Extractability Oracle
Extractability Oracle is a deterministic, pay-per-call API that scores any web page from 0 to 100 for how well AI crawlers and answer engines can understand and quote it (GEO). Every paid response is ECDSA-signed, backed by a public hash chain, and returns a per-check breakdown plus prioritized fixes. Paid in USDC on Base via x402 — no accounts, no API keys.
The rubric — 6 categories, 100 points
Scoring is 100% deterministic: the same input always yields the same score, with no LLM or third-party call — which is exactly what makes the signed result meaningful. Rubric version 2.0.
| Category | Points | What it measures |
|---|---|---|
| Structured data / JSON-LD | 25 | Valid JSON-LD, recognized @types, schema richness (≥5 properties) |
| Extractable facts & patterns | 20 | Statistics, dates, FAQ or comparison tables, a front-loaded definitional lead |
| Metadata | 15 | Title 10–60, meta description 50–160, canonical, OpenGraph / Twitter |
| Structure & headings | 15 | Single H1, logical H2/H3 hierarchy, lists and tables |
| Machine readability / AI access | 15 | llms.txt depth, robots allowing AI crawlers, semantic HTML5 + ai.txt |
| Authority & entity | 10 | Brand / Organization, author & freshness, citations, internal links, breadcrumbs |
Endpoints
GET /extractability?url=— signed score + subscores + fixes · $0.01POST /extractability— score raw HTML in the body · $0.01GET /extractability/preview?url=— free, unsigned, rate-limitedGET /provable/head·/provable/verify— the tamper-evident scan chain
Machine-readable: /.well-known/x402 · /openapi.json · /llms.txt · /ai/summary.json. Built on the open x402 protocol and schema.org vocabulary; part of the nsgoods suite.
Frequently asked questions
What is an extractability score?
A 0–100 measure of how easily an AI crawler or answer engine can understand and quote a page, from its structured data, extractable facts, metadata, structure, machine-readability, and authority signals.
Is the score deterministic?
Yes — the same input HTML and origin-file state always produce the same score. No LLM or third-party call is involved, which is what lets each paid response be ECDSA-signed and hash-chained.
How does an agent pay?
Over x402 on Base mainnet in USDC: the agent calls the paid endpoint, receives HTTP 402 with the price, submits a signed USDC payment, and retries. The free preview needs no payment.