API reference

Authentication

All requests need an API key, sent as a bearer token. Create and manage keys on the developers page. Keep keys secret and server-side — never ship one in client code.

Authorization: Bearer scitrue_sk_live_…

Base URL: https://api.scitrue.ai/v1

Each fresh call spends one credit from your account; an identical repeat is served from cache for free. Fair-use rate limits apply.

Try it

Paste a key, pick a core operation, and send a real request.

Core operations

The three tools below are the product surface. Each returns evidence cards in one shared, flat shape (see Response fields), and every card carries a corpus_id you can hand to ask-paper.

POST/v1/verify

Verify a scientific claim against published evidence. Returns a verdict, an executive summary, evidence cards, and numbered references.

curl -X POST https://api.scitrue.ai/v1/verify \
  -H "Authorization: Bearer $SCITRUE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"claim": "Creatine improves memory", "k": 10}'

POST/v1/check-citations

Check whether cited sources support, overstate, contradict, or don't address their claims. Every checked citation carries a full evidence card.

curl -X POST https://api.scitrue.ai/v1/check-citations \
  -H "Authorization: Bearer $SCITRUE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"text": "Green tea improves alertness (Jones, 2018).\n\nReferences\nJones 2018, ..."}'

POST/v1/add-citations

Find real sources for uncited claims and return suggested sources (each a full evidence card), a citation-ready rewrite, and a numbered reference list its [n] markers map to.

curl -X POST https://api.scitrue.ai/v1/add-citations \
  -H "Authorization: Bearer $SCITRUE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"text": "Regular aerobic exercise increases hippocampal volume."}'

Ask a paper — the attribution follow-on

Ask-paper isn’t a fourth operation you invoke cold — it’s the follow-on for a paper the three tools already surfaced. The typical flow: call /v1/verify (or check / add), take the corpus_id off any evidence card, and pass it here to interrogate that source — Doğru Bilim fetches its full text for you. For a standalone paper (not from a prior call), pass a doi or the paper’s abstract / passages instead.

curl -X POST https://api.scitrue.ai/v1/ask-paper \
  -H "Authorization: Bearer $SCITRUE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"question": "What was the study design and the sample?",
       "context": {"corpus_id": "279946754",
                   "claim": "Creatine improves working memory"}}'

Response fields

Responses carry the same rich, structured evidence the Doğru Bilim app shows — so you can render a full report in your own product. Every evidence card is the same flat shape, whether it comes from /v1/verify (in evidence_cards), a /v1/check-citations result (in evidence_card), or a /v1/add-citations suggested source (in evidence_card). Each card includes:

• relevant_sentence — the key sentence the verdict rests on.
• source_excerpt — the exact paragraph it was drawn from (with source_excerpt_truncated).
• supporting_conditions / refuting_conditions — the validity conditions (population, dose, design caveats) under which the finding holds. Each is an object: { text, source, section, implied } — the condition, the verbatim quote it's grounded in, the section that quote came from, and whether it was implied rather than explicitly stated.
• study_type_label, evidence_strength — the design and how strong it is.
• credibility_tier, credibility_score — the journal’s reputation (SJR quartile).
• corpus_id — the paper’s Semantic Scholar id; hand it to /v1/ask-paper to chat with this source.
• stance, relation, source link / DOI, and badges.

/v1/verify adds an overall accuracy_label verdict, an executive_summary, numbered references, and a report_url to the full interactive report.

/v1/add-citations returns candidates (suggested sources grouped by claim), a reframed_text citation-ready rewrite with [n] markers, a numbered references list those markers map to (author, year, title, journal, link — identical to the app), and a proposal_id for a stage="refine" follow-up.

/v1/check-citations returns a results array — each with a verdict, reasoning, and the full evidence_card above.

/v1/ask-paper returns the answer, the verbatim quotes it relied on, and grounding — full_text (fetched the paper’s body), snippets, or supplied (answered from the text you passed).

Errors

• 401 — missing or invalid API key.
• 402 — out of credits. Sign in at scitrue.ai and click your credit balance to buy a pack.
• 422 — the input was understood but couldn’t be verified (e.g. not a scientific claim). The body carries an error message.
• 429 — rate limited; back off and retry.

Questions

For higher volume or licensing, contact scitrue.ai@gmail.com.