The AI verdict endpoint ($0.01/call) has always returned a confidence score (0–100) with every verdict. The gate turns that number into a promise: if a paid verdict's confidence falls below 70, that same response automatically includes a free single-use re-check token. The list price does not change.
| Term | Value |
|---|---|
| Threshold | 70 (confidence < 70 triggers compensation) |
| Compensation | One free re-check of the same contract, single-use, valid 24 hours |
| How it appears | _confidence_gate block in the paid response: status: "below_threshold", recheck_token, recheck_expires |
| How to redeem | Repeat the same POST /api/verdict with header X-Confidence-Recheck: <token> and no payment header |
| Passing verdicts | _confidence_gate.status: "pass" — the block is on every paid response, so the terms are auditable per call |
| Re-checks | A redeemed re-check is marked as such and never mints a second token — no free loop |
| Price | Unchanged: $0.01 per paid call |
The endpoint's confidence comes from its own verdict pipeline: the analysing model reports 0–100 per verdict. Every fallback path in that pipeline reports at least 70 — the deterministic heuristic maps its risk score to 70/80/85, and an unparseable model reply defaults to 70. So a sub-70 confidence can only come from the model itself saying, in effect, "I am genuinely unsure about this one" — exactly the verdict a buyer should not have to pay twice to double-check. The gate triggers on that signal and nothing else; no new score was invented for it.
POST https://api.mekongtradeintel.com/aiverdict/api/verdict
{ "chain": "base", "contract": "0x..." } + Payment-Signature
→ 200 {
"verdict": "RISKY", "confidence": 62, ...,
"_confidence_gate": {
"threshold": 70, "confidence": 62, "status": "below_threshold",
"compensation": "free_recheck",
"recheck_token": "rcg_...", "recheck_expires": "2026-...T...Z",
"how_to_recheck": "Repeat the SAME POST /api/verdict (same
contract) within 24h with header X-Confidence-Recheck:
rcg_... and NO payment header — served free, once."
},
"_receipt": { ... }
}
The same policy line is printed in the endpoint's 402 response bodies, so the terms are visible before you pay. Sibling r730 tools: the paid fleet index and team trial-share codes.