The Imagination Trap

What cognitive science actually says about a leader's memory, and why the fix is the same one we build for machines.

The directive that was never given

A leader spends a Sunday evening working through a pivot. They map the sequence, decide who owns what, picture the conversation where they hand it over. It's vivid, it's detailed, and by Monday it feels settled.

Three weeks later the work doesn't exist. The leader is annoyed. The team is baffled, because nobody ever told them.

Everyone in that room is telling the truth. The leader genuinely remembers assigning it, the team genuinely never heard it, and the conversation happened exactly once — in one head, then filed alongside the ones that happened in rooms.

I've been on both sides of this. It isn't a failure of character on either side, and treating it as one is how it becomes permanent.

What the research actually says

Two findings matter here, and both are more careful than the way they usually get repeated.

The first is the constructive episodic simulation hypothesis, from Daniel Schacter and Donna Rose Addis. Recalling a past event and imagining a future one draw on a shared core network in the brain — medial temporal, medial prefrontal, midline and parietal regions. Memory isn't a recording being played back. It's a reconstruction assembled from fragments, and imagination runs on much the same machinery, because imagining a future event means recombining pieces of remembered ones.

It's worth being precise about the limit, because the popular version overshoots. Schacter and Addis titled their paper "Constructive episodic simulation of the future and the past: distinct subsystems of a core brain network mediate imagining and remembering." Shared network, distinct subsystems — not identical pathways. If you've read that the brain simply "cannot tell the difference," that's a stronger claim than the study makes, and worth being suspicious of whoever sold it to you that way.

The second finding is narrower and, for this problem, more damning. It's called imagination inflation. In 1996 Maryanne Garry, Charles Manning, Elizabeth Loftus and Steven Sherman asked people how confident they were that certain things had happened in their childhood, had them vividly imagine some of those events, then asked again. Confidence went up for the imagined ones. Picturing something that never happened made people more sure that it had.

That's the mechanism — not that memory is unreliable in some general, hand-waving way, but that the specific act of rehearsing something in detail makes it feel more like a thing that occurred. Which is precisely what a leader does on a Sunday evening when they think a plan through properly.

I want to be careful about the size of the claim. Imagination inflation is a shift in confidence measured in a lab, not proof that executives routinely hallucinate whole meetings. The honest version is narrower and still uncomfortable: the more vividly you plan something, the more it feels like you already said it.

Why this is worse for people in charge

An ordinary person misremembering a conversation has a disagreement. A person with authority misremembering one has a verdict.

The asymmetry does the damage. When a leader and a team member remember a conversation differently, the leader's version usually wins — not because it's better evidenced, but because disputing it is expensive. So the reconstruction becomes the record, and the team learns something corrosive: what happened matters less than what the boss recalls happening.

The same asymmetry runs through performance judgement. Recency effects are well established, and a single vivid failure from last fortnight will sit heavier in recall than three quiet months of competence — vividness is what makes a memory easy to retrieve, not what makes it representative. When retrieval is the assessment, the loudest fragment wins.

None of this requires anyone to act in bad faith. That's the part worth sitting with. Every person in the chain can be honest and the outcome is still unfair.

The fix is externalisation, and it's boring

There's no technique for remembering better. The reconstruction is how the system works, and you can't introspect your way out of it. What you can do is stop making memory load-bearing.

Write the decision down the same day, where the other party can see it. Not minutes, not a document nobody opens — three lines in the channel where the work already lives, naming what was decided, who owns it, and by when. The value isn't the record. It's that somebody who was there can read it and say that isn't what I agreed to, while the disagreement is still cheap.

Decide on the shared screen, not afterwards. If the owner and the date get typed while everyone is looking at them, the gap between what you think you said and what the room heard closes in the meeting instead of six weeks later. Ninety seconds, and most of the problem goes away.

Before you rule on a dispute, write your version first, then go and look. Put down what you remember, then read the actual thread, the actual commits, the actual dates. Not to catch yourself out, but because comparing the two is the only way to learn whether your recall of this kind of thing runs hot or cold. Treat the memory as a claim awaiting evidence rather than as the evidence.

The third one is the one people skip, and it's the only one that improves your judgement rather than just protecting the team from it.

We already built this for machines

Here's what made me want to write this down.

I've spent this year auditing software practices against a single question: did this exist because humans have judgment, or because human throughput was the bottleneck? One of the entries in that ledger is the fact graph — a store of claims a human has explicitly approved, which gates what a machine may assert, and which names everything it hasn't been told.

We build that for models because we don't trust a fluent system to distinguish what it verified from what it generated. The output is confident either way, and confidence carries no information about which one you're looking at.

That description fits the Sunday evening exactly. A leader running a plan through their head is a fluent system producing something indistinguishable, from the inside, from something retrieved. We already decided this problem was serious enough to engineer around — when the fluent system was a language model.

The discipline isn't new either. Ship logs, lab notebooks, contracts, minutes: every field that ever depended on getting the record right converged on writing it down at the time, in front of the other party. What's new is that we spent a decade calling that overhead, and are now rebuilding it for machines while still skipping it for ourselves.

What I'm not claiming

I haven't run a study. This is two published findings, one organisational pattern I've watched from both sides, and a practice I adopted because I kept being wrong.

I don't know how often a vivid Sunday plan actually converts into a false memory of having communicated it. The lab work shows confidence shifting; it doesn't tell you the base rate inside a real company, and anyone quoting you one is guessing.

What I'll defend is the asymmetry. Writing the decision down costs about ninety seconds. Being wrong about whether you said it — when you're the one whose version wins by default — costs somebody their quarter and some of their trust in you. You don't need a precise base rate to take a bet priced like that.

References

1. Schacter, D. L. & Addis, D. R. Constructive episodic simulation of the future and the past: distinct subsystems of a core brain network mediate imagining and remembering. Neuropsychologia, 2008. Also Schacter & Addis, The cognitive neuroscience of constructive memory: remembering the past and imagining the future. Philosophical Transactions of the Royal Society B, 2007.
2. Garry, M., Manning, C. G., Loftus, E. F. & Sherman, S. J. Imagination inflation: imagining a childhood event inflates confidence that it occurred. Psychonomic Bulletin & Review, 1996.