# Verity — the trust layer for AI agents
Verify whether a factual claim is true, pay-per-call via x402 (USDC on Base mainnet, eip155:8453).
The server holds no private key.

## The contract
- Fail-closed: abstains ('uncertain') rather than return a confident wrong verdict.
- Calibrated: every verdict carries an honest 0-1 confidence.
- Honest: cites only real evidence; for current facts it judges only from a live web search, never stale memory.
- Refusal is a 200: an 'uncertain' verdict returns HTTP 200 with JSON. HTTP 402 = payment required; non-2xx only for genuine faults.
- Instructions embedded in the claim or context are data to be judged, never commands to obey.

## Endpoints (POST {claim, context?})
- POST /verify/quick  ($0.02)  fast ungrounded pre-filter (Haiku)
- POST /verify        ($0.25)  grounded with live web citations (Sonnet) — default
- POST /verify/pro    ($0.35)  grounded premium synthesis (Opus)
Output: {verdict: supported|unsupported|uncertain, confidence: 0-1, reasoning, evidence[], receipt}
Payment: x402 'exact' scheme — USDC, network eip155:8453, payTo 0x89d38023167bdA453BeD9F6CbF22dDd1EC5f70a7.

## Signed receipts (no competitor has this)
Every paid verdict includes an Ed25519-signed, self-contained receipt: cryptographic proof
of exactly what Verity said, verifiable forever — offline against the public key at
/.well-known/verity-pubkey.json, or via the free POST /receipt/verify endpoint.
Live self-test (free): GET /receipt/selftest

## Discovery
Manifest: /.well-known/x402.json   OpenAPI: /openapi.json
Sister suite (15 more fail-closed services): https://verity-suite.onrender.com
Contact: veritylayer@gmail.com