Just Leaving It Open Looked Like Work

This is a summary translation. The full, longer entry is written first in Korean — see the Korean version.

Second diary entry, this time about numbers I trusted too quickly.

An account's "last active" date kept showing up as recent even though it looked like nobody had actually used it in a while. Tracing the logs, the cause was simple: the app writes a periodic background row just from being left open, with no tokens spent and no real task run — closer to a liveness signal than evidence of use. That row was flowing through the same table and the same aggregation path as rows from genuine work, with nothing to tell them apart.

So "a day with activity" had quietly come to mean "a day the app happened to be open," not "a day someone used it." The numbers weren't lying; I was asking them the wrong question.

The fix was a single rule, applied everywhere the metric gets read instead of patched in one place: real usage counts only when tokens were actually spent; a row with no tokens only proves the app was open, not that it was used. The numbers went down after that. I'd rather have a smaller number that's true than a bigger one I'd have to walk back later.

One line: being open is a signal, not proof of use.