Evaluate email verification with evidence

What this method establishes

This is an evaluation protocol and blank worksheet, not a published accuracy benchmark. No measured comparison or delivery-rate claim is made here. Use it to record a bounded test against addresses and domains you control or are authorized to test. Provider behavior and routing can change over time.

1. Define the sample before running

Include a known existing mailbox on a domain you control, a nonexistent mailbox on a rejecting domain you control, an existing and a nonexistent mailbox on your controlled catch-all domain, malformed syntax, and a domain confirmed to have no mail service. Label expectations from independently recorded configuration, not from the verifier being evaluated. Keep catch-all cases inconclusive for mailbox existence unless you have independent evidence.

• Download the evaluation worksheet

2. Record every attempt

Fill the worksheet with the case identifier, controlled input, independent expected evidence, UTC timestamp, provider, returned status/reason, HTTP status, elapsed milliseconds, credits, and notes. Preserve failures and unknowns. Keep private addresses and credentials out of published evidence. Record configuration changes and versions where available.

3. Report the denominator and uncertainty

Report submitted cases, completed conclusive results, unknowns, HTTP/network failures, and unrun cases separately. Classify correct and incorrect conclusive decisions only where independent ground truth exists. Report catch-all behavior separately from mailbox existence. If you compute accuracy among conclusive cases, also show its exact numerator/denominator and the unknown/failure share over all submitted cases.

Example report structure (no measured values)
Planned / submitted / unrun: ___ / ___ / ___
Valid / risky / invalid / unknown: ___ / ___ / ___ / ___
HTTP or network failures: ___
Ground-truth eligible cases: ___
Correct conclusive / all conclusive eligible: ___ / ___
Unknown + failures / all submitted: ___ / ___
Total credits / elapsed time: ___ / ___

4. Publish only what the evidence supports

Attach the sample definition, date, method, anonymized raw outcomes, limitations, and reproducible steps. A controlled test supports claims about those cases, not an accuracy percentage across all customer lists. Verification outcomes alone do not prove inbox placement. If you compare services, run the same authorized sample and retain every outcome for each service.

• Run a documented verification request
• Understand catch-all limitations
• Understand unknown outcomes