After a company closes, a particular clarity arrives. You read back over decisions made months earlier and they look wrong in a way you feel you should have caught. The pivot you delayed. The hire you made when the numbers were already slipping. The funding conversation you kept alive past the point where the model was working. The errors arrange themselves into a pattern that feels, from where you stand now, almost impossible to have missed. That feeling is information, but not the information you think it is.
It is a named cognitive phenomenon, studied in detail for decades, and it does not tell you what your mistakes were. It tells you what your brain does after the fact.
How the past becomes inevitable in retrospect
The mechanism was documented by Kahneman and colleagues starting in the early 1970s, and the finding is consistent across many experimental designs. Once people know how a situation resolved, they substantially overestimate the probability they would have predicted that outcome in advance. Before an event, asked to estimate the chance of a particular outcome, people give a number. After the event, asked to recall what they estimated, they produce a number that is considerably higher. The original estimate is not retrieved from memory. It is reconstructed, and the reconstruction is shaped by the ending they now know.
The process is automatic and largely invisible. System 1, the fast and parallel part of the mind that Kahneman describes throughout his work, does not hold the raw past in storage and hand it to you on request. It assembles a plausible version of the past, shaped by everything you now know, including how it ended. The result feels like memory. In large part it is confabulation: a coherent story your brain builds backward from the outcome.
This is not a weakness or a sign of poor self-awareness. It is how the system works for everyone, operating on every kind of event. It fires especially hard on high-stakes outcomes you lived inside, which is precisely what a company you built and then closed is.
What "I should have seen it coming" costs you
The practical consequence, for a founder in the first 90 days after the close, is significant. If the failure was predictable, the quiet conclusion is: you were present at the decision, you had the information, and you chose wrong. Seligman's research on how people explain adverse events identifies this as an internal and stable attribution. Stable means it reflects something that persists, something about your judgment or character that will not be different next time without serious corrective work. That is the conclusion hindsight bias quietly installs, and it installs it without your noticing.
The attribution matters because it drives the corrective action you take, or do not take. A founder who concludes they were present for a predictable failure and missed it will often respond in one of two directions: extreme risk aversion, refusing to act without certainty they could not have had anyway, or redoubling on the same methods, convinced that more rigor will close a gap that was not actually there to close. Either way, the corrective is aimed at a diagnosis that was never accurate.
Why complex failures are not predictable from inside them
Edmondson's research on failure taxonomy offers the complementary piece. Complex failures, the category that includes most startup shutdowns, do not have a single cause that was legible at the time. They arise from multiple conditions converging: market timing, team configuration, product assumptions, external events, capital access, competitor moves. Each condition, examined individually at the moment a decision was made, may have been ambiguous or within an acceptable range. The convergence that produced the failure only becomes visible once you know where the system ended up.
The pattern looks obvious in hindsight precisely because you are reading the conditions through a frame that selects for the ones that matter, given the outcome. At the time, there were many other signals, many other plausible outcomes. The mind that was inside those months did not have the frame you have now. It could not. The information set was genuinely different, not because you were inattentive, but because the ending was not yet known and therefore could not organize what you were seeing.
Running the analysis without the contamination
There is a practical approach, and it is not an exercise in self-forgiveness. For each decision you now read as obviously wrong, reconstruct the information that was actually available at the moment it was made: what you knew, what the alternatives looked like, what comparable situations suggested at the time. The question is not "was this the right decision given what I know now" but "was this a reasonable decision given what was knowable then."
The two questions produce different answers, and only the second one yields a lesson you can act on. A lesson drawn from what was knowable at the time is the one that changes a decision you might face again. A lesson drawn from what is obvious now, with the benefit of a resolved outcome, is a lesson about a situation that no longer exists and will not recur in the same form.
The goal is accuracy. The hindsight version of the failure is both harsher than necessary and less useful than a real account. It attaches the moral weight of "you should have known" to a situation that was genuinely uncertain, and it produces a corrective aimed at certainty you cannot manufacture. The accurate version of what happened is harder to find and uncomfortable in different ways, because it requires holding complexity rather than settling blame. It is, however, the only version that tells you something worth carrying into the next two years.