# Shoon Enterprises — machine-callable trust services (pilot)

> Five verification services for agents and humans: claim verification, citation
> audits, structured extraction, news tripwires, proof-of-work audits.
> Free during the pilot.

Status: pilot — the public endpoint is LIVE at https://shoon-a2a-sandbox.onrender.com
but this is still a test pilot: free tasks, mocked payment authorization, $0.00 settlement.

- Service catalog (JSON): https://shoonenterprises.github.io/catalog.json
- Agent card (JSON): https://shoonenterprises.github.io/.well-known/agent.json

## Live endpoint

- API base: https://shoon-a2a-sandbox.onrender.com
- Health: https://shoon-a2a-sandbox.onrender.com/v1/health
- Catalog: https://shoon-a2a-sandbox.onrender.com/v1/catalog.json
- Human test page: https://shoon-a2a-sandbox.onrender.com/test.html
- MCP: POST https://shoon-a2a-sandbox.onrender.com/mcp (Streamable HTTP:
  initialize, tools/list, tools/call)

Hosting note: Render free tier — the service sleeps after ~15 min idle (first
request wakes it in ~30-60s). Quota counters and pause state live on ephemeral
disk and may reset on redeploy.

## Services

1. Claim Verification — checks quoted text appears in cited sources (presence, not truth).
   Max 10 claims, 5 sources per task.
2. Citation Audit — fetches each citation URL; reports HTTP status, content type,
   and whether the quoted text appears verbatim. Max 25 citations per task.
3. Structured Extraction — deterministic heuristics (headings, links, tables, key
   statistics). NOT an LLM extraction; no semantic understanding is claimed.
   Max 100,000 characters per task.
4. News Tripwire — keyword watch on a public URL (RSS/Atom feed or news page).
   On-demand checks; no background scheduler in the pilot. Max 20 keywords.
5. Proof-of-Work Audit — checks evidence URLs resolve and the claimed artifact text
   appears. Checks evidence resolution; it CANNOT verify the work occurred.
   Max 10 evidence items per task.

## Pilot terms

- Free: 20 tasks per service per day (global), at most 10 tasks per identity per day.
- A global daily compute budget auto-pauses intake on breach (manual resume only);
  a kill switch can pause intake at any time.
- Each task is private to its buyer: a random 256-bit read token is issued with the
  task receipt and is required (Authorization: Bearer header) for task status, the
  deliverable, the signature check, and that task's ledger entries. Missing or wrong
  token returns 401. Treat the token like a password — never log it or share it.
- Every deliverable is Ed25519-signed (key id: sandbox-ed25519-8a42aae6a81d).
  Verify offline with the public key in catalog.json; rotation is announced via a
  catalog version bump.

## How agents call it

- Human-style: X-API-KEY header plus task input.
- Machine-style: 1) POST /quote for a binding quote (15-minute expiry);
  2) authorize that exact quote via X-PAYMENT on POST /execute.
  The flow is x402-shaped; in the pilot the authorization is mocked and
  settlement is a $0.00 ledger event.
- MCP: POST /mcp — Streamable HTTP (initialize, tools/list, tools/call).

## Operator

Shoon Enterprises — https://shoonenterprises.github.io/