Skip to content

v1.2.0

Latest

Choose a tag to compare

@moonweave moonweave released this 08 Jun 14:03
30709ab

First release that ships the materials/physical-science accuracy work: scientific units, condition-qualified values, an OpenAlex abstract source, and a regression ship-gate. pip install ref-verify now matches main.

Added

  • ref-verify check-file for JSONL and CSV DOI/claim batch checks.
  • Fixture-backed numeric claim eval coverage for repeated-use workflows.
  • DOI-bound OpenAlex abstract fallback before Semantic Scholar and PubMed.
  • A CLI regression corpus and manual Live Smoke ship gate for release-readiness checks.
  • English and Korean scope guidance: what the tool verifies, what it does not verify, and how to interpret non-ACCEPT verdicts.

Fixed

  • Composite scientific units such as MV/m were misread as numerator-only units.
  • Numeric claim support for common physical-science units such as eV, Ω·cm, S/m, and MPa.
  • estimated to be <value> is treated as a reported value while predictive estimated to exceed frames stay conservative.
  • Comparative evidence such as >220 °C no longer accepts an exact 220 °C claim.
  • Physical measurement conditions such as 1.7 eV in the temperature range and 5 S/m at 1 kHz are allowed without relaxing count-claim scope guards.
  • Semantic Scholar 429 rate limits are separated into SOURCE_RATE_LIMITED and retried once before the source is marked unavailable.
  • Comma-clause splitting can bind a number to a subject across descriptive commas when no same-unit competing value is present.
  • Claim-side up to <value> comparators are recognized for percentage and unit/count claims while exact-claim guards stay conservative.
  • Clarified that npx skills add installs the agent skill but does not pip-install the Python CLI.

Full changelog: see CHANGELOG.md.