You shut the company down weeks or months ago, and you are still angry. Maybe at your co-founder, maybe at an investor who pulled at the worst moment, maybe at the market, maybe at yourself for the thing you still cannot name precisely. The standard advice is to process it, which is a noun in search of a verb. What the research actually says is more useful: the anger is not the problem you need to make go away. It is a scanner running the wrong search, and the work is to give it a better one.
Anger is a threat-detection system, not an emotional failure
The fast, automatic side of the mind, the one Kahneman called System 1, does not respond to failure with a neat narrative. It responds with threat appraisal: what here could hurt me again. Anger is part of that scan. It is the system flagging that something in the recent past is worth staying alert to. That is a legitimate function. It evolved to be useful.
The reason it does not feel useful after a shutdown is that it is imprecise. System 1 is fast and coarse-grained. It will identify the most salient person or event in proximity to the failure: the last difficult conversation, the most visible disagreement, the decision you keep returning to. Then it tries to lock onto that as the threat, whether or not that is the actual cause. You carry the anger toward something that may be at most one thread in a failure that had twenty.
That is not a character problem. It is how the fast system works. The slow, effortful system is what can actually evaluate cause. But the slow system rarely gets the assignment, because the advice to "feel your anger" does not hand the problem to the part of the mind equipped to resolve it.
The attribution, not the emotion, is where the distortion lives
Seligman's research on explanatory style maps how people attribute bad outcomes across three dimensions: whether the cause feels internal (my fault) or external (outside me), whether it feels permanent (this is how things are) or temporary (this was a specific situation), and whether it feels pervasive (this will keep happening in every domain) or specific (this one situation, this once). The variable that predicts how well someone moves forward is not the emotional intensity. It is the accuracy and specificity of the attribution.
This matters for anger directly. Anger typically sits with external attributions: something or someone outside you is coded as responsible. That can be accurate or distorted, but it is not inherently less healthy than guilt, which sits with internal attributions. Founders who flip from anger to shame in the belief that shame is more evolved have not found a more accurate attribution. They have traded one imprecise appraisal for another. Shame coded as "I am a bad founder" is a permanent, pervasive attribution, which is the one Seligman's work consistently links to longer recovery times. Anger that can be made specific is more workable than guilt that stays general.
The relevant question is not "should I be angry or should I feel responsible." It is: is the attribution accurate.
What the failure type tells you about the target
One reason the attribution stays imprecise is that most founders have not classified what kind of failure they actually had. Edmondson distinguishes three types: preventable failures, which come from known processes going wrong and which someone could have caught in advance; complex system failures, which emerge from multiple interacting factors with no single decisive cause; and intelligent failures, which are well-designed experiments that ran into genuine uncertainty and did not produce the hoped-for outcome.
Most startup failures are complex system failures. Many have a strand of intelligent failure woven through them. Very few are entirely preventable. Anger is a particularly bad match for a complex system failure, because complex systems do not have a single villain, a single decision point, a single moment where things could cleanly have gone otherwise. Anger searching for that moment will not find it, because it did not exist in that form.
This is not a reason to suppress the anger. It is a reason to classify first. If the classification reveals a genuinely preventable thread, something that was a clear error with information available at the time, the anger has a real target: a process to change, a judgment to revisit, a specific thing to not repeat. If the classification reveals a complex or intelligent failure, the anger loses its target not because you willed it away but because the terrain does not support one. Either way, you can move.
The question the anger is actually asking
Under the noise, the anger is asking a practical question: what do I avoid next time. That is not an unreasonable thing to want to know. The reason the question stays stuck is that the fast system, running on coarse threat appraisal, is not equipped to answer it. The slow system is. But the slow system rarely gets the assignment, because "process your anger" has come to mean "let yourself feel it," not "classify what kind of failure this was and what, if anything, is genuinely avoidable."
So the work is less emotional than it sounds and more cognitive than most advice suggests. It is not permission to stay angry and it is not a technique for releasing anger. It is a classification task: what kind of failure was this at the level of its causes, and what follows from that answer. Once the classification is done, the angry scan either finds something real to point at, or the target it was looking for turns out not to exist in the form it assumed. Both of those are movement.