I hear that argument a lot. It assumes spammers need published signals to adapt. They don't. They A/B test. They rotate domains. They monitor delivery rates through their own infrastructure. The people who actually need published signals are legitimate senders trying to diagnose problems they didn't cause.

Fine, but give me one first.

And the response was to remove the signal entirely, which punished everyone. Here's what that decision actually looked like from the sender side: a legitimate brand that monitored complaint volumes daily, acted on spikes within hours, and used that data to run sunset campaigns, suddenly had nothing.

They went from same-day complaint visibility to aggregate data with a two-day lag, available only in certain tools, only for certain sending volumes. The spammers adapted in a week. The legitimate senders lost a core diagnostic tool permanently.

I am describing a power transfer. Before FBLs died, a sender could self-diagnose. After, that diagnosis moved inside the mailbox provider. You no longer know what your complaint rate looks like at Gmail in real time. You see what Google decides to show you, when they decide to show it, in a format they control.

That is not a transition cost. That is a structural shift in who has information and who doesn't.

That argument only works if the models are never wrong. They are wrong. Not often, but at scale even a small error rate represents millions of legitimate messages landing in spam folders every day with no path to remediation.

When a rule-based filter blocks your mail, you can at least form a hypothesis. When an ML model blocks your mail, the answer is "the model decided." There is no appeal. There is no explanation. There is no path forward beyond "try sending differently and see what happens."

Nobody had a right to inbox placement, but they had access to enough signal to understand what was happening and take action. That access has been systematically reduced. Google Postmaster Tools used to show domain reputation at a granular level. That granularity has shrunk. FBLs are gone. SNDS is still around but shows less over time.

The trajectory is clear and it is not moving toward more transparency.

Agreed, filters exist to protect recipients. But recipients are also harmed when legitimate mail doesn't reach them. The person waiting for a password reset, an order confirmation, a medical appointment reminder — those people are recipients too.

When ML filtering incorrectly bins legitimate mail and there is no mechanism for senders to understand or challenge that classification, recipients suffer alongside senders.

Not full documentation. That would be exploited immediately and you'd be right to oppose it. What I want is enough signal to distinguish between "your content has a problem," "your list has a problem," "your infrastructure has a problem," and "we can't tell you, the model just decided."

Right now we often can't make those distinctions. The signal is too thin. That is the gap.

I think the gap is **convenient**. Mailbox providers don't need to answer for classification decisions they can't explain. Opacity protects them as much as it protects the filters. That doesn't require a conspiracy. It just requires that the incentives point in the same direction.

The current balance isn't neutral. It consistently favors large senders with the resources to maintain relationships with mailbox providers, and consistently punishes smaller operations doing everything right but flying blind. That asymmetry is worth naming even if you think the filters are broadly working.

Reinstate some form of complaint feedback for authenticated, high-reputation senders. You don't need to give it to everyone. Give it to the senders who have demonstrated they act on it responsibly. That's a targeted improvement that helps the people who deserve the signal without publishing it to the entire internet.