Three friends arrived to sit a week and then to speak, and speak, and speak. Each knew exactly why it fell. Not one was there to see it well.
Eliphaz, Bildad and Zophar come to comfort Job, and for seven days they manage it admirably by saying nothing at all. Then they begin to explain.
Their explanations are coherent and mutually reinforcing. Suffering follows wrongdoing; Job is suffering; therefore Job did wrong, and the sensible thing is to confess it. The logic is tidy and the rhetoric occasionally superb. None of the three witnessed the heavenly wager that set Job's ruin in motion. They are reasoning backwards from the outcome to a cause that fits it.
At the end of the book they are rebuked. "Ye have not spoken of me the thing that is right, as my servant Job hath," the King James text has it (42:7). The comforters had the better argument and the wrong account.
The rationale that arrives late
Organisations do exactly this, usually with better formatting.
Once the March model is retired, the October reviewer faces an awkward gap. The historical record shows what was done. The original reasoning, if there was any beyond the threshold, left with the system. And the hearing needs a justification, since Who Is the Explanation For? found that excuses alone cannot answer the claimant.
So someone writes one. They consult the policy, the risk framework, the model's validation report, and perhaps the successor model, which is fluent and very good at producing plausible rationales for decisions it did not make. The result is a clean, well-structured justification for the March determination, drawn entirely from documents that were never consulted in March.
It may even be correct, in the sense that the decision could have been justified that way. It is still an account written after the fact by people who were not there, and it describes a reasoning process that did not occur.
Resemblance again
The Container Is the Tell found a citation checker that certified invented references because their strings resembled real ones. A post-hoc rationale works on the same principle, relocated from citations to reasons. It resembles the justification the institution would have given, had it given one, and the resemblance is persuasive precisely because it is built from the institution's own materials.
The Accused Writes the Log examined the case where the system produces the evidence about its own conduct. This is the institutional version, and in some ways harder to catch. The log was at least generated at the time. The rationale is generated at the hearing, and it arrives with all the authority of an official document and none of the provenance.
The law already refuses it
American administrative law has a name for this move and a long-standing objection to it. In SEC v. Chenery Corp. (1943), the Supreme Court held that an agency's order could be upheld only on the grounds the agency actually relied on when it acted. A court could not substitute a better rationale supplied afterwards. The principle was restated forcefully in Department of Homeland Security v. Regents of the University of California (2020), where the Court declined to consider a justification memo issued months after the decision under review, describing it as an impermissible post hoc rationalization.
European law approaches the same point from the drafting end. Article 296 of the Treaty on the Functioning of the European Union requires legal acts to state the reasons on which they are based, and the case law treats those stated reasons as the ones against which the act is tested. A court reviewing the act examines the reasons given with it.
The underlying intuition is shared. Reasons invented later cannot be tested against the decision, because the decision was never made on them. Accepting them rewards the institution for having had no reasons at the time.
What an honest hearing says instead
The alternative is uncomfortable, and it is the only honest one. Where the original reasoning cannot be recovered, the institution says so, plainly, in the account given to the claimant: this is what the record shows and what can no longer be reconstructed, and here is what we will do about the gap.
The gap then becomes part of the case. It may justify reconsidering the determination on its merits. It may, in some regimes, weigh against the institution. It will certainly be less pleasant to read than a polished rationale. It has the singular advantage of being true.
The comforters were rebuked for speaking wrongly about the decision-maker. They were also, in their way, trying to help: they could not bear a suffering that had no reason, so they supplied one. Institutions writing reasons after the fact are often moved by the same discomfort. The cure is the same. Say what you know, and stop there.
Companions
- What the claimant is owed: Who Is the Explanation For?.
- Resemblance certified as identity: The Container Is the Tell.
- The system writing its own record: The Accused Writes the Log.
- SEC v. Chenery Corp., 318 U.S. 80 (1943); DHS v. Regents of the University of California, 591 U.S. 1 (2020); Treaty on the Functioning of the European Union, Article 296.
- The fiction: the Book of Job, chapters 4 to 31 and 42:7, King James Version, Project Gutenberg ebook 10.
These notes come out of Sociable Systems, a practice that reads AI-shaped documents the way a hostile reviewer will, before a lender or a court finds the gap. The argument has an operational form: the Interim Protocol sets out four rules for AI use in environmental and social deliverables, covering disclosure at touch-point grain, evidence custody, the phrases no automated screening may settle, and a hostile read before anything ships. Free, and written to be cited or retired once institutional guidance arrives.