The account that was removed

Consider a woman who has spent six years building a small business inside someone else's walls. Her workshop is real — she makes things with her hands — but her customers, her reputation, and most of her income arrive through a single account on a single platform. One morning the account is gone. In its place is a notice: her content violated the community guidelines, and the account has been removed. It does not say which content. It does not say which guideline. There is a button to appeal. She presses it, describes her situation, and waits. A day later the reply arrives, and it is the same notice, arrived at by the same process, carrying the same verdict. No person ever names the specific post, the specific rule, or the specific signal that ended six years of work. She is not told she was wrong. She is told the decision has been made, and that the decision is final.

Set aside, for a moment, the question everyone reaches for first: was she actually guilty of something? It may be that she was. Platforms carry real obligations — to their users, to the law, to the basic hygiene of a shared space — and they cannot litigate every removal in open court. The platform's authority to remove is not what is in question here. Its terms of service reserve that authority in plain language, she agreed to them, and she has no standing to claim she was owed a permanent seat. Authorization, in the narrow legal sense, is not the problem.

The problem is legitimacy, which is a different and harder thing. A private company that has become the public square for millions of people is exercising a governance function over people who never consented to be governed by it — only to use it. And legitimacy, unlike authorization, is not conferred by a document the governed had no power to negotiate. It is earned, continuously, by being answerable for the exercise of the power. A platform that can end a livelihood but cannot say, to the person whose livelihood it ended, exactly why, has all the authority and none of the legitimacy. It is acting within its rights and failing the only test that would make those rights defensible.

A verdict is not an account

Moderation at scale does not produce reasons in the sense a person would recognize as reasons. It produces scores. A classifier assigns a probability that a piece of content falls into a prohibited category; a threshold converts that probability into an action; the action is logged as a policy violation. Somewhere upstream a human wrote the policy, but no human read this post and concluded it broke that rule. So when the notice says the content violated the community guidelines, it is not describing a judgment. It is citing a category. And a citation is not an account. It tells you the drawer your case was filed in, not what was found inside it.

This is why the appeal that returns the same verdict is not a malfunction but the design working as intended. You cannot contest what you cannot see. To mount a defense she would need to know which post, measured against which rule, on the strength of which signal — and it is precisely those three things the process withholds. An appeal against an unnamed charge is not an appeal. It is a form that lets you feel heard while the machine re-runs the same inputs to the same output. The absence of the specifics is not an oversight in the system. It is the system.

A platform earns the right to remove you the same way any authority earns the right to punish — by being willing to show you, in specifics you can actually argue with, exactly what you did and which line it crossed.

Removal is a sentence, not a filter

We talk about content moderation as though it were filtration — a screen that lets the good through and holds the bad back, adjustable, reversible, no great matter if a grain slips the wrong way. For the platform, that is roughly true. Reinstating an account costs a few keystrokes; the record was never really deleted, only flagged; a reversal is, institutionally, trivial. But the asymmetry is total, because for the person on the other side removal is not a filter setting. It is a sentence. It is fast, it is nearly free to impose, and it is, in practice, close to irreversible — the audience disperses, the ranking resets, the momentum of six years does not come back because a switch was flipped a month too late. The thing that is reversible in principle for the platform is unrecoverable in fact for her.

An account of a decision has to be built to match the stakes as they fall on the party who bears them, not as they register on the party who acts. When the same event is a shrug on one side of the glass and a catastrophe on the other, calibrating the process to the shrug is how power quietly excuses itself from the weight of what it does. The measure of what is owed is set by the person with the most to lose, because they are the only one for whom the decision is not abstract.

What legitimacy would require here

It would not require the platform to keep her, or to be right, or to move slowly. It would require three specific things she does not have. The content actually cited — this post, quoted back to her. The rule it was measured against — this clause, not the whole guideline document. And the signal that triggered the action, rendered into something a person can read and dispute, rather than a score she is never shown. Add to that an appeal that reaches someone with the authority to reverse, and that leaves a durable record either way — a preserved account of the decision that says what was decided, on what basis, and what the review concluded, so that the outcome can be examined later by someone other than the party that produced it.

None of this exposes proprietary machinery, and none of it is beyond a platform that already logs everything for its own purposes. Which is the uncomfortable part. A platform that could show its work and declines to is not protecting a secret; it is choosing opacity, and choosing it against the person least able to bear the cost of it. Sometimes the choice is defensible — there are genuine cases where naming the signal would hand a playbook to the people gaming the system. But that defense has to be made, and made narrowly, not assumed as the default posture toward everyone. Opacity offered as a blanket policy, from an actor with this much power over people with this little, is not a security measure. It is where legitimacy goes to die — quietly, behind a form, one final verdict at a time.

The scenario above is illustrative — a composite drawn to show a pattern, not an account of any real person, company, or event.

— Dispatches · Summit Cognitive