"""Shared, dependency-free evaluation primitives for Fold benchmarks.

Runners collect artifacts.  This package owns canonical finding identities,
ground-truth matching, metrics, and repeated-run stability so those semantics do
not drift between benchmark entry points.
"""

from .confirmation import summarize_confirmation_results
from .dispositions import compare_dispositions, validate_disposition_baseline
from .metamorphic import compare_metamorphic_findings
from .metrics import (
    aggregate_classification_metrics,
    classification_metrics,
    distribution_summary,
    paired_comparison,
)
from .performance import compare_performance_lanes, optimization_surface
from .project import (
    load_project_manifest,
    manifest_fingerprint,
    validate_project_manifest,
)
from .schema import validate_benchmark_case, validate_run_artifact
from .scoring import (
    negative_cases_from_universe,
    score_grouped_findings,
    score_labeled_cases,
)
from .stability import stability_summary

__all__ = [
    "aggregate_classification_metrics",
    "classification_metrics",
    "compare_dispositions",
    "compare_metamorphic_findings",
    "compare_performance_lanes",
    "distribution_summary",
    "load_project_manifest",
    "manifest_fingerprint",
    "negative_cases_from_universe",
    "optimization_surface",
    "paired_comparison",
    "score_grouped_findings",
    "score_labeled_cases",
    "stability_summary",
    "summarize_confirmation_results",
    "validate_benchmark_case",
    "validate_disposition_baseline",
    "validate_project_manifest",
    "validate_run_artifact",
]