Corrections

What was wrong

auto-editor. Edition No. 2, written 27 August 2026, recorded version 31.5.0. A later note said that version did not exist and that the real one was 30.5.0, released 9 June. That note was wrong. 31.5.0 was released on 13 August 2026 and was the current version on the day the edition was written. The entry was right; the correction replaced it with a version two and a half months older.

umami. Edition No. 1 recorded v3.3.1, and it was corrected to v3.1.0, released 16 April. The same error. v3.3.1 was released on 20 August 2026 and was current. The correction replaced it with a version four months older.

Why it happened

Both figures were read from GitHub’s releases web page rather than from its API. The page returned a partial list, and both times that partial list was trusted over the figure already recorded.

The specific wrong assumption was that the recorded number had been a typing mistake. The note written at the time says so: “the 31.5.0 figure appears to have been a transcription error.” It was not a typing mistake. It was a real release that the web page did not show. Assuming the error was in the record rather than in the source is what made both corrections go the wrong way.

The same method produced every version number in the first thirteen editions, so this was not two accidents. Re-checking the whole archive against the API found 78 further versions that had fallen behind for the same reason. All of them have been corrected.

What changed

Every version is now checked on the day it is published — on the project’s own page, or against the package registry the project actually releases to where the two differ, because some projects publish to PyPI or npm and tag GitHub separately. Every entry carries the date it was checked, shown under the version. Where a project publishes no tagged release at all, the entry says so rather than showing a number.

All 219 repositories on file were re-checked on 7 September 2026. One of them, Castopod, could not be reached, and it is held back rather than published with a figure that cannot be confirmed.

A correction is trusted more than the thing it corrects, because someone has already looked once. That is exactly why it needs more checking than the original claim, not less. Both of these got less.