Corrections

The ledger

1. 0001 The dissent rate was divided by a denominator that included non-votes Recorded absences and pairings are rows in the source table but are not votes cast. The headline figure was understated by about a fifth — 0.120% published, 0.143% correct. It was also converting a difference in attendance into an apparent difference in independence between two parties. Partially applied2026-08-06
   Headline · arithmetic

One entry. This is the first, filed the day the process was written, on a site nobody has been told about yet.

It is not finished. The front page and /design carry the corrected figures; /discipline is generated from twelve queries that all share the fault and is still serving the withdrawn denominator. It is labelled as such wherever it is linked, rather than left to look correct. Rule 07 is the reason there is no promised date attached to that sentence — and the count above is the reason not having one is not a way of avoiding it.

How this works

The rules, before there is any pressure on them.

Written now, while there is nothing at stake, because a corrections process first used under pressure is not a process. It is a press release.

1. Every correction is permanent and numbered. Entries are never deleted, never renumbered, never silently amended. If a correction is itself wrong, that gets its own number and both stay up.
2. The withdrawn figure is shown at the size it was published at. Struck through, beside its replacement, on the same page. A correction set in smaller type than the error is a retraction pretending to be a footnote.
3. Corrections are counted, never scored. There is no accuracy rating, no health score, no single number summarising how honest this site is. Counts can be checked; a score is an opinion with a decimal point, and it would be the easiest thing here for an opponent to attack.
4. The query that produced the wrong number is published beside the right one. Both, in full, runnable against the same hashed snapshot. A correction you cannot reproduce is an assurance, not a correction.
5. How the error was found is part of the record. Found by us, or reported by a reader, or forced by someone who disagreed with the result. Named where the finder wants to be named. The useful measure of this site is not how few corrections it has, but how many were found here rather than by someone else.
6. Near misses are filed too. Errors caught in draft, before publication, are recorded on the correction they were caught during. A page that only admits the mistakes that escaped is not being honest about how mistakes work.
7. No service-level promise. There is deliberately no published target for how fast a correction gets resolved. A stranger who wanted to make this site look negligent could simply file enough reports to breach it. What is published instead is the count still standing, and the age of the oldest one.
8. The record is a file, not a page. Each correction is a JSON document with a SHA-256 over its exact bytes as served. The HTML is a rendering. Anyone can verify the record independently of how it happens to look today.