Changelog
This page documents methodology changes, prompt updates, and model version changes for transparency. It is not a release log for code changes — see CHANGELOG.md for that.
2026-05-17 · v1 baseline
launch trust layer + accessibility · compliance foundation goal
- Initial v1 methodology published (claim extraction → claim-scope check → fact-check → verdict).
- Model: Anthropic Claude Opus 4.7 for fact-check; Haiku-class support for claim extraction and scope checks.
- Transcription: Deepgram Nova-3.
- Taxonomy: 123-entry bias/fallacy/rhetoric taxonomy (CC-BY-4.0).
- Claim-scope policy defined; hard refusals include private-individual harassment, doxxing, hate speech, CSAM, and extremism.
- Trust pages published: /about, /methodology, /privacy, /terms, /subprocessors, /accessibility, /contact, /pricing, and /faq.
- WCAG 2.2 AA baseline: skip-to-content, focus rings, 44px touch targets, prefers-reduced-motion, aria-live regions.
- DPIA completed (docs/dpia.md).
- AI-generated content disclosure added to all session views.