Suppose the software knows. Not that the market is about to fall — nobody knows that — but something narrower and far easier to know: that the person with a thumb over the sell button is about to do a thing they will undo in a fortnight, at their own expense. What does it owe them?
This used to be an idle question, because the knowing was the hard part. It is getting cheap. A system that has watched a few million sessions can learn the shape of the minutes before a reversal — the app opened at an odd hour, several screens touched in seconds, the whole holding sold in one go rather than trimmed. I have not seen published work putting numbers on this, so treat it as reasoning about how such systems work rather than evidence. The shape of the claim matters more than its accuracy, and it is unusual: this is not a prediction about the market but about the person.
That distinction carries most of the weight, so let me make it plain. “Prices will recover” is a forecast, and a hard one. “This user has sold in a hurry four times, and bought the same thing back within a fortnight at a worse price three of those times” is not a forecast at all. It is a record. The product is not claiming to know the market better than its user; it is claiming to know its user’s habits better than the user does in the middle of one. That is a much smaller claim, and probably true, because we do not keep score on ourselves and the software does.
The reversal is also the honest measure of regret. Someone who sells and stays out made a decision and lived with it; maybe their view changed, maybe they needed the money. Someone who sells and buys the same thing back higher two weeks later has paid for the round trip and ended up where they started. Nobody has to judge whether they were wrong about the world. They undid their own action and it cost them.
So what should the product do at the moment of the tap? Refusing is out. A system that will not let you have your own money is not a product, and the one occasion when the hurry is right is the occasion when refusal is unforgivable. Which leaves a delay, a warning, or nothing.
The case for a delay is better than I expected. It does not argue with anyone. The state that produces the panicked sell — heart going, a headline half-read — passes on a timescale of minutes to hours, so a twenty-minute hold is not a veto; it hands the same decision back to the same person once the rush has gone. Put that way it sounds almost unobjectionable.
Then argue against it, properly. If the price keeps falling during those twenty minutes, the delay has cost real money, and “you would have regretted it” is worth nothing to someone watching a number drop with their hands tied. And notice what the design has assumed: that the calmer self is the real one. Sometimes the person in a hurry has understood something first. Fear is occasionally information, and a product that treats every fast decision as a malfunction will smother the good ones with the bad.
Here is the objection that actually moved me, though. Friction is never neutral. Suppose a platform adds a confirmation step and a short hold to selling, and leaves buying at one tap. Follow the money: a hundred thousand people who would have sold at ten now sell at ten-twenty or nine-eighty. Supply that would have hit the market this morning arrives this afternoon, or never. That is a market view — prices recover, hold on — taken on behalf of a million people who were not asked. And the stated reason is never trustworthy on its own: a business earning a fee per trade wants no friction anywhere, and one earning a percentage of assets held wants friction on the way out only. Whatever the explanation offered, the design will tend to agree with the revenue.
So I do not think a product owes anyone prevention. It owes two narrower things, both defensible without knowing anything about where prices go. The first is symmetry: whatever friction sits on the way out sits on the way in. A confirmation screen for selling and a single tap for buying is not caution, it is a thumb on the scale. The second is the record — before the action, not after, the user’s own history with it, stated flatly: you have done this four times, three of them you bought back higher. That is information they already own. Withholding it because they might change their mind is the paternalism that friction only gets accused of.
I am not sure where showing someone their record ends and steering them begins, because a well-timed reminder of past regret is also a persuasion technique, and the timing is what makes it work. That line will not settle in the abstract. But I notice that the version I would be comfortable defending out loud is the symmetric one — the same speed bump in both directions, and the record handed over either way. It is a crude test. It is also the only one that does not depend on being right about the market.