The stats row counted voters, nominees and the last write, and never the votes themselves. It is the headline number of a poll, the one a reader means when they ask how it is going, and the site could tell you 1,348 people had voted without telling you what for.
It is one more aggregate over rows the query already scans, so it costs nothing:
COUNT(*) beside the COUNT(DISTINCT did) that was there, filtered by the same
status = 'counted' as every other total here, which is what keeps an
over-limit or self vote out of it.
The lexicon gains the field because it has to. Output is validated against it, so a number that is not declared cannot be returned at all — it would be a 500 rather than a missing key.
votes leads the row and takes the full width on a phone. Four cards into two
columns is three singles and one double, which leaves a hole wherever the double
falls; giving the headline the wide cell fills the grid and puts the biggest
total where the eye starts. The other double is "Last activity", and that one is
not cosmetic: before hydration it renders a whole UTC timestamp and only becomes
"9h ago" once mounted.
Two tests, because votes is the one total here that is not a distinct count
and so the one that would agree with voters if it were wrong. One asserts they
disagree — four votes from two voters — which a COUNT(DISTINCT did) copied by
mistake would fail and every other test in the file would not. The other pins
the status filter.
Co-Authored-By: Claude Opus 5 noreply@anthropic.com