1 oracle.pipelines.measurement_sensitivity_report

oracle.pipelines.measurement_sensitivity_report

Read a frozen endpoint-preparation audit without recomputing its metric.

The audit compares different observation preparations on the same selected trajectory. It is diagnostic evidence about measurement sensitivity, not an estimate of the benefit of a data-generation recipe. Raster changes, rebuilt roots, and mesh retessellation have distinct meanings and remain separate.

1.1 Classes

Name Description
MeasurementSensitivityReport Validated local audit projection with fixed source and assignment identity.
MeshSubdivisionSensitivity Effect of changing triangle representation after the original crop.
PreparationSensitivityRow One measured selected-depth treatment of a fixed rollout target.

1.1.1 MeasurementSensitivityReport

oracle.pipelines.measurement_sensitivity_report.MeasurementSensitivityReport(
    source_revision,
    assignment_sha256,
    rows,
    mesh,
    content_sha256,
)

Validated local audit projection with fixed source and assignment identity.

1.1.1.1 Attributes

Name Description
source_revision Frozen generation source revision declared by the audit assignment.
assignment_sha256 SHA256 of the frozen audit assignment bytes.
rows Validated selected-depth treatments; observations are correlated within case.
mesh After-crop tessellation controls, separate from preparation treatments.
content_sha256 SHA256 of every audit artifact validated in the same read transaction.

1.1.2 MeshSubdivisionSensitivity

oracle.pipelines.measurement_sensitivity_report.MeshSubdivisionSensitivity(
    case,
    original_faces,
    subdivided_faces,
    original_gain,
    subdivided_gain,
)

Effect of changing triangle representation after the original crop.

1.1.2.1 Attributes

Name Description
case Audit case identifier matching one fixed target and rollout.
original_faces Face count of the original cropped GT surface.
subdivided_faces Face count after four-way planar subdivision of every original face.
original_gain Root-normalized gain under the original triangle representation.
subdivided_gain Root-normalized gain under the subdivided triangle representation.

1.1.3 PreparationSensitivityRow

oracle.pipelines.measurement_sensitivity_report.PreparationSensitivityRow(
    case,
    stride,
    voxel_m,
    endpoint_gain,
    root_error_m2,
    final_error_m2,
    root_points,
    final_points,
    measurement_seconds_cpu,
    root_mode,
    selected_depth_raster_hw,
    source_depth,
    root_accuracy_m2,
    root_completeness_m2,
    final_accuracy_m2,
    final_completeness_m2,
    repeat_max_abs_delta,
    noop_endpoint_gain,
    noop_error_delta_m2,
)

One measured selected-depth treatment of a fixed rollout target.

1.1.3.1 Attributes

Name Description
case Audit case identifier; cases are not independent scene estimates.
stride Pixel stride used to unproject retained selected depth.
voxel_m Fusion voxel width in metres; root is rebuilt for this treatment.
endpoint_gain Finite root-normalized gain under this observation preparation.
root_error_m2 Prepared root point–mesh squared error in square metres.
final_error_m2 Prepared terminal point–mesh squared error in square metres.
root_points Target-crop root point count after preparation.
final_points Target-crop final point count after preparation.
measurement_seconds_cpu Measured endpoint computation time on CPU, including audit controls.
root_mode Root preparation mode; every retained treatment rebuilds the root.
selected_depth_raster_hw Height and width of the retained selected-depth raster.
source_depth Selected-depth storage/reload protocol, or None for older evidence.
root_accuracy_m2 Root point-to-mesh accuracy component, when retained by the artifact.
root_completeness_m2 Root mesh-to-point completeness component, when retained by the artifact.
final_accuracy_m2 Terminal point-to-mesh accuracy component, when retained by the artifact.
final_completeness_m2 Terminal mesh-to-point completeness component, when retained by the artifact.
repeat_max_abs_delta Maximum absolute difference across repeated endpoint measurements.
noop_endpoint_gain Measured gain for the root-to-root no-op control.
noop_error_delta_m2 Measured error difference for the root-to-root no-op control.

1.2 Functions

Name Description
measurement_sensitivity_identity Hash the retained audit inputs for cache invalidation on app reruns.
read_measurement_sensitivity Validate and project one frozen preparation-sensitivity summary.

1.2.1 measurement_sensitivity_identity

oracle.pipelines.measurement_sensitivity_report.measurement_sensitivity_identity(
    summary_path,
)

Hash the retained audit inputs for cache invalidation on app reruns.

This reads only small JSON artifacts, never the source mesh or rollout store. Validation still belongs to read_measurement_sensitivity.

1.2.2 read_measurement_sensitivity

oracle.pipelines.measurement_sensitivity_report.read_measurement_sensitivity(
    summary_path,
)

Validate and project one frozen preparation-sensitivity summary.

Every selected-depth row is bound to its raw JSON by SHA256. The assignment hash and source revision are retained, and the raw no-op/repeat controls must pass. The returned content digest is computed from the bytes actually parsed; a final full-content scan rejects a mixed or changed snapshot. The reader never reruns rendering, unprojection, or point–mesh scoring.

Back to top