“It is the peculiar and perpetual error of the human understanding to be more moved and excited by affirmatives than by negatives.”

— Francis Bacon

A room is about to approve a new initiative, and one person in the corner feels a cold flicker of recognition. We tried almost exactly this two years ago, she thinks. It failed. But the thought dies where it starts, because the memory is written nowhere. The person who ran the last attempt has left. The post-mortem was never filed. Nothing in the room will surface the old failure at the moment it matters. So the initiative is approved, on the same reasoning, heading for the same wall. And the strangest part is that the room is full of intelligent people. That was never the problem.

The error is not that people are unintelligent. It is that we have no way to let what we already learned defeat the story we now want to believe. That is a structural gap, not a personal one, which means no amount of individual brilliance will close it. So what closes the gap? Not brilliance in the room, but infrastructure around it. Call it evidence infrastructure: the standing machinery that keeps a causal story defeatable, that forces the room to test it against reality before it hardens into identity. Every causal trap you can name, a correlation read as a direction, a predictor mistaken for a lever, a filtered sample taken for the whole world, has the same cure. Not a smarter person in the room. A structure that forces the catch when the smart person is tired, outranked, or gone.

Why does this need structure at all? Because left to individuals, doubt always loses to momentum. The person who questions the favoured story pays a social cost, while the person who pushes it forward earns one, so rooms drift toward whatever is easiest to justify, and the first plausible story quietly becomes the only one allowed. Infrastructure fixes this by funding the dissent no individual can safely supply, and it works on two fronts.

The first is designed dissent, and the treatment can be brief, because the moves are known. Assign someone to argue that the plan is wrong, out loud, before it commits, so the strongest case against it is made by role and not by courage. That much a good room already tries to do.

The second front is the one almost everyone forgets, and it is where the effort belongs: memory. Most institutions have none. Not a shortage of documents, a shortage of usable memory, a living record of what was tried, what moved the outcome, and what did not, surfaced at the moment of the next decision instead of buried in a drive no one opens. Designed dissent catches the mistake standing in front of you. Memory catches the one you already made, and are about to make again.

What happens without it? We learn in episodes while the world compounds. Every team starts from zero. The confounder that fooled one division fools the next. The intervention that failed quietly last year is proposed confidently this year. And because no one checks the new story against the old wreckage, we mistake motion for progress, right up until the bill arrives. Worse, the causal story does not just persist. It becomes identity. Questioning it starts to read as disloyalty, so the very people who could trigger a rethink learn that rethinking is unsafe, and go quiet. That is the trap at full size: a capable organisation committing the same expensive error again and again, its confidence rising exactly as its memory fades.

So the move is to stop hoping a brilliant individual will save the room, and start building the structure that saves it by default. The good news is that it takes only a few pieces, and they are cheap beside the failures they prevent.

You will object that this is exactly the bureaucracy that strangles fast institutions, the premortems and red teams and review panels that turn a nimble company into a committee. Against bad infrastructure, you are right. A checklist long enough to feel responsible is just ritual, and ritual is what fast organisations correctly refuse. But real evidence infrastructure is the opposite of a long checklist. It is the minimum friction that blocks the most expensive errors, and nothing else. It does not cost you speed. It buys you speed, because the fast-and-brittle organisation pays later, and catastrophically, for the reversals and the funded illusions a little friction up front would have caught. The choice was never fast against careful. It is fast-and-corrigible against fast-and-brittle, and only one of them compounds.

So install the few pieces that earn their cost, and make sure of four things before anything large scales. Someone was assigned to make the strongest case that the plan is wrong, and you heard that case in the room rather than filing it. You checked the plan against your own memory, the prior attempts and the known counterexamples, so you are not paying full price to relearn an old lesson. The claim is written so anyone can later recover what was assumed, what the data was conditioned on, and what would have proven it wrong, and read by more than one kind of expert, so the design flaw and the plausibility flaw both surface before you commit. And one person you name holds the authority to pause the whole thing, without punishment, when the disconfirming signal finally arrives. Strip any of the four and the machine can see the error but cannot stop it.

Good judgment under pressure is not a gift a few people carry into the room. It is a structure the room carries for them, standing there before the decision arrives, ready to let reality defeat the story while the story is still just a story, and not yet who you are. So do not staff for genius and hope for the best. Build the memory, fund the dissent, make the claim checkable, and protect whoever calls the stop. That is how we learn faster than we forget.

Decision Rule — The Infrastructure Card

Before a major causal claim scales, require four things to be standing, not improvised:

• Standing dissent. Is arguing the strongest case against the plan an owned role in how we decide, not a favour asked of one brave person this once?
• Consulted memory. Have we checked our own record, the prior attempts and known counterexamples, before we spend to relearn them?
• Auditable claim. Can we state, in writing, what we assumed, what we conditioned on, and what would prove us wrong, read by more than one kind of expert?
• Protected revision. Who is authorised to pause or stop, without punishment, when the disconfirming signal arrives?
• Keep the gate minimal.

The moment it becomes paperwork, it has failed.