Index #1 · rank 26 of 47 · DevInfra · scanned 2026-07-19
Vercel
Agent readability of https://vercel.com/docs — scored from a reproducible crawl snapshot under rubric v0.
Deterministic · 40%
90
LLM rubric · 35%
72
Task probes · 25%
67
Band
amber
Machine-checkable fundamentals
✓llms.txt at the rootllms.txt present · 66 sections · 1716 links100%
✕OpenAPI spec discoverable & validOpenAPI found (https://vercel.com/login?next=%2Fopenapi.json) but did not validate40%
✓Heading structuremean heading-quality 100% over 60 pages100%
✓Code samples on reference pages57/59 reference pages have code samples97%
✓Auth documented with an exampleauthentication documented with a code example100%
✓Pages within agent token budget60/60 pages within 8000-token budget100%
✓Links resolve0/10160 links broken100%
✓Crawler & meta hygienemean meta-hygiene 98% over 60 pages98%
Can an agent actually finish?
✓Find the authentication methodnamed an auth method with a cited sourcePASS
✕Find the rate limitsanswer states no concrete numeric rate limitFAIL
✓Construct a first valid requestproduced a request construct with a cited sourcePASS
Executed by a pinned claude-sonnet-5 agent over the crawl snapshot; graded on substance + citation, transcripts stored.
Prose quality, graded by model
Clarity
2.8/4
Completeness
2.8/4
Consistency
3/4
Key pages assessed
6
Is this your documentation? Every score is reproducible from stored inputs, and corrections are part of the methodology — if a page we couldn't reach is reachable to a normal agent, we'll re-scan and publish the update. See the methodology or grab your badge.