Spring cleaning in October: what we retired and why
We asked three questions of everything we had published: does it work, do we use it, is it true. Here is what did not survive, and what the cull taught us.
A catalogue gets big enough that nobody remembers what is in it. On 2026-10-20 we went through everything published under the studio's name, on npm, on GitHub and on this site, and put three questions to each thing.
- Does it work today?
- Do we use it, or does anybody?
- Is the page about it true?
Anything that failed one of the three was rebuilt or retired. Nothing got a fourth chance because we liked the idea.
The numbers#
| Where | What happened |
|---|---|
| npm | 23 packages rebuilt and republished with honest descriptions, 11 deprecated |
| GitHub | 21 repositories deleted, 23 brought in line with the rebuilt packages |
| This site | every page about something that no longer runs removed, the old posts included |
The deleted repositories and the old pages sit in a private archive, so nothing is lost to us. It is gone for the public, which was the point.
What did not survive#
- Tools whose engine was a call to a hosted language model. They looked clever in a demo and did not give answers we would stand behind. A tool you cannot trust is worse than no tool, because people use it.
- Small tools and extensions that leaned on a backend we no longer run. The front half still worked. The back half was gone.
- Contest entries, a duplicate and a fork. Built for a weekend, never meant to be kept.
- Pages that described all of the above in the present tense.
Deprecate, do not vanish#
On npm the eleven are deprecated, not deleted. A deprecation is a message on the package page and it can be taken back. An unpublish cannot, and it breaks anybody who still depends on the package. The reversible option was good enough, so we used it.
What we learned#
A listing is a promise. Every page that says "this does X" is a small present-tense claim. When the thing stops working, the page becomes a lie that nobody maintains.
Run it. The method was dull: install each package fresh into an empty folder, run it, read what it prints. That found things reading the code did not. Some of the tools we kept still print a line of old wording in their own output, and we only know because we ran them.
Cheap to publish, expensive to keep honest. Putting a package on a registry takes an afternoon. Keeping its description true, its output current and its dependencies sane goes on for as long as it stays up. We now think about the second cost before the first.
The public list should be the whole list. There is no second shelf of "coming back soon" things. If it is not on the tools page, the scrapers page or the lab, it is not something we stand behind today.