> A format like YYYY-MM-SSTHH:MM:SS.SSS suggests to me as a physicist that it is an absolute time value, and I expect the number of digits to roughly represent the accuracy of this value.
As an amateur student of human nature, I would expect the number of expressed digits to allow precision somewhat in excess of "accuracy" only if the expresser is trained in, and/or respects the traditions of, the sciences where meaningless precision is avoided and eschewed. In this context, there is little reason to have such expectation. And, as is brought out elsewhere in this thread, "accuracy" can mean either "incremental precision" or "absolute accuracy", at the library user's option.
This "precision" or "accuracy" ambiguity seems like good enough reason for the SQLite library's julianday('now') to express whole seconds. Anybody who wants more precision at least is nudged to ask: Is there really any more meaningfully available?