Dataset Selection

The repository defaults to polished_dataset and retains simplified_dataset as a compatibility option.

Polished schema

Each polished row provides four model inputs:

theta
theta_dot
tau_load
T

The target is theta_TE. Direction is resolved from the first-level forward or backward folder and is not added to the input tensor.

Command-line selection

Training, validation, smoke-test, visualization, split-export, and campaign entry points expose a --dataset selector:

python -B scripts/training/validate_training_setup.py `
  --config-path config/training/feedforward/presets/trial.yaml `
  --dataset polished_dataset

Use --dataset simplified_dataset to preserve the legacy five-feature contract.

Stage 1 campaign

Validate the prepared polished-dataset smoke campaign without training:

.\scripts\campaigns\cross_wave\run_polished_dataset_stage1_smoke_campaign.ps1 -PreflightOnly

Early-wave polished retraining

Validate the prepared 36-run early-wave polished-dataset campaign without training:

.\scripts\campaigns\cross_wave\run_polished_dataset_early_wave_parallel_training_campaign.ps1 -PreflightOnly

Launch the local early-wave batch only after confirming the active campaign state intentionally records the RCIM run as parallel on another workstation:

.\scripts\campaigns\cross_wave\run_polished_dataset_early_wave_parallel_training_campaign.ps1

Wave 5.2B polished offset and harmonic guided campaign

Validate the prepared 12-run Wave 5.2B polished-dataset campaign without training:

.\scripts\campaigns\wave_5_2\run_wave52b_offset_harmonic_guided_campaign.ps1 -PreflightOnly

Launch locally only after confirming the prepared campaign state is intended:

.\scripts\campaigns\wave_5_2\run_wave52b_offset_harmonic_guided_campaign.ps1

Remote launch delegates to the repository-owned remote campaign runner:

.\scripts\campaigns\wave_5_2\run_wave52b_offset_harmonic_guided_campaign.ps1 -Remote

Shape-gate loss v2 checkpoint-selection pilot

Validate the prepared one-run polished setpoint forward pilot without training:

.\scripts\campaigns\cross_wave\run_shape_gate_loss_v2_checkpoint_selection_pilot_campaign.ps1 -PreflightOnly

Run one-batch validation without launching training:

.\scripts\campaigns\cross_wave\run_shape_gate_loss_v2_checkpoint_selection_pilot_campaign.ps1 `
  -PreflightOnly `
  -RunOneBatchValidation

Launch locally only after confirming the prepared campaign state is intended:

.\scripts\campaigns\cross_wave\run_shape_gate_loss_v2_checkpoint_selection_pilot_campaign.ps1

Remote launch delegates to the repository-owned remote campaign runner:

.\scripts\campaigns\cross_wave\run_shape_gate_loss_v2_checkpoint_selection_pilot_campaign.ps1 -Remote

Causal offset / mean calibration pilot

Validate the prepared two-arm polished_dataset setpoint Fw pilot without training:

.\scripts\campaigns\cross_wave\run_causal_offset_mean_calibration_pilot_campaign.ps1 `
  -PreflightOnly `
  -RunOneBatchValidation

Launch the pilot on the remote workstation with:

.\scripts\campaigns\cross_wave\run_causal_offset_mean_calibration_pilot_campaign.ps1 `
  -Remote

Shape-gate loss v2 bounded TE Curve Verification screen

After the checkpoint-selection pilot is closed out, validate the bounded polished_dataset setpoint Fw verification-screen package with:

.\scripts\campaigns\track_2\run_shape_gate_loss_v2_bounded_track2_screen.ps1 `
  -PreflightOnly

Launch the bounded screen on the remote workstation with:

.\scripts\campaigns\track_2\run_shape_gate_loss_v2_bounded_track2_screen.ps1 `
  -Remote

Parallel shape-objective follow-up

Validate the prepared three-arm polished_dataset setpoint Fw follow-up without training:

.\scripts\campaigns\cross_wave\run_parallel_shape_objective_followup_campaign.ps1 `
  -PreflightOnly `
  -RunOneBatchValidation

Launch the follow-up on the remote workstation with:

.\scripts\campaigns\cross_wave\run_parallel_shape_objective_followup_campaign.ps1 `
  -Remote

Shape-objective bounded TE Curve Verification screen

After the shape-objective follow-up is closed out, validate the bounded polished_dataset setpoint Fw verification-screen package with:

.\scripts\campaigns\track_2\run_shape_objective_bounded_track2_screen.ps1 `
  -PreflightOnly

Launch the bounded screen on the remote workstation with:

.\scripts\campaigns\track_2\run_shape_objective_bounded_track2_screen.ps1 `
  -Remote

Shape-first training-rule distillation pilot

Validate the prepared two-arm polished_dataset setpoint Fw pilot without training:

.\scripts\campaigns\cross_wave\run_shape_first_training_rule_distillation_pilot_campaign.ps1 `
  -PreflightOnly `
  -RunOneBatchValidation

Launch the pilot on the remote workstation with:

.\scripts\campaigns\cross_wave\run_shape_first_training_rule_distillation_pilot_campaign.ps1 `
  -Remote

Shape-first distillation bounded TE Curve Verification screen

After the shape-first distillation pilot is closed out, validate the bounded polished_dataset setpoint Fw verification-screen package with:

.\scripts\campaigns\track_2\run_shape_first_distillation_bounded_track2_screen.ps1 `
  -PreflightOnly

Launch the bounded screen on the remote workstation with:

.\scripts\campaigns\track_2\run_shape_first_distillation_bounded_track2_screen.ps1 `
  -Remote

Causal offset bounded TE Curve Verification screen

After the causal offset / mean calibration pilot is closed out, validate the bounded polished_dataset setpoint Fw verification-screen package with:

.\scripts\campaigns\track_2\run_causal_offset_bounded_track2_screen.ps1 `
  -PreflightOnly

Launch the bounded screen on the remote workstation with:

.\scripts\campaigns\track_2\run_causal_offset_bounded_track2_screen.ps1 `
  -Remote

After normal campaign closeout, the separate operator-launched official TE Curve Verification Pipeline refresh evaluates the selected harmonic-profile registry candidates across global, Fw, and Bw. Run it locally with:

.\scripts\campaigns\wave_5_2\run_wave52b_te_curve_verification_refresh.ps1

Or delegate it to the remote campaign runner:

.\scripts\campaigns\wave_5_2\run_wave52b_te_curve_verification_refresh.ps1 -Remote

Reduced non-MMT cross-wave evaluation

The reduced cross-wave launcher keeps forward and backward surfaces separate across polished setpoints, simplified setpoints, and polished actual values. Print the six-cell plan without evaluating models:

.\scripts\campaigns\track_2\run_reduced_selected_track2_reports.ps1

Run the prepared evaluation locally or on the configured LAN workstation:

.\scripts\campaigns\track_2\run_reduced_selected_track2_reports.ps1 -Run
.\scripts\campaigns\track_2\run_reduced_selected_track2_reports.ps1 -Remote -Run

The launcher excludes MMT and global. It synchronizes required code, configuration, documentation, and selected model archives before remote execution, then retrieves the generated reports, matrix artifacts, and logs. Official promotion remains a separate curve-first review step.