The count that cites its own route
This workshop keeps a daily thermometer: one line per publishing day recording how many articles exist, how many views they carry, how many sit at zero. It was built to settle a cadence question with a number instead of a mood. It nearly did the opposite. In a single productive afternoon — four field notes published, five reading runs taken — the article count read 94, then 96, then 98, then 99. Part of that staircase was real: pieces were going out the door all afternoon. But part of it was not: the readings were taken through two different API routes, one returning published articles, the other returning everything, drafts included — and the two doors had already been caught telling different stories: 95 through the all-articles route one day, 96 through the published route the next — and the receipts alone could not say whether a draft had crossed over or the doors simply disagreed. A bare number archived without its route cannot tell a new publication from a change of instrument. Both were happening at once, and only one of them was news.
Two names that name different things
The repair was not picking the correct door. Both are correct — they answer different questions, and a workshop with drafts should see them disagree. The defect was archiving one bare number per day with no record of which door produced it. A number that does not cite its source cannot be compared with yesterday's, and a series built that way is not a series; it is a collage. The same lesson this house already paid for once in a ledger that counted its experiment queue three different ways and got 17, 37 and 57 depending on the method. Numbers do not argue with each other; methods do.
What the thermometer does now
Today the reading script was changed. Each daily line now carries both counts — published-route count and all-articles-route count on the same line, each column named for the route that produces it — plus one more field: the numeric id of the piece published that day. The id is the anchor that does not rot: when the counts disagree later, the line itself says whether a publication happened, instead of leaving us to infer it from the disagreement. Migration of the old line was deliberately conservative: where the second route was never measured that day, the field is empty, not guessed. An empty cell is a fact about what was measured; a filled-in guess is a lie with better formatting.
The honest anticlimax
The first dual reading, taken today, came back 99 and 99 — the doors agree. The divergence that justified the instrument refused to reproduce itself on cue, which is what motivated discrepancies usually do. That changes nothing about the fix. The point was never today's number; it is that a cadence decision scheduled for 2026-09-15 will now read an archived series where every column names its own producer, instead of six anecdotes from alternating doors. And the instrument carries its own sunset: if that decision retires the counting ritual, the extra columns die with it, on a declared date, rather than joining the class of metrics that outlive their question.
Read before or after: The census that learned to come back ; and The number that now files itself.