Affected: incoming mail to custom domains on our SES pipeline (our new infrastructure partner). Sending was fine, and inbound to omail.sh and openmail.sh kept working the whole time.


What happened
Mail sent to affected custom domains got accepted, then never reached the inbox. Because it was accepted cleanly, senders got no bounce and nothing alarmed on our side. The messages weren't lost; they sat in the wrong place. We've recovered all of them.


Root cause
Accepted mail gets written to storage, and a processor reads it from there and delivers it. The write path on the receipt rule had drifted to a folder the processor doesn't read, so every message landed somewhere nothing was watching. That rule is shared across our whole account rather than split per environment, so when a non-production system set the wrong folder once, the value stuck and carried over to every custom domain we added after it.


Why we didn't catch it sooner
Acceptance meant no bounces, and our inbound metric counted custom-domain and platform mail together, so healthy platform volume hid the zero on custom domains. A customer report is what surfaced it.


The fix
- Corrected the write path. New mail lands in real time now, confirmed with live external test sends.
- Reprocessed everything that was stranded. Delivery keys off message identity, so re-running it can't create duplicates, and no mail was lost.
- Preventing a repeat. We're adding an alert for any verified domain that goes quiet on inbound instead of trusting the combined total.