Deliverability

Reputation

SPF, DKIM and DMARC answer the question of whether a message is genuine. Whether it lands in the inbox is decided by another authority — one that publishes no rules, gives no reasons and has no appeals procedure.

The ticket is not the seat

It is a widely shared disappointment: all three procedures are in place, the checking tools report nothing but green ticks, and the message still lands in the spam folder. Nothing is broken there. The procedures answer one question only — does this message really come from this sender? — and the answer "yes" is the precondition for being assessed at all.

The actual decision comes afterwards, and it reads: do our users want mail from this origin?

What a reputation is made of

No large provider publishes its scoring rules, for an understandable reason: published rules get gamed. From what providers describe in their guidance for senders, though, a clear picture of the signals emerges:

• How recipients react. The strongest signal there is. Is the message opened, answered, rescued from the spam folder? Or deleted, ignored, reported as spam? A single "report as spam" click weighs more than a thousand delivered messages.
• Sending behaviour. Steady volume from the same source looks different from a sudden burst. An address that has been quiet until now and sends ten thousand messages tomorrow looks like a compromised account — because it often is.
• Invalid recipients. A lot of bounces suggests bought or stale lists. Those who work cleanly have few. More on that under bounces.
• The neighbourhood. With shared sending address ranges, the behaviour of everyone sending from there counts. You inherit your neighbours' reputation without ever meeting them.
• Age and consistency. A domain that has sent the same kind of mail from the same direction for years is predictable. Predictability is everything here.

Why a new sender has a hard time

Reputation is a history. Having none does not make you neutral, it makes you unknown — and in case of doubt unknown is treated as unwanted, because that is the cheaper wrong decision.

Which is exactly where the asymmetry that shapes today's system comes from: a large sender with a well-kept name gets through reliably. A small, technically impeccable sender struggles — not because it is doing anything wrong but because it sends little and therefore produces few of the signals trust could be built from.

Greylisting: the other direction

One procedure gets by without any history at all and is nevertheless astonishingly effective: the receiving server answers an unknown sender's first attempt with a temporary rejection. A proper mail server remembers the message and comes back minutes later — which is exactly what SMTP prescribes for 4xx replies.

Bulk senders frequently do not come back, because their software is built for throughput rather than persistence. The price is a one-off delay of a few minutes per new sender. For a contact mailbox that is a good trade; for sending sign-up confirmations it is a bad one.

What you can influence

Reputation cannot be configured, only earned. What actually helps is unspectacular: send only to recipients who expected it; evaluate bounces and remove invalid addresses; act on unsubscribes immediately and make them easy to find; send marketing and operational mail from separate sender addresses, so that a failed newsletter does not drag the invoice mail down with it; increase volume slowly.

And the uncomfortable truth behind it: there is no body you can appeal to. Some providers maintain contact routes for senders, but a right to delivery does not exist. The receiving server decides whose mail it inflicts on its users, and owes nobody an explanation.

Why it turned out this way

This layer is no part of the design. It grew because the protocol had no way of limiting unwanted mail at the source — sending costs nothing, as described on the page about the invention of spam.

Where a brake is missing, you build one in at the other end. That this brake is opaque, names no rules and structurally disadvantages some participants is not by design but a consequence of its having had to come about after the fact.