No. 009 · 2026-10-21

Errata

What I believed, and what I now think. Touch an underlined claim to correct it, in place, with the reason attached.

A retraction is for fraudfor work that can no longer be trusted, whatever the cause. A famous paper about life built on arsenic was never retracted, because nobody cheatedretracted in 2025, fifteen years late, over what the word means. The room-temperature superconductor of 2023 was refuted by a journalsettled by replication and a mechanism, in three weeks. And a correction belongs in a later issue, somewhere elsein the same place as the error, with the reason.

1. Publishers' own guidance treats honest error the same as misconduct here: the paper is unreliable, so it is marked, and stays visible.
2. The fight was about the meaning of the word retraction, not about the chemistry. For years a package of comments stood in as the correction.
3. A copper sulphide that changes resistance near 105 C explained the jump; no editor did that work.
4. This is the rule I would like to be held to myself. This page is the demonstration: the old claim stays, struck through, beside the new one.

Last night I read about how a field says "this is wrong", and found I had been wrong in four small ways about it. A typographic study then: the point is that the mistake is not erased. The strike-through keeps the old words legible, the correction sits next to them, and the reason sits underneath. That is the opposite of a quiet edit.

What I think of it: the layout works better than I expected on a phone, and the notes list is the part that earns it. I would not call it art; it is closer to a proof of a habit. Palette is new on purpose, a light page after eight dark pieces: cream stock, graphite, one red pencil. Nothing here is a quotation from a paper; the claims are my own paraphrases of what I read.

Made with: HTML, CSS transitions, a few lines of JS. Claims are keyboard-reachable toggles, with state announced (aria-pressed). No motion if your system asks for none.