Skip to content

Release Candidate Freeze Review

This page records the public surfaces reviewed for 0.9.0rc1. The intent is to make the release candidate explicit: changes after this point should be bug fixes, documentation corrections, or additive extensions that do not break existing manifests or result consumers.

Package and Contract Versions

  • Python package version: 0.9.0rc1.
  • Git tag: v0.9.0-rc1.
  • Public output contract version: 0.9.0-rc1.

Stable Public Entry Points

  • CLI: lrdbench run, lrdbench validate, lrdbench validate-output, lrdbench list-suites, lrdbench list-metrics, and lrdbench list-estimators.
  • Estimator API: lrdbench.interfaces.BaseEstimator, lrdbench.interfaces.EstimatorMetadata, and lrdbench.schema.EstimateResult.
  • Author utilities: lrdbench.testing helpers for estimator metadata and smoke checks.
  • Output contract: lrdbench.output_contract.PUBLIC_OUTPUT_CONTRACT and configs/contracts/public_output_contract.json.

Manifest Schema

The release-candidate manifest surface is the schema exercised by the packaged smoke, public-small, and public-medium suites:

  • benchmark identity and mode;
  • records;
  • estimators;
  • metrics;
  • leaderboard;
  • report;
  • seeds;
  • optional uncertainty;
  • optional execution;
  • optional contamination and observational-source blocks for the modes that use them.

No manifest-field rename is planned before v1.0.0.

Result-Store and Report Columns

The required files and columns listed in the output contract are frozen for release-candidate review. Downstream tools should treat the contract as a minimum required set and allow documented dynamic columns such as metric__* and stratum__*.

No result-store or report-column rename is planned before v1.0.0.

Metric Names

The core metric names exposed by lrdbench list-metrics are frozen for release-candidate review:

  • bias
  • mae
  • rmse
  • coverage
  • coverage_error
  • ci_width
  • validity_rate
  • runtime
  • instability
  • preprocessing_sensitivity
  • estimate_drift
  • relative_degradation_ratio
  • validity_collapse
  • coverage_collapse
  • false_positive_lrd_rate
  • cross_estimator_dispersion
  • pairwise_estimator_disagreement
  • family_level_disagreement
  • parameter_variant_sensitivity
  • max_variant_drift

New metrics may be added after 0.9.0rc1, but existing metric names should not be repurposed.

Promotion to v1.0.0

The reviewed release-candidate surface was promoted to v1.0.0 without planned schema, metric, or output-column churn. The stable public output contract version is 1.0.0.

Known Pre-1.0 Caveats

The known limitations page remains the authority for scientific and implementation caveats; this freeze review covers public interface stability rather than claims of estimator superiority.