round 7: checks, logs, round-4 checkpoint

chess-sim: perturbed-teacher (DART) experiment: checkpoint, training log, paired evaluation

part 2 summary: cause of run-to-run noise not yet established

part 2 summary: correct the interpretation (inconclusive, not equivalence)

part 2: summary

part 2 arm B: training log

part 2 arm B: reports

part 2 arm B: checkpoint (weights, config, normalization)

part 2 arm A: training log

part 2 arm A: reports

part 2 arm A: checkpoint (weights, config, normalization)

pipeline check: weight identity

pipeline check: comparison and closed-loop tests

pipeline check: M0 processors and configs (weights identical to the baseline)

pipeline check: P0 processors and configs (weights identical to the baseline)