Passcode failures no longer blame the member’s email provider for our errors
Shipped 2026-10-20
Before this change, every failed passcode email showed “your email provider rejected the message”. During the 2026-10-19 outage (TUN-897) the failure was ours, a send error before Mandrill was ever called, and around 30 members were told otherwise.
What a member sees now
describeDeliveryFailure now returns a cause:
| Cause | When | Message |
|---|---|---|
rejected | Mandrill rejected the recipient (bounce, spam, unsubscribed) | The provider-rejection copy as before. The “mailbox may be full” hint only appears for a soft bounce |
send_error | Transport error, Mandrill API error, no recipients accepted, or anything unrecognised | ”Something went wrong on our side… try again in a few minutes”, plus the same fallback options |
An unknown or missing cause counts as send_error. The member’s provider is
only blamed when Mandrill says so, and only for reasons about the address
itself: hard-bounce, soft-bounce, spam, unsub, invalid, custom.
invalid-sender, unsigned, test-mode-limit and rule are our sending
config, so they count as send_error.
Logging
Mandrill’s webhook never reports rejected messages, so these log lines are the only record of them. Every email failure, not just passcodes, now logs at a level that says whose problem it is:
| Level | When | Use |
|---|---|---|
error | Our failure: transport, Mandrill API, no recipients, our-config rejections | Alert on it |
warn | Member-side rejection | Watch the rate. A spike points to a sending reputation or DNS problem |
Each line carries event: email.send_failed with reason and
reject_reason. The logger only lifts error into the message at error level,
so search by attribute: @event:email.send_failed status:error.
passcode.send_failed now logs at warn, as flow context, so one failure
isn’t counted as two errors.
Which flows unlock fallbacks
A failed passcode email opens the 15-minute security-question login fallback
(TUN-870). That unlock is now opt-in through
sendPasscodeToEmail({ unlockFallbacks }):
- Login and account recovery pass it. Both send to the member’s stored address, so a failure there unlocks the fallback whatever the cause, and email-only members can still get in during an outage.
- Sign-up and a profile email change don’t. Both send to an address the
person has only just typed. The sign-up endpoint is unauthenticated and takes
member_idfrom the request body, so anyone could post a member’s id with an address that bounces and unlock security-question login for that account, replacing the emailed code for anyone who has the password.
Non-passcode emails (logbook, bookings, notifications) never touched the unlock and still don’t.
Not covered
Sign-up and the profile email change still ignore the send result and always report “code sent”. That needs UI work in both flows and is tracked separately.
- Jira: TUN-898
- Author: @alexhobday