First defined in Outpaced by AI by Waydell D. Carvalho.
Outcome: A decision gets defended with an account that was never how it was made.
The explanation an organization offers for a decision is not the same as the mechanism that produced it. The first is built for the audience that asks. The second is whatever the system actually did. Post-hoc Rationalization is the gap between them, and it grows whenever the real mechanism is opaque and an explanation is still demanded.
COMPAS, a proprietary risk-scoring tool used in US courts, shows the pattern. In one Wisconsin case, a judge cited a defendant's high-risk score in handing down a sentence. The defendant had never seen how the score was calculated, and no one in the courtroom could explain it; the algorithm was a trade secret. The score arrived as a number, and the reasoning was supplied around it after the fact.
When the tool was challenged, every party had a different explanation of what it did, and each was internally consistent. The vendor said the scores were statistically calibrated. Journalists said the error rates fell unevenly across racial groups. The court said it was a validated tool to be used with caution. The judge had the number. The defendant had an opaque product that had called him dangerous. None of these accounts was simply wrong. None of them described the same object.
That is the structure of the failure. When the actual mechanism cannot be inspected, explanation detaches from cause and becomes a function of who is asking and what they need to hear. The decision is fixed; the rationale is fitted to it. Everyone in the chain can give a confident answer, and no one can give the real one, because the real one is not available to them.
AI makes post-hoc rationalization the default, because modern models are not built to expose their reasoning and organizations still have to answer for their outputs. The result is a layer of plausible explanation laid over decisions whose true basis no one can reconstruct. The danger is not that the explanations are dishonest. It is that they are convincing, they are accepted, and they were never how the decision was actually made.