Documentation debt

The accumulating gap between what a dataset or model actually is and what its documentation claims, growing every time the underlying data changes without a corresponding datasheet or AI bill of materials update; compounds the same way technical debt does, and is often unrecoverable once the people who remember the original context leave.

The deferred cost of writing documentation under deadline pressure, treated as a legitimate engineering tradeoff only when explicitly named, assigned an owner, and given a plan to close it, rather than silently passed to whoever needs the missing information later.

The gap between a written policy claim (a retention period, an access review cadence, a deletion commitment) and the operational evidence that would prove the policy is actually followed. Distinct from having no documentation at all, because it appears compliant until specifically tested.