Wave 5.2 Frozen-Baseline Residual Replay

Replay frozen Wave 5.2 baselines over canonical dataset splits.

This non-training workflow resolves four accepted polished-setpoint ONNX baselines, reconstructs their original train/validation/test datasets, and emits per-curve residual metrics compatible with the Wave 5.2 MMT residual-explanatory diagnostic.

class scripts.reports.analysis.build_wave52_frozen_baseline_residual_replay.BaselineReplaySpecification(candidate_id, surface, direction_label, architecture_class, registry_path, reference_inventory_path)[source]

Bases: object

Store one frozen baseline replay contract.

Parameters:

  • candidate_id (str)

  • surface (str)

  • direction_label (str)

  • architecture_class (str)

  • registry_path (Path)

  • reference_inventory_path (Path)

candidate_id: str
surface: str
direction_label: str
architecture_class: str
registry_path: Path
reference_inventory_path: Path
class scripts.reports.analysis.build_wave52_frozen_baseline_residual_replay.ResolvedBaseline(specification, model_entry, training_config, registry_run_instance_id, archive_run_instance_id, registry_archive_run_aligned)[source]

Bases: object

Store validated baseline provenance and runtime objects.

Parameters:

  • specification (BaselineReplaySpecification)

  • model_entry (ExportedModelEntry)

  • training_config (dict[str, Any])

  • registry_run_instance_id (str)

  • archive_run_instance_id (str)

  • registry_archive_run_aligned (bool)

specification: BaselineReplaySpecification
model_entry: ExportedModelEntry
training_config: dict[str, Any]
registry_run_instance_id: str
archive_run_instance_id: str
registry_archive_run_aligned: bool
scripts.reports.analysis.build_wave52_frozen_baseline_residual_replay.parse_arguments()[source]

Parse command-line arguments.

Return type:

Namespace

scripts.reports.analysis.build_wave52_frozen_baseline_residual_replay.resolve_project_path(path_value)[source]

Resolve one repository-relative path.

Parameters:

path_value (str | Path)

Return type:

Path

scripts.reports.analysis.build_wave52_frozen_baseline_residual_replay.format_project_path(path_value)[source]

Return a stable repository-relative path when possible.

Parameters:

path_value (str | Path)

Return type:

str

scripts.reports.analysis.build_wave52_frozen_baseline_residual_replay.load_yaml_dictionary(path_value)[source]

Load one YAML dictionary.

Parameters:

path_value (str | Path)

Return type:

dict[str, Any]

scripts.reports.analysis.build_wave52_frozen_baseline_residual_replay.write_yaml(path_value, payload)[source]

Write one stable YAML artifact.

Parameters:

  • path_value (str | Path)

  • payload (dict[str, Any])

Return type:

Path

scripts.reports.analysis.build_wave52_frozen_baseline_residual_replay.write_csv(path_value, row_list)[source]

Write one stable CSV artifact.

Parameters:

  • path_value (str | Path)

  • row_list (list[dict[str, Any]])

Return type:

Path

scripts.reports.analysis.build_wave52_frozen_baseline_residual_replay.format_float(value)[source]

Format one finite float for machine-readable output.

Parameters:

value (float)

Return type:

str

scripts.reports.analysis.build_wave52_frozen_baseline_residual_replay.normalize_surface(surface_value)[source]

Normalize a configured surface to the repository ONNX surface label.

Parameters:

surface_value (str)

Return type:

str

scripts.reports.analysis.build_wave52_frozen_baseline_residual_replay.build_baseline_specification_list(config_dictionary)[source]

Build and validate configured baseline specifications.

Parameters:

config_dictionary (dict[str, Any])

Return type:

list[BaselineReplaySpecification]

scripts.reports.analysis.build_wave52_frozen_baseline_residual_replay.resolve_frozen_baseline(specification)[source]

Resolve one registry-backed archived ONNX baseline.

Parameters:

specification (BaselineReplaySpecification)

Return type:

ResolvedBaseline

scripts.reports.analysis.build_wave52_frozen_baseline_residual_replay.build_permissive_shape_thresholds()[source]

Build thresholds that retain metrics without making promotion claims.

Return type:

ShapeGateThresholds

scripts.reports.analysis.build_wave52_frozen_baseline_residual_replay.build_candidate_metric_entry(resolved_baseline, curve_sample, angular_position_deg, target_curve_deg, prediction_curve_deg)[source]

Build one shape-metric-compatible candidate entry.

Parameters:

  • resolved_baseline (ResolvedBaseline)

  • curve_sample (dict[str, Any])

  • angular_position_deg (ndarray)

  • target_curve_deg (ndarray)

  • prediction_curve_deg (ndarray)

Return type:

dict[str, Any]

scripts.reports.analysis.build_wave52_frozen_baseline_residual_replay.evaluate_baseline_split(resolved_baseline, split_name, split_dataset, session, harmonic_order_list, mdn_playback_channel)[source]

Evaluate one frozen baseline over one canonical dataset split.

Parameters:

  • resolved_baseline (ResolvedBaseline)

  • split_name (str)

  • split_dataset (Any)

  • session (onnxruntime.InferenceSession)

  • harmonic_order_list (list[int])

  • mdn_playback_channel (str)

Return type:

list[dict[str, Any]]

scripts.reports.analysis.build_wave52_frozen_baseline_residual_replay.update_p95_metric_values(row_list)[source]

Populate per-candidate, per-split P95 percentage-error values.

Parameters:

row_list (list[dict[str, Any]])

Return type:

None

scripts.reports.analysis.build_wave52_frozen_baseline_residual_replay.build_split_coverage_row_list(residual_row_list)[source]

Build per-candidate split and direction coverage evidence.

Parameters:

residual_row_list (list[dict[str, Any]])

Return type:

list[dict[str, Any]]

scripts.reports.analysis.build_wave52_frozen_baseline_residual_replay.build_baseline_manifest(resolved_baseline_list)[source]

Build machine-readable frozen baseline provenance.

Parameters:

resolved_baseline_list (list[ResolvedBaseline])

Return type:

dict[str, Any]

scripts.reports.analysis.build_wave52_frozen_baseline_residual_replay.build_aggregate_summary_row_list(residual_row_list)[source]

Summarize residual metrics per candidate and split.

Parameters:

residual_row_list (list[dict[str, Any]])

Return type:

list[dict[str, Any]]

scripts.reports.analysis.build_wave52_frozen_baseline_residual_replay.run_self_test(output_directory, harmonic_order_list)[source]

Run a bounded synthetic metric and schema self-test.

Parameters:

  • output_directory (Path)

  • harmonic_order_list (list[int])

Return type:

Path

scripts.reports.analysis.build_wave52_frozen_baseline_residual_replay.build_report(report_path, run_instance_id, config_path, output_directory, resolved_baseline_list, split_coverage_row_list, aggregate_summary_row_list, residual_row_count)[source]

Generate the residual replay analytical report.

Parameters:

  • report_path (Path)

  • run_instance_id (str)

  • config_path (Path)

  • output_directory (Path)

  • resolved_baseline_list (list[ResolvedBaseline])

  • split_coverage_row_list (list[dict[str, Any]])

  • aggregate_summary_row_list (list[dict[str, Any]])

  • residual_row_count (int)

Return type:

Path

scripts.reports.analysis.build_wave52_frozen_baseline_residual_replay.run_replay(config_path, output_directory_override, run_self_test_first)[source]

Run the full non-training frozen baseline residual replay.

Parameters:

  • config_path (Path)

  • output_directory_override (Path | None)

  • run_self_test_first (bool)

Return type:

list[Path]

scripts.reports.analysis.build_wave52_frozen_baseline_residual_replay.main()[source]

Run the command-line entry point.

Return type:

None