Six weeks after the launch failed, the team sat down to talk about what went wrong. Within twenty minutes, someone said the sentence that ends every honest conversation before it starts: We should have seen this coming. Heads nodded. Someone half-smiled at the ceiling. And a person who is very good at their job — who, back in March, made a defensible call with the information on the table — quietly began building a story about themselves in which they are not.

Here is the uncomfortable part. Almost nobody in that room did see it coming. You can check. The Slack threads exist. The doc comments exist. Nobody wrote this will fail in March. What changed between March and now is not the evidence. It's the memory of the evidence — and human memory, once it learns how a story ends, cannot be trusted to describe how the story looked from the middle.

Creeping determinism, the thing eating your retros

In 1975, the psychologist Baruch Fischhoff gave people historical scenarios with genuinely uncertain outcomes and asked them to estimate the probability of each possible ending. Some participants were first told which ending had actually occurred. Those participants rated that outcome as far more predictable than participants who didn't know it. Fischhoff called the effect creeping determinism: once we know what happened, the past reorganizes itself into a straight line pointing at it.

Worse, in a companion study with Ruth Beyth, he asked people to predict outcomes of Nixon's 1972 diplomatic trips to China and the Soviet Union — and then, weeks later, to recall the predictions they themselves had made. People misremembered their own forecasts, shifting them toward whatever had actually happened. Not other people's forecasts. Their own. Written down, in their own handwriting, weeks earlier.

That is the mechanism running underneath your Friday retrospective. Psychologists Neal Roese and Kathleen Vohs, reviewing decades of this work, describe hindsight bias as operating at three levels that stack: memory distortion (I said it would go badly), inevitability (it was always going to go badly), and foreseeability (I knew it would go badly). Each one is a small lie, and together they make an honest reconstruction of a decision essentially impossible without help.

The specific damage this does to a team

Start with who gets blamed. Judging a decision by how it turned out rather than by how it was made has its own name — outcome bias — and Jonathan Baron and John Hershey demonstrated it cleanly in 1988. They gave people descriptions of decisions, including a surgeon choosing whether to operate. Same information, same reasoning, same probabilities. When the patient died, people rated the surgeon's decision as worse. Not the luck. The decision.

Your retrospective does this constantly. The engineer who shipped on a reasonable estimate and got hit by a vendor outage is discussed differently from the engineer who shipped on the same estimate and got lucky. Both made the same call. Only one of them has to sit through a meeting about it.

Then consider what the team learns. If the failure was obvious, there is nothing to learn about process — only something to learn about the person who missed the obvious. So the retro produces no systemic change, just a faint social residue. And the people who watched it happen draw the lesson that actually gets learned: do not be the one holding the decision when the dice land. They start hedging. They stop putting their name on hard calls. They add a reviewer nobody needed. The organization gets slower, and no one can point to the meeting where that was decided, because it wasn't decided in a meeting. It was absorbed from one.

The cruelest version is private. Hindsight bias doesn't only run in groups. It runs in the shower at 11pm, when you replay the call you made and cannot access the version of yourself who made it — the one who didn't know. You are, quite literally, judging a stranger with information he never had. You'd never do that to a colleague.

Foresight leaves a paper trail. Memory doesn't.

The fix is not try to be objective. Fischhoff's later work found that simply warning people about hindsight bias barely dents it. Knowing about the distortion does not undo the distortion, in the same way that knowing about optical illusions does not make the lines look equal.

What does work is refusing to rely on memory at all. If a decision's inputs — what we believed, what we predicted, what we were unsure about, what would have changed our minds — exist in writing before the outcome arrives, then the retrospective becomes an act of comparison rather than an act of recall. You are no longer asking did we know? You are reading what you knew.

There's a second lever, and it's stranger. Gary Klein's premortem asks a team, before a project begins, to imagine it is a year later and the thing has failed catastrophically — and then to write the history of that failure. Klein points to work by Deborah Mitchell, Jay Russo and Nancy Pennington finding that prospective hindsight — imagining an event has already occurred — substantially improved people's ability to generate correct explanations for future outcomes. It's the same cognitive machinery that corrupts your retrospective, pointed forward, where it becomes useful. Certainty about the past is a bug. Borrowed certainty about the future is a tool.

The third lever is the oldest one in the debiasing literature: consider the opposite. Explicitly generating reasons an alternative outcome could have occurred reduces hindsight bias, because it forces the mind to rebuild the branches it pruned. In a retro, that means one person is assigned, on purpose, to argue that the decision was good and the outcome was bad luck — and to be taken seriously.

Your next moves

  • Write a one-paragraph decision record before every consequential call. Three lines is enough: what we believe, what we predict, and what evidence would change our minds. Timestamp it. Put it where the retro will find it. This single artifact does more than any facilitation technique, because it makes memory distortion falsifiable.
  • Open the next retrospective by reading the original decision doc aloud, before anyone discusses the outcome. Not summarized. Read. If no such doc exists, spend the first ten minutes reconstructing what was known without mentioning what happened — and notice how hard the room finds this.
  • Ask one question in every retro: "With only what we knew then, would we make this call again?" Separate that answer from "did it work?" Sometimes a good decision has a bad outcome. A team that cannot say that sentence out loud is a team that will start optimizing for defensible decisions instead of correct ones.
  • Assign a devil's advocate for the defense. One named person argues the decision was sound and the world was noisy. Rotate it. It's a role, not an opinion, which is exactly what makes it safe to play.
  • Run a fifteen-minute premortem on the biggest thing currently in flight. Prompt: It's six months from now and this failed badly. Write the story. Everyone writes silently for five minutes before anyone speaks — otherwise the first voice anchors the room. Then collect the failure modes and pick the two you can actually cheapen or prevent.

The version of you who didn't know

Be gentle with the person who made the call in March. He was working in fog, with a deadline, using the best map anyone had drawn. You have the luxury of standing in the clearing and calling the fog a personal failing. That isn't wisdom. It's a trick of the light — and a team that learns to name it becomes, over time, a team where people are willing to decide things.

This is the whole reason MeetingMortem exists: to hold the record of what a room actually knew, decided, and predicted, so that when the outcome arrives the conversation can be about the process instead of the memory of the process. Notes taken before the ending are the only ones that stay honest. If your retrospectives keep concluding that it was obvious, have a look at meetingmortem.lumenlabs.works — or just start the decision doc yourself, today, and read it back in six weeks. You'll be surprised who you were.