Omitted raw outputs
===================
To keep this archive small, bulky raw run logs were left out of outputs/.
Everything omitted is regenerated by the pipeline (see README.md, sections 6-7).

Kept
  - benchmark_analysis/        final report, figures and summary tables
  - pid_flat_tuning_v2/        PID tuning study and selected gains
  - tensorboard/               training logs
  - */calibration_analysis/, */hpo/   calibration and HPO summaries
  - offpolicy-phase4a-v3/      final benchmark: per-episode results (episodes.csv, all
                               checkpoints) and, at the final checkpoint, metadata,
                               hparams.json and observation-normalisation stats
  - benchmark/                 PID benchmark results
  - seed 202, episode 0 trajectories (the data behind the report trajectory figures)

Omitted
  - per-episode trajectory CSVs for all other episodes
  - per-trial metadata.json / episodes.csv from calibration, search and validation runs
  - intermediate model checkpoints

Trained policy weights are NOT under outputs/. They are in ../models/ (see models/README.md).
