Your documentation has a new reader. It isn't human.
It doesn't skim, guess, or watch the walkthrough video. An AI agent fetches your page, parses the text, and either completes the task — or gives up. Prolectio scores how well agents can actually read and use your API docs, on a 0–100 scale.
Create a payment
Authorization: Bearer •••
{ "amount": 4200 }
Two very different readers.
The same page, read two completely different ways.
Documentation has been written for humans for thirty years. But a fast-growing share of the traffic hitting your docs is agents acting for a developer — and they read nothing like a person does.
Forgiving. Resourceful. Patient.
Renders the whole page in a browser, skims, infers intent, and fills the gaps with a search, a forum post, or a support ticket.
- Tolerates ambiguity and missing steps
- Sees JavaScript-rendered content
- Asks a colleague when stuck
Literal. Fast. Unforgiving.
Fetches the raw page, parses the tokens it receives, forms a plan, and calls the API — or abandons the task the moment the page can't answer it.
- Needs auth, limits & a runnable request, stated plainly
- Often sees only what loads without JavaScript
- Doesn't guess — it stops
One number an agent can trust — built from three.
Task completion is weighted over surface structure: you can't top the Index by dropping in an llms.txt and calling it done.
Deterministic checks
Clarity, graded by model
Real task probes
We scored 47 platforms your agents already read.
An open, reproducible benchmark of public developer documentation. Here are the platforms that ship docs an agent can read cold.
Scan, score, and get the fix — reproducibly.
Scan
We crawl your public docs as an agent would — one request a second, robots-respecting, no logins. A snapshot is captured so every score is reproducible.
Score
Three categories, a pinned model, and full provenance — the exact prompt and output stored with every result. The methodology is published and versioned.
Fix
A specific, ranked list of what an agent couldn't read — the missing auth example, the JS-only reference, the rate limit buried in prose — with the transcript that proves it.
Make your docs the ones agents can actually use.
Scan your documentation, see exactly where an agent stalls, and ship the fixes. When you clear the green bar, show it.