We measured our filter over a full month of real mail, on our own production server. Here are the numbers.
Spam doesn’t reach you
97% of hostile mail is stopped before your mailbox. Most of it is refused at the connection, never touching your disk.
Your real messages do
This is the number that actually matters, and the one almost nobody publishes: one legitimate message in 180 was misfiled. That’s 0.56%.
A spam that slips through costs you three seconds. An important message that disappears can cost you a client. We optimise for the second one.
Three times less spam than the reference open-source filter
We compared OxiMail to rspamd — the filter most self-hosted servers run — on exactly the same mail, tuned so it makes the same rate of mistakes we do.
It lets nearly three times more spam through.
And you’ll always know why
Every filter decision is stored with its reasons, in plain words: sender never seen before, suspicious reply address, server on a blocklist. You can look them up months later, for any message.
No hosted service will tell you that. There, the explanation is that the model decided.
Measured in August 2026 on our production server, across all mailboxes. Detailed method available on request.