Releases
v0.4.0
Compare
Sorry, something went wrong.
No results found
Added lower-is-better --max gates, reusable YAML policies, coverage/sample gates, provenance detail, and direction-aware recommendations.
Added configurable group_by, stratified sampling, dataset manifests, CSV/JSON/XLSX/Parquet import hooks, PII/secret redaction, near-duplicate linting, and a starter benchmark manifest.
Added claim-support, citation-span, context diversity/redundancy, rank-sensitivity, unanswerable-correctness, token throughput, judge agreement, and optional embedding similarity metrics.
Added multi-run trend summaries with bootstrap intervals, local regression bisect, JUnit/SARIF output, side-by-side baseline samples, group heatmap data, and richer probe output.
Added explicit missing-environment diagnostics, custom streaming token paths/done markers, async-compatible adapter calls, judge prompt fingerprints, and circuit-breaker limits.
Extended the reusable GitHub Action and CI smoke gates with policy inputs, HTML/XML/SARIF artifacts, dataset lint, manifests, and trend output.
You can’t perform that action at this time.