[probst]

There is actually a very good reason to drop these email addresses, and the reason is that a high-rate of non-deliverable emails hurt your sender score. It's a total pain to get emails delivered to the major email providers in the first place, and you immediately land in spam (or with emails not delivered at all) if they don't trust the sending email server or your score is anything but stellar!

[mehrdadn]

I have 2 responses to the sender reputation concern:

1. If the user's mail service penalizes you equally regardless of whether the recipient's addressed existed 1 day vs. never existed, that itself is absolutely inexcusable nonsensical behavior that needs to be fixed. You shouldn't do that, just as you shouldn't shoot the mailman (or even arm yourself...) merely because he knocked a second time.

2. Notwithstanding the previous point, I don't buy this as valid justification anyway. The proposal isn't that you should blast 100 emails toward the mailbox every time you get a bounce due to an address not existing. The idea was to just exercise some intelligence in the matter. Like maybe just retry a couple times, spaced out by a day or two. The bounce rate increase due to such an event is very negligible here—people don't suddenly delete their accounts en masse. When that happens, it's clearly due to an outage, not because half the users at that domain suddenly decided to delete their accounts. (Which is something you can also easily detect across the domain as another useful signal to drastically lower the bounce rate across the entire domain, btw, if you're absolutely paranoid about your immaculate delivery rate dropping by an epsilon. But it shouldn't be necessary given how negligible the impact should be.)

So I don't buy this excuse one bit.

[judofyr]

> The proposal isn't that you should blast 100 emails toward the mailbox every time you get a bounce due to an address not existing. The idea was to just exercise some intelligence in the matter. Like maybe just retry a couple times, spaced out by a day or two.

What you're proposing is to explicitly ignore the specification (which says that you should _not_ retry when you receive a 550) and try to implement a custom smart retry logic that handles temporary error cases, but also does not get you blocked.

> So I don't buy this excuse one bit.

I'm all for building resilient services, but "try to detect when a server incorrectly returns 550" is not something I would prioritize at all. I'll happily manually clean up after this occurrence than to have this complicated retry logic. It's not an "excuse", it's a very sensible trade-off.