v0.7.0

• Add bioevidence review (check, agreement, adjudication-sheet, score, freeze)
  implementing the gold-standard protocol: strict annotation checks, Krippendorff's α with
  bootstrap interval and Cohen's κ (both cross-checked against reference implementations),
  adjudication of disagreements, scoring on the test split with hash-bound joins, and frozen
  manifests. It never produces labels.
• Add a blinded ClinVar expert-review kit (evaluation/clinvar_review/): all 95 variants where
  the validator and NCBI disagree plus 95 stratum-matched controls, an Excel workbook with
  dropdowns, a reviewer rubric, a private key, and an importer into the protocol format.
• CI runs ruff and mypy, and enforces a 95% test-coverage minimum (currently 98%).
• Add CONTRIBUTING, CODE_OF_CONDUCT (Contributor Covenant 2.1), SECURITY, issue forms and a
  pull request template.
• Add community/profiles/: contributed domain profiles whose example cases are built,
  validated and checked against expected outcomes in CI; _template/ to copy.
• Add a documentation site (MkDocs, strict link checking) published to GitHub Pages.
• The canine 0.3 implementation is also preserved at the canine-0.3 tag.
• No change to validation decisions or report format; both benchmarks reproduce 0.6.0 exactly
  apart from validator_version.

Also in this release: 0.6.0 (not published separately)

0.6.0 was never released on its own, so upgrading from 0.5.0 also brings:

• A second real-data case, ClinVar germline classifications: 5,026 variants from the 2023-09 release validated with a ClinVar-style profile, compared with NCBI's own review status (98–99% agreement) and with 2026-09 outcomes. Where the validator is stricter, classifications were about 4–5× more likely to be reclassified or in conflict three years later (1★ and 2★).
• docs/STANDARDS.md mapping the record model to ECO, Biolink 4.4.4, GA4GH VA-Spec 1.0.1 and PROV-O; ECO meanings on ExtractionMethod in the LinkML schema.

See the CHANGELOG for details.