Things You Know and Cannot Show

A fire lieutenant took his crew into what everyone thought was a kitchen fire. Standard call, single storey house, fire in the back. Inside, the living room was hotter than a kitchen fire should make it, and the place was too quiet. Fires are noisy. This one was not.

He ordered his men out. Moments after they cleared the building, the floor where they had been standing collapsed. The fire had been in the basement the whole time, directly beneath them.

Years later, asked why he pulled everyone out, he told the researcher interviewing him that it had been ESP. A sixth sense. He was insistent about it.

Here is the part almost every retelling gets wrong, and it is the part that matters.

He did not work out that there was a fire in the basement. He did not know the house had a basement. What his experience gave him was not a diagnosis. It was an alarm. In Gary Klein's account, the lieutenant's expectations were violated and he realised he did not know what was going on, and that is why he left. The heat was wrong and the silence was wrong, because a floor was muffling the fire underneath it, and hundreds of fires had taught him what a kitchen fire feels like well enough to know that this was not one.

So the story is not really about a man who knew something. It is about a man whose instinct correctly reported this does not match anything I have seen, and who was disciplined enough to act on that alone without inventing an explanation.

Hold onto that, because it is the difference between a good read and an accusation.

What our rule actually says

We are a company built on the idea that you do not get to assert what you cannot show. So let us be clear about what that rule is and is not.

It governs publication. It has never been a claim that unproven things are unreal. The lieutenant proves they are not, and so does every nurse who calls a doctor before the chart changes.

The reason we hold the line hard on our side of it is narrower. The standard of proof scales with who pays if you are wrong. Wrong about a hunch on your own work, you pay. Wrong about a person in public, they pay, and they never agreed to the wager.

You could argue that bar should slide. A cheap, well hedged statement costs less than a confident one. We do not slide it, because a bar that slides is a bar that gets argued down in the exact moment it matters most.

When a hunch is worth something

In 2009 Daniel Kahneman and Gary Klein published a paper together called Conditions for Intuitive Expertise: A Failure to Disagree. They had spent their careers on opposite sides. Kahneman ran the experiments where human judgment came out flawed. Klein went into firehouses and intensive care units and found judgment working.

They converged almost completely on the substance and stayed divided on temperament. Kahneman still finds the errors more interesting; Klein still flinches when people say "bias." Their own line is that the sharpest differences between them were emotional rather than intellectual, which is a more honest ending than most collaborations get.

What they agreed on: skilled intuition needs an environment with real, learnable regularities in it, and prolonged practice with feedback that is rapid and unequivocal. Both are necessary. Neither is sufficient, and they add that talent still matters.

They also add a warning that is easy to skip. Hogarth called them wicked environments: settings where feedback exists and teaches you the wrong thing. Their example is a physician who could reliably predict which patients would develop typhoid, and confirmed his intuitions by palpating their tongues without washing his hands. Fast feedback. Unambiguous feedback. Confirming feedback. He was creating the outcome he was predicting.

Which brings us to your lobby

Now the honest part, because the easy version of this argument is wrong and you would catch it.

The easy version says a player's suspicion about a stranger gets no feedback, so it never becomes skill. That is not true and you know it is not true. You spectate on death. You pull the demo and watch the round back from his seat. You get the notification saying action was taken against someone you reported. Ban waves land. People get caught. Some of you have thousands of hours of that, and it has made you genuinely good at reading a match.

We are not going to tell you your read is noise. It probably is not.

Here is the real problem, and it is the lieutenant's problem. Skill is fractionated. Kahneman and Klein are blunt about it: a financial analyst can be genuinely skilled at judging whether a company will do well and have no skill at all at judging whether its stock is underpriced, and neither he nor anyone watching him can feel where the boundary is.

Your version of that boundary runs between two questions that feel identical from the inside. Is something off in this match? and is that specific person cheating? The first is anomaly detection, which is exactly what long exposure trains and exactly what saved that fire crew. The second is causal attribution about a stranger, and almost nothing in your experience corrects you on it specifically. You find out someone got banned. You rarely find out about the person you were sure about who was clean, because nothing happens to them and nobody tells you.

And the Discord thread that agrees with you is not feedback. It is a wicked environment with a tongue and no handwashing.

So the read stands and the verdict does not. You are the lieutenant. You correctly detected that something does not match. You do not know there is a basement.

The same test, turned on us

It would be cheap to apply this to you and not to ourselves, so here is where we sit.

Vera is young. Our environment has real regularities, because process behaviour and system state genuinely differ between an honest machine and a compromised one. Our feedback is thinner than we would like. When a record clears someone, we usually do not find out whether we were right, for the same structural reason you do not: nothing happens, and nothing happening produces no signal.

That is precisely why we build the way we do. We are not calibrated enough to be trusted on instinct, so we do not ask you to trust our instinct. We show the collected evidence and let you read it. The discipline is not humility for its own sake. It is what you do when you know your own feedback loop is incomplete.

The working rule

Steer with what you cannot prove. Publish only what you can.

Almost nothing gets verified that somebody did not first suspect, so the hunch is what aims the search. Plenty of the best things in this project started with someone saying "that feels wrong" about something they could not yet explain, and then going and finding out. What the hunch never becomes is the finding.

The practical thing

If a match feels wrong, treat the feeling the way the lieutenant treated his, which is as a reason to move rather than a conclusion to announce.

Save the clip and note the timestamp while you still have it. Pull the demo and watch the sequence back from the other seat, because that is the one place your read gets tested instead of repeated. Report through the system that actually produces a record, so somebody with more than a feeling can look. And keep the name out of public until you have something a stranger could check for themselves.

Then go play. If you are the one somebody else is certain about, the record is already there, and you should never have to win an argument about your own honesty.

Sources: Gary Klein, Sources of Power: How People Make Decisions (MIT Press, 1998), chapter 4, where the fire case appears as "The Sixth Sense." Daniel Kahneman and Gary Klein, "Conditions for Intuitive Expertise: A Failure to Disagree", American Psychologist 64(6), 2009, for the two conditions, the fractionation of skill, wicked environments, and the finding that subjective confidence is an unreliable guide to whether a judgment is any good.