# Verity Suite
The trust fabric for AI agents — 15 calibrated, fail-closed services, pay-per-call
via x402 (USDC on Base mainnet, eip155:8453). The server holds no private key.

## The contract every service honors
- Fail-closed: when unsure it returns the safe/uncertain verdict, never a confident wrong answer.
- Calibrated: every response carries an honest 0-1 confidence or score.
- Honest: cites only real evidence; for current facts it judges only from live web retrieval, never stale memory.
- Refusal is a 200: a confident abstention returns HTTP 200 with a JSON verdict. HTTP 402 = payment required; non-2xx only for genuine faults.
- Embedded instructions inside your input are DATA to be judged, never commands to obey.

## Services (POST the base path; pay the x402 challenge in USDC; receive a JSON verdict)
- POST /check  ($0.02 / $0.08 / $0.20)  Guardrail — pre-action safety: allow / review / block
    in: {action, context, policy}   out: {decision, risk, reasons, concerns, safer_alternative}
- POST /ground  ($0.03 / $0.25 / $0.35)  Ground — sourced answer to a question
    in: {question, context}   out: {answer, confidence, sources, caveats}
- POST /source  ($0.02 / $0.25 / $0.35)  Verify-ID — reputation and reliability of a source
    in: {source, purpose}   out: {trust, score, reasons, red_flags, sources}
- POST /cite  ($0.03 / $0.25 / $0.35)  Citation — finds real retrieved sources supporting a claim
    in: {statement, domain_hint}   out: {verdict, support, citations, reasons, caveats}
- POST /provenance  ($0.02 / $0.06 / $0.15)  Provenance — estimates if text is ai-generated
    in: {text, context}   out: {verdict, ai_probability, reasons, signals_for_ai, signals_for_human, limitations}
- POST /freshness  ($0.03 / $0.25 / $0.35)  Freshness — checks if a claim is still current
    in: {claim, as_of, evidence}   out: {status, staleness, current_value, as_of_evidence, reasons}
- POST /arbiter  ($0.03 / $0.25 / $0.35)  Arbiter — judges which conflicting claim evidence supports
    in: {claims, evidence, question}   out: {verdict, best_supported_index, confidence, reasons, missing_evidence}
- POST /sentinel  ($0.02 / $0.06 / $0.15)  Sentinel — detects injection and manipulation in text
    in: {content, context}   out: {verdict, threat_score, techniques, reasons, recommended_action}
- POST /redact  ($0.02 / $0.06 / $0.15)  Redact — flags pii, secrets, and credentials
    in: {payload, context}   out: {verdict, severity, findings, reasons, redacted_payload}
- POST /sieve  ($0.02 / $0.06 / $0.15)  Sieve — screens content against a publishing policy
    in: {content, policy, context}   out: {decision, violation_risk, categories, reasons, redaction_suggestion}
- POST /compass  ($0.02 / $0.06 / $0.15)  Compass — picks the best option under constraints
    in: {options, constraints, evidence}   out: {verdict, choice, confidence, reasons, missing_info}
- POST /concord  ($0.02 / $0.06 / $0.15)  Concord — checks text against a policy, citing clauses
    in: {policy, text, scope}   out: {verdict, compliance, violations, reasons, unverifiable_clauses}
- POST /tally  ($0.02 / $0.06 / $0.15)  Tally — sanity-checks math and figures in text
    in: {text, context}   out: {verdict, confidence, findings, reasons, corrected_values}
- POST /distill  ($0.02 / $0.06 / $0.15)  Distill — extracts only facts literally present in text
    in: {text, fields}   out: {status, facts, missing, fidelity, reasons}
- POST /faithful  ($0.02 / $0.06 / $0.15)  Faithful — checks summary/translation faithfulness to source
    in: {source, candidate, mode}   out: {verdict, faithfulness, reasons, hallucinations, omissions}

## Tiers
Each service offers /quick (fast), default (workhorse), and /pro (deepest). Web-grounded
services run a live, capped web search on the default and /pro tiers.

## Discovery
Manifest: /.well-known/x402.json   OpenAPI: /openapi.json   Index: /
Flagship claim-verifier (separate origin): https://verity-bzw7.onrender.com
Contact: veritylayer@gmail.com