[OK] Verified
The paper exists and the cited title, authors, year, and journal match the registry record.
diff --git a/docs/index.html b/docs/index.html index a9d9b1a..3f1e842 100644 --- a/docs/index.html +++ b/docs/index.html @@ -1,91 +1,189 @@ -
- -Citation verifier · CLI · MCP · Skill · v1.1.0 MIT
-Deterministic registry checks against CrossRef, PubMed, OpenAlex, arXiv, and ClinicalTrials.gov. A 4-tier verdict — Verified, bibliographic mismatch, hallucination, or content review needed. Not a vibes checker. No API key.
-Install in 60 seconds ★ Star on GitHub
-Engine: npx evidentia check · Skill: npx skills add kgraph57/evidentia
$ npx evidentia check examples/inputs/ai-generated-answer.md - -Evidentia: 4 citations — 1 verified, 1 mismatch, 2 hallucinated (75.0% fabrication rate) - [OK ] doi:10.1136/bmj.i6583 — Paper exists and the cited metadata matches the registry record. - [HAL] doi:10.1056/nejmoa2105512 — DOI does not resolve in CrossRef or OpenAlex, and no matching paper was found. - [HAL] pmid:18768876 — Identifier resolves to a different paper than the one cited. - [MIS] doi:10.1002/14651858.cd012734 — Paper exists, but cited metadata disagrees with the record (year).-
Committed example — vitamin D / childhood infections. Four citations, one real.
A Lancet audit (Topaz et al., 2026) of biomedical papers found fabricated references in 1 in 2,828 papers in 2023, 1 in 458 in 2025 (about sixfold), and 1 in 277 in the first 7 weeks of 2026 — about a 10-fold rise from 2023, in step with AI writing tools. A fabricated DOI looks exactly like a real one. Evidentia resolves each identifier and tells you which is which.
Topaz et al., Lancet 2026;407(10541):1779–1781. doi:10.1016/S0140-6736(26)00603-3 · STAT, 7 May 2026 · PMC open-access subsetMost citation checkers stop at a shrug. Evidentia resolves the identifier and says why a citation is suspect.
-The paper exists and the cited title, authors, year, and journal match the registry record.
Registries cannot confirm it (book, guideline, title-only), or a real paper still needs a human or LLM to check whether it is used in the right context.
A real paper exists, but the DOI/PMID is wrong or the metadata disagrees — careless citation, or a fabricated identifier bolted onto a real title.
The identifier resolves to nothing, or to a completely different paper. The signature of AI-generated text.
Split on purpose: existence is deterministic. Honesty of use still needs judgment.
-Pure registry lookup. DOI, PMID, arXiv, NCT vs CrossRef, PubMed, OpenAlex, arXiv, ClinicalTrials.gov. No API key, no LLM, no hallucination of its own.
-npx evidentia check · claude mcp add evidentia -- npx -y evidentia-mcp
Claim ledger, then four named loops: engine, semantic, adversarial, correction. The 15-criteria rubric is one pass inside that pipeline. KILL or MAJOR cannot ship as A.
-npx skills add kgraph57/evidentia
The 15-criteria rubric is one pass inside the pipeline, not the product. Existence is the engine. Honesty is the skill. A fabricated DOI presented as real is a KILL — overall ≤ D, F if fabrication. That is the system working.
-Registry lookup. Retry once if unreachable; then unresolved, never a guessed hallucination.
One extra abstract lookup per Tier 1 cite. Primary outcome, population, direction of effect. Does not upgrade a T3 or T4.
Five lenses. KILL / MAJOR / MINOR / PASS. KILL or MAJOR cannot ship as A. Max 3 passes.
Cap 3. Then stop and report what is still open. Do not raise the letter while a T4 remains.
Worked example: vitamin D — claim ledger → engine → semantic → KILL.
No API key. MIT. Pick the path that matches how you write.
-# one-off, no API key -$ npx evidentia check your-article.md
Works because SKILL.md lives at skills/medical-fact-check/SKILL.md.
# Claude Code, Cursor, Codex, and other agents that load skills -$ npx skills add kgraph57/evidentia
$ /plugin marketplace add kgraph57/evidentia -$ /plugin install evidentia@evidentia
MCP, if you want the engine inside any agent: claude mcp add evidentia -- npx -y evidentia-mcp
Evidentia is a pre-publication aid for writers, editors, and researchers — not clinical decision support. It does not diagnose, treat, or replace professional medical judgment.
Evidentia is a protocol for medical writing before it is published. Quality is guaranteed by a gated loop, not by a scoring rubric. Existence is a registry lookup. Honesty is a skill. A fabricated identifier presented as real is a KILL.
+CLI + MCP engine · skills/medical-fact-check · MIT · not CDS
content
loop 5 → 2 Engine · cap 3
+$ npx evidentia check examples/inputs/ai-generated-answer.md + +Evidentia: 4 citations — 1 verified, 1 mismatch, 2 hallucinated (75.0% fabrication rate) + [OK ] doi:10.1136/bmj.i6583 — Paper exists and the cited metadata matches the registry record. + [HAL] doi:10.1056/nejmoa2105512 — DOI does not resolve in CrossRef or OpenAlex, and no matching paper was found. + [HAL] pmid:18768876 — Identifier resolves to a different paper ("Trafficking of antigen-specific + CD8+ T lymphocytes…") than the one cited. + [MIS] doi:10.1002/14651858.cd012734 — Paper exists, but cited metadata disagrees with the record (year).+
+ Committed vitamin D run. Four citations; one real. The Cochrane DOI stays T3. + Case study: claim ledger → engine → semantic → KILL. +
+npx evidentia check your-article.md +npx skills add kgraph57/evidentia+
引用検証 · CLI · MCP · Skill · v1.1.0 MIT
-CrossRef、PubMed、OpenAlex、arXiv、ClinicalTrials.gov への決定論的なレジストリ照合。判定は4段階 — 検証済み、書誌の不一致、ハルシネーション、内容レビューが必要。雰囲気で「怪しい」と言うチェッカーではない。APIキー不要。
- -エンジン: npx evidentia check · スキル: npx skills add kgraph57/evidentia
$ npx evidentia check examples/inputs/ai-generated-answer.md - -Evidentia: 4 citations — 1 verified, 1 mismatch, 2 hallucinated (75.0% fabrication rate) - [OK ] doi:10.1136/bmj.i6583 — Paper exists and the cited metadata matches the registry record. - [HAL] doi:10.1056/nejmoa2105512 — DOI does not resolve in CrossRef or OpenAlex, and no matching paper was found. - [HAL] pmid:18768876 — Identifier resolves to a different paper than the one cited. - [MIS] doi:10.1002/14651858.cd012734 — Paper exists, but cited metadata disagrees with the record (year).-
収録の実行例 — ビタミンDと小児感染症。引用4件のうち、実在は1件。
生物医学論文を対象にした Lancet の監査(Topaz et al., 2026)では、捏造文献は 2023年は2,828本に1本、2025年は458本に1本(約6倍)、2026年最初の7週間は277本に1本。2023年比で約10倍で、生成AIライティングの普及と時期が重なる。捏造DOIは本物と見分けがつかない。Evidentia は識別子を一つずつ解決して、どれが本物かを示す。
Topaz et al., Lancet 2026;407(10541):1779–1781. doi:10.1016/S0140-6736(26)00603-3 · STAT、2026年5月7日 · PMCオープンアクセス収載分多くの引用チェッカーはそこで止まる。Evidentia は識別子を解決し、なぜ怪しいのかを示す。
-論文は実在し、引用のタイトル・著者・年・誌名がレジストリ記録と一致する。
レジストリでは確定できない(書籍・ガイドライン・タイトルのみ)、または実在論文でも、正しい文脈で使われているかの確認が人間かLLMに残る。
実在する論文だが、DOI/PMID が誤っているかメタデータが食い違う。雑な引用か、本物のタイトルに捏造識別子を付けたもの。
識別子が何も解決しない、またはまったく別の論文を指す。AI生成テキストの典型。
分けているのは意図的。存在の確認は決定論。使い方の誠実さには判断が要る。
-純粋なレジストリ照合。DOI / PMID / arXiv / NCT を CrossRef、PubMed、OpenAlex、arXiv、ClinicalTrials.gov へ。APIキー不要、LLMなし、エンジン自身はハルシネーションしない。
-npx evidentia check · claude mcp add evidentia -- npx -y evidentia-mcp
まず主張台帳。つづいてエンジン、意味、敵対、修正の4ループ。15項目はその中の一工程。KILL または MAJOR は A では出せない。
-npx skills add kgraph57/evidentia
15項目のルーブリックはパイプラインの中の一工程であって、商品そのものではない。存在はエンジン、誠実さはスキル。実在として出した捏造DOIは KILL — 総合 ≤ D、捏造なら F。それが故障ではなく、仕組みが動いているということ。
-レジストリ照合。届かなければ1回だけ再試行。それでもダメなら unresolved。推測のハルシネーションは付けない。
Tier 1 だけ、抄録を1回余分に取る。主要アウトカム、対象集団、効果の向き。T3 や T4 は格上げしない。
レンズは5つ。KILL / MAJOR / MINOR / PASS。KILL または MAJOR は A では出せない。最大3パス。
上限3回。その後は止まって残件を書く。T4 が残っているあいだは評点を上げない。
APIキー不要。MIT。書き方に合う経路を選ぶ。
-# 単発・APIキー不要 -$ npx evidentia check your-article.md
SKILL.md が skills/medical-fact-check/SKILL.md にあるので、このコマンドで入ります。
# Claude Code、Cursor、Codex など、スキルを読むエージェント -$ npx skills add kgraph57/evidentia
$ /plugin marketplace add kgraph57/evidentia -$ /plugin install evidentia@evidentia
エンジンをエージェントに載せるなら MCP: claude mcp add evidentia -- npx -y evidentia-mcp
Evidentia は執筆者・編集者・研究者のための公開前支援であり、臨床判断支援ではありません。診断・治療・医療専門家の判断を代替しません。
Evidentia は、医学文章を公開する前の手順です。品質は宣伝用の採点表ではなく、ゲート付きのループで担保します。存在はレジストリ照合。使い方の誠実さはスキル。実在として出した捏造識別子は KILL です。
+CLI + MCP エンジン · skills/medical-fact-check · MIT · 臨床判断支援ではない
content
loop 5 → 2 エンジン · 上限 3
+$ npx evidentia check examples/inputs/ai-generated-answer.md + +Evidentia: 4 citations — 1 verified, 1 mismatch, 2 hallucinated (75.0% fabrication rate) + [OK ] doi:10.1136/bmj.i6583 — Paper exists and the cited metadata matches the registry record. + [HAL] doi:10.1056/nejmoa2105512 — DOI does not resolve in CrossRef or OpenAlex, and no matching paper was found. + [HAL] pmid:18768876 — Identifier resolves to a different paper ("Trafficking of antigen-specific + CD8+ T lymphocytes…") than the one cited. + [MIS] doi:10.1002/14651858.cd012734 — Paper exists, but cited metadata disagrees with the record (year).+
+ 収録済みの実行。ビタミンDと小児感染症。引用4件、実在は1件。Cochrane の DOI は T3 のまま(T4 に上げていない)。 + ケーススタディ: 主張台帳 → エンジン → 意味 → KILL。 +
+npx evidentia check your-article.md +npx skills add kgraph57/evidentia+