Skip to content

Releases: kgraph57/evidentia

Release list

v1.1.0 — arXiv verification and Codex plugin packaging

v1.1.0 — arXiv verification and Codex plugin packaging Latest
Latest

Choose a tag to compare

View all tags
@kgraph57 kgraph57 released this 18 Jun 03:20
v1.1.0
7708ce9

Highlights

  • Adds deterministic arXiv ID verification alongside DOI, PMID, NCT, ISBN, and title-only citation handling.
  • Adds machine-readable lookup traces via lookupVerified and resolverOutcomes so agents can explain citation verdicts without guessing.
  • Adds --cache for stable, faster repeated registry checks.
  • Adds Codex Desktop plugin metadata and updates Claude/Codex medical fact-check skill guidance to prefer the local evidentia binary.
  • Expands regression coverage and benchmark cases for preprints, fabricated identifiers, and manual-review sources.

Verification

  • npm test
  • npm run build
  • live sample: arXiv:1706.03762 verifies as Tier 1 and a fabricated DOI is flagged as Tier 4.

Note: npm registry publication is still pending npm authentication; the attached tarball is the packed evidentia@1.1.0 artifact.

Assets 3
Loading

v1.0.0 — Evidentia becomes a runnable tool

v1.0.0 — Evidentia becomes a runnable tool

Choose a tag to compare

View all tags
@kgraph57 kgraph57 released this 12 Jun 08:57
v1.0.0
6b98a44
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

Evidentia is now a deterministic citation-verification engine (CLI + MCP server) plus the 15-criteria appraisal skill — not just a prompt.

Highlights

  • Verifies every citation against CrossRef, PubMed, OpenAlex, and ClinicalTrials.gov — no API key.
  • 4-tier classification that distinguishes "real paper, wrong DOI" from "this paper does not exist".
  • Catches what humans miss: on a real AI-generated answer it flagged a fabricated DOI, a PMID pointing to an unrelated paper, and a wrong year (75% fabrication rate).
  • Clinical trials & grey literature: resolves NCT trial IDs; never false-flags a guideline, book (ISBN), or other non-indexed source as a hallucination.
  • CLI (evidentia check, --fail-on-fabrication for CI), MCP server (claude mcp add), and a Claude Code plugin.
  • 46 unit + 7 live tests; a 19-case live benchmark (100% pass).

Install

Claude Code plugin

/plugin marketplace add kgraph57/evidentia
/plugin install evidentia@evidentia

CLI (once published to npm)

npx evidentia check article.md

From source

git clone https://github.com/kgraph57/evidentia.git
cd evidentia && npm ci && npm run build
node dist/cli.js check article.md

Pre-publication aid for writers, editors, and researchers — not clinical decision support.

See the full changelog.

Loading