Accuracy
How we measure verification accuracy
Accuracy means a verdict that matches a mailbox we already know is real or not. This page is the method. A completed run has not been published.
Method
How a run would work
Four steps. No results table until the first run is finished.
1. Build a seeded ground-truth list
A seeded list of known-good, known-bad, and known-catch-all mailboxes across major providers (Gmail, Outlook/Microsoft 365, Yahoo, Workspace-hosted domains) plus a long tail of self-hosted mail servers.
2. Run the same list through each API
Each API is run against the identical list in the same window, using default settings and no manual tuning.
3. Score against ground truth
Precision, recall, and F1 for each result set, plus a catch-all resolution rate: the share of catch-all addresses turned into a usable verdict instead of "unknown" or "risky".
4. Publish the numbers
Results would be refreshed as mail infrastructure changes, including months where the score drops. We have not published a completed run yet.
What we publish today
An internal target, not a published score
Our stated target is 99.5% on an internal seeded list. That figure is our target, not an independently published result. If an address we mark deliverable hard-bounces within 30 days, we refund 3× the credits spent verifying it.
FAQ
Accuracy methodology FAQ
How do you measure verification accuracy?
Against a seeded ground-truth list of known-good, known-bad, and known-catch-all mailboxes. We score precision, recall, and F1, plus how many catch-all addresses receive a usable verdict. This page describes that method. A completed run has not been published.
What does catch-all resolution rate mean?
A catch-all mail server accepts messages to any address at its domain, so a plain SMTP handshake cannot confirm a specific mailbox exists. The resolution rate is the share of those addresses that still receive a send-safe or do-not-send verdict from pattern and reputation signals, instead of "unknown" or "risky".
When will results be published?
When a full seeded-list run is complete. Until then this page stays a description of the method, with no accuracy table.
Is the seeded test list public?
The list itself is not published. Publishing known-good and known-bad addresses would let a checker tune to the test. The list composition and scoring method on this page are what a later run will follow.
Verify a list and read the verdict
100 free credits after a work-email application. No card required.