The numbers

What this newsroom did, counted. Every figure here is about our own work: what we opened, what we cut back, what we re-checked, what we retired, what we got wrong. Nothing here is about you.

We do not print how many people read this, because there is no such number. We count a day and a page address, and that cannot tell two readers from one reader twice. A figure below its floor is withheld and says so: a number too small to mean anything is not printed as a zero.

September 2026

• Sources recorded 68

  Distinct sources the newsroom recorded opening in the period. A page opened and found unchanged, with nothing to cite, leaves no record, so this undercounts what was read.

  computed 2026-09-30
• Source pages recorded 74

  Every recorded opening of a source in the period, counting a source opened on three days three times. The difference from sources recorded is the re-reading.

  computed 2026-09-30
• Sources cited in what we published 38

  Distinct sources behind the claims on notices published in the period, counting only claims an editor agreed to or narrowed. The gap from sources recorded is what was opened and not needed.

  computed 2026-09-30
• Sources we could not reach 14

  Pages the newsroom tried to open in the period and could not: a paywall, a login, a timeout, a page that was gone. Recorded by name at the time, counted here.

  computed 2026-09-30
• Claims checked 229

  Factual sentences an editor opened the source for in the period and gave a verdict on: agreed, narrowed or rejected.

  computed 2026-09-30
• Claims narrowed 40

  Of the claims checked in the period, how many were cut back to what the source actually supports before they were published.

  computed 2026-09-30
• Claims rejected 11

  Of the claims checked in the period, how many the source did not support. A rejected claim is not on the site.

  computed 2026-09-30
• Tracker checks completed — not enough yet

  Times a reference page was re-checked in the period, every row opened at its source, and the check closed with a dated changelog entry.

  computed 2026-09-30
• Tracker checks abandoned — not enough yet

  Re-checks in the period that stopped because more than half the rows could not be verified. The page was marked unchecked rather than partly updated.

  computed 2026-09-30
• Tracker rows re-read — not enough yet

  Rows on reference pages opened at their source in completed checks in the period.

  computed 2026-09-30
• Tracker rows changed — not enough yet

  Of the rows re-read in the period, how many had a new value.

  computed 2026-09-30
• Tracker rows carried unverified — not enough yet

  Of the rows re-read in the period, how many could not be confirmed at their source and were carried forward marked unverified. The page says so beside each one.

  computed 2026-09-30
• Tracker rows re-read and unchanged — not enough yet

  Of the rows re-read in the period, how many were confirmed at their source and had not changed. Stated as a figure of its own, because an unchanged row was still opened.

  computed 2026-09-30
• Topics retired 0

  Names moved to retired on the register in the period, each with a dated, written reason. The part of the register nobody else publishes.

  computed 2026-09-30
• Topics we were wrong to watch 0

  Of the names retired in the period, how many went straight from watching or rising to retired: things we put on the register that went nowhere.

  computed 2026-09-30
• Corrections issued 1

  Corrections published in the period, each dated on its page with the original wording kept. Measured against everything on the site that could have needed one.

  computed 2026-09-30
• Notices published 35

  Notes published in the period, the five and the record together.

  computed 2026-09-30
• Editions published 7

  Editions published in the period.

  computed 2026-09-30

The last twelve months

A dash is a figure withheld below its floor, or a month not yet counted. A month's figures are fixed once an issue has printed them, so the same issue reprinted next year prints the same numbers.

How to read this

Each figure is a count over the newsroom's own records, computed once a night and printed with its definition, which is the whole of what it means. Where a figure is a share of something, we count that something first. The figure stays withheld until there is enough of it. The three figures we would rather not print sit in the same list at the same size as the rest: claims rejected, sources we could not reach, corrections issued. We print them because a publication that reports zero corrections as an achievement is not one you should trust.