Round 6 uses the Round-3 full reproducibility package as its immutable base.

Latency sensitivity:
  python round6_latency_build.py --delay {10|20|30|40} --output <dir>
  then run baselines/run_baselines.py + baselines/augment_assessments.py and analysis/analyze_baselines.py on the generated development and confirmation inputs.

Resource sensitivity:
  python round6_resource_driver.py --input <round3 causal input> --input-role <role> --input-manifest <manifest> --output <dir> --config-name <name> [--pb ... --emin ... --emax ... --pr ... --re ...]

All headline results remain the frozen 20-min baseline. Sensitivities are not used for tuning.

Post-freeze scoring-definition sensitivity (frozen DMPC replays only):
  python scoring_definition_sensitivity.py --output-dir .
This relabels stored trajectories at scoring thresholds 0.005, 0.01, 0.02 and 0.03 pu and over 30-min, 60-min and full active windows. It does not rerun or retune the controller and does not change the physical executor or synthetic obligation.
