Acceptance is not delivery

A receiving mail server accepting your message and a human being seeing it are two different events, separated by an unknown amount of filtering. Most platforms report the first and let you assume the second.

Here is what a 250 2.0.0 OK actually tells you: a server on the other end took custody of the bytes. That is it. What happens next — inbox, spam folder, quarantine, silent discard — happens inside systems that deliberately do not report back, because reporting back would help spammers calibrate.

Why the blur is convenient

If you are selling email infrastructure, “98.7% delivered” is a wonderful number. It is technically defensible, it is large, and it is almost entirely uninformative. The customer reads it as “98.7% of my users saw this”, which is not what it says and not what anyone promised.

We think the vagueness is the product problem, so we use four words instead of one:

• Acceptance — the receiving server took the message.
• Deferral — the receiving server asked us to try later, usually with a reason.
• Inbox placement — where an accepted message actually landed.
• Latency — time from your API call to the provider’s answer.

The one that earns its keep

Deferrals. A 450 4.7.1 is the receiving provider telling you something, in words, before anything has actually failed. Concentrated deferrals at one provider almost always mean a reputation signal rather than an addressing problem — and the response text usually says which outright.

This is the event most platforms either hide or roll into a retry counter. It is the single most useful early warning available, and it is free: the provider is volunteering it.

What we do about it

Every event OutSend stores keeps the raw payload it came from. Not a normalised summary alone — the actual response, the actual code, the actual reporting host. You can hand that to a person, an auditor, or an agent, and all three can reason about it.

We do not currently measure inbox placement systematically. When we do, it will arrive with its methodology published, because a placement number without a method is exactly the kind of claim this post is complaining about.