vertebrae.results

Structured benchmark results.

Classes

`ExtractorResult`	Result data for one evaluated extractor.
`BenchmarkResult`	Aggregated result for a benchmark run.

Functions

`benchmark_result_columns`([metric_names, observed_columns])	Return deterministic canonical columns for benchmark and monitoring rows.
`null_benchmark_result_row`([columns])	Return a null row matching the canonical result schema.

Module Contents

vertebrae.results.benchmark_result_columns(metric_names=None, observed_columns=None)[source]

Return deterministic canonical columns for benchmark and monitoring rows.

Parameters:

metric_names (Optional[List[str]])
observed_columns (Optional[List[str]])

Return type:

List[str]

vertebrae.results.null_benchmark_result_row(columns=None)[source]

Return a null row matching the canonical result schema.

Parameters:: columns (Optional[List[str]])
Return type:: Dict[str, Any]

class vertebrae.results.ExtractorResult[source]

Result data for one evaluated extractor.

name[source]: Extractor name.

extractor_type[source]: Extractor family/type metadata.

metrics[source]: All normalized metric results, including overlap when enabled.

primary_metric_name[source]: Name of the metric used for ranking.

stability[source]: Optional stability-analysis summary.

separatix[source]: Optional Separatix diagnostic summary.

embedding_metadata[source]: Metadata for the embedding artifact.

runtime[source]: Runtime timing metadata by benchmark stage.

warnings[source]: Warnings produced during evaluation.

recommendation[source]: Recommendation label for this extractor.

label_view[source]: Label-view metadata for the scoring target.

target_view[source]: Target-view metadata for the scoring target.

weakest_class[source]: Class with the lowest per-class score when available.

weakest_class_score[source]: Score for weakest_class when available.

resource_profile[source]: Optional measured extraction and footprint profile.

to_dict()[source]

Serialize the extractor result to a JSON-safe dictionary.

Returns:: JSON-compatible result data.
Return type:: Dict[str, Any]

property overlap: vertebrae.scoring.metrics.MetricResult | None[source]

Return the overlap metric result without duplicating serialized state.

Return type:: Optional[vertebrae.scoring.metrics.MetricResult]

property primary_score: float[source]

Return the aggregate score used for ranking.

Return type:: float

class vertebrae.results.BenchmarkResult[source]

Aggregated result for a benchmark run.

dataset_summary[source]: Summary of the labeled dataset.

extractor_results[source]: Per-extractor results.

recommendations[source]: Practitioner-facing benchmark recommendations.

metadata[source]: Reproducibility metadata for the run.

to_dict()[source]

Serialize the benchmark result to a JSON-safe dictionary.

Returns:: JSON-compatible benchmark data.
Return type:: Dict[str, Any]

ranked_results()[source]

Return extractor results sorted by the selected aggregate metric.

Returns:: Sorted extractor results.
Return type:: List[ExtractorResult]

quality_cohort(tolerance=None)[source]

Return candidates within an absolute primary-score tolerance of the best.

Parameters:: tolerance (Optional[float])
Return type:: List[ExtractorResult]

to_dataframe(*, include_invalid=False)[source]

Convert benchmark results into a pandas DataFrame.

Parameters:: include_invalid (bool) – Whether results with an invalid primary aggregate should remain in the table with a null rank. The default preserves the ranked, valid-only benchmark view.
Returns:: A pandas DataFrame with one row per extractor.
Return type:: Any

save_json(path)[source]

Save the benchmark result as JSON.

Parameters:: path (str)
Return type:: None

save_markdown(path)[source]

Save the benchmark result as a Markdown report.

Parameters:: path (str)
Return type:: None