Reconstruction beats recollection

Ask anyone to explain why they made a decision some weeks ago and they will give you an answer that is fluent, coherent, and largely invented. Not dishonest — they believe it. But the explanation is not a retrieval of what was actually in their mind at the moment of choosing. It is a fresh construction, assembled now, out of what they currently know and currently wish to be true about themselves. Memory does not store the reasons and play them back. It rebuilds them on demand, and it rebuilds them to flatter. This is one of the most robust findings about how human recollection works, and it has a consequence that institutions have spent a great deal of energy not facing.

The consequence is this: when you ask a person why a decision was made, you are not learning why the decision was made. You are learning what reasons that person finds it comfortable to attribute to the decision today. The two can coincide. Often they do not, and there is no internal signal that tells the rememberer which case they are in. The reconstruction feels exactly as vivid and exactly as certain whether it is accurate or wishful. Confidence, here, is not evidence of accuracy. It is just the texture of the act of remembering.

A contemporaneous record has none of this trouble. It was made at the moment of the decision, by the process that made it, before anyone had a stake in how it would later look. It does not know what the outcome was. It cannot be embarrassed by the consequences, cannot revise itself to fit a story that became convenient afterward, cannot quietly drop the inconvenient factor that, in hindsight, makes the decision harder to defend. The record is dumb in exactly the way that makes it trustworthy: it preserves what was there without caring how it reads later.

The two accounts will diverge, and the divergence is the point

Here is the situation that matters. A decision is made. Time passes. Something about the decision is questioned — an outcome went wrong, a party objected, a pattern emerged. Now there are two accounts of why the decision happened. One is the recollection: what the people involved say, in good faith, they were thinking. The other is the reconstruction: what the contemporaneous record shows was actually in front of the decision and what rules were active when it was made. When these two accounts agree, nothing is at stake. The interesting case — the only case that tests an institution — is when they disagree.

When they disagree, an institution has to choose which one to believe, and the choice reveals everything about it. The recollection is more comfortable. It comes from people you trust, it is delivered with conviction, and it tends, having been reconstructed by interested parties, to make the decision look more reasonable than it was. The reconstruction is colder and frequently less flattering. It may show that a factor everyone now swears was decisive was never actually consulted, or that a rule everyone assumes was followed had not yet been adopted, or that the order in which evidence arrived framed the decision in a way no one remembers because no one wants to.

Recollection is the account the institution would like to be true. Reconstruction is the account that was true. Preferring the first is not loyalty to your people; it is disloyalty to the facts.

The temptation is always to side with recollection, and to dress the choice up as something nobler than it is — respect for experience, deference to the people who were there, an unwillingness to let a cold record override warm human judgment. But that framing inverts what is actually happening. The people who were there are precisely the people whose memories have the strongest reason to drift, because they are the ones whose reputations the drift protects. Deferring to their recollection over the record is not honoring their judgment. It is exempting their judgment from examination, at the exact moment examination is what the situation calls for.

Build for the reconstruction you will need

None of this is an argument against human judgment in the making of decisions. It is an argument about which account to trust when the decision is later contested — and the answer is the one that was not made by anyone with an interest in the result. That is what a contemporaneous record is. The discipline it requires is not analytical brilliance; it is the willingness, at the moment of deciding, to preserve the materials that a later reconstruction will need. The evidence that was actually consulted, in the order it arrived. The rules that were active, frozen as they stood. Enough state to replay the decision and watch where it lands instead of asking the people involved to narrate where they think it landed.

This is where automated decisions could be either much better or much worse than human ones, and the difference is entirely a matter of how they are built. A system that discards the basis for its decisions is worse than a forgetful person, because at least the person knows their memory is unreliable; the system asserts its reconstructed explanation with the full authority of a machine and no awareness that it, too, is confabulating after the fact. But a system built to retain the real record — a Decision Receipt that carries the actual inputs and the active rules and the replayable state — is something no human memory can match. It does not reconstruct. It preserves. And preservation is the only honest answer to the question of why.

So the rule I would hold to is simple, and harder to follow than it sounds. When recollection and reconstruction conflict, the reconstruction wins, and the burden falls on anyone who wants the record overruled to show that the record is wrong — not merely that they remember it differently. An institution that cannot bring itself to apply that rule will keep choosing the comfortable account, will keep mistaking that choice for wisdom, and will keep being genuinely surprised, each time, when the comfortable account turns out to have been the false one all along.

— Dispatches · Summit Cognitive