Design partners · Reliability API

What it is

A score, and every claim behind it.

Ask for a subject by name. You get a score from 0 to 100 for each kind of claim (factual, promise, guidance, prediction) with a range and a confidence level, and each claim it rests on: the quote, the source, the evidence and the verdict.

33public figures and organisations, each returned in under a second

28,548claims extracted from what they said

2,653of them checked and given a verdict

Dataset v1.1.2, 10 November 2026. The claims in the browser are live and can differ by a few verdicts corrected since.

How it works

The same five steps for every subject.

Read. A random sample of what the subject has said in public, spread across the years.
Extract. Checkable claims, each quote matched word for word to its source.
Fix the test. What would count as true or false is written down before any research.
Research. Dated evidence for and against, preferring sources independent of the subject.
Judge and score. A model judge (Claude Opus) gives the verdict. A second judge from another model family (OpenAI GPT) reads the same claim and evidence independently; where it disagrees, the claim is flagged. Too little evidence gives “unresolvable”, never a guess.

How it works, in full · Method and measurements

Read with care

What to expect.

Many claims end unresolvable. They are left out of the score, not counted against anyone.
A score describes a sample of what the subject said, not everything.
Two judges agreeing is consistency, not accuracy. What has been measured is on the method page.
A new subject takes 10 to 17 minutes. A stored one comes back at once.
Public figures and organisations only, from their public professional output.

The arrangement

What you get, and what we ask.

You get

A key for the API, the MCP server and the bulk export.
Reports on the subjects you choose.
A direct line to us, and a say in what we build next.

We ask

Use it on real work.
Tell us when a verdict looks wrong.
Tell us what is missing.

Start

Try it now.

Reading is open: the dataset, the claims and single lookups need no key and no sign-up. Running reports on new subjects, the full API and the bulk export are by invitation, for design partners.

curl "https://epistacklabs.com/v1/lookup?name=Jensen%20Huang"

With a key, ask for any subject, or add it to your assistant:

curl -X POST https://epistacklabs.com/v1/reports \
  -H "Authorization: Bearer $RAPI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"name": "Ørsted"}'

claude mcp add --transport http reliability https://epistacklabs.com/mcp \
  --header "Authorization: Bearer $RAPI_API_KEY"

Become a design partner Developer docs