AI Scoring API
How CorrectInbox turns raw signals into an honest confidence score.
The four verdict states
Verification is probabilistic, especially for catch-all domains. Rather than collapse everything into valid/invalid, we return one of four states plus a 0–1 confidence score.
Reading the score
- 0.90–1.00 — high confidence; safe to act on.
- 0.70–0.89 — likely, but apply your own threshold.
- below 0.70 — treat with caution or re-verify later.
Why “unknown” exists
When a true catch-all genuinely can't be resolved, we return “unknown” with a score — never a guessed “valid” to inflate numbers. This protects your sender reputation.
Thresholding example
const { result, confidence } = await verify(email);
const accept = result === "valid"
|| (result === "catch-all" && confidence >= 0.8);