A summary in English. The full record is in Chinese, at changelog/zh. Wall v30, record v34, this summary v1, 2026-08-19.
Every version of the Imagery Wall is logged with what changed, why it changed, and what the judgement behind it got wrong. Thirty-eight entries so far, covering fourteen days. This page describes the record and picks out four turning points. It is not a translation of it.
Each entry carries four fields: what prompted the change, what was done, how it was verified, and what was left unresolved or uncertain. The last field is the one that makes the record worth keeping. It is where the unverified assumptions, the changes made on the author's judgement rather than on evidence, and the things nobody has looked at in a browser yet all get written down.
Rejected options are logged next to adopted ones, in two lists: settled and open. Something that was ruled on and then reversed keeps both entries.
It is a working document in working register. It was written for the person building the piece and for whoever maintains it next, not for a reader arriving cold, and it reads that way: dense, abbreviated, and full of internal references. It also names which parts of the site's copy were drafted by an AI system and which decisions the author made, because that division is one of the things the project is trying to keep visible rather than tidy away.
Nothing in it has been reviewed by anyone outside the project.
Early versions animated the transformation: the apple visibly turning, peeling, splitting. A round of checking established that the classic experiments on mental transformation infer intermediate states from response times. They do not establish that anyone consciously watches a continuous motion, and no study was found that had systematically asked participants what the process looked like to them. Showing a smooth animation risked teaching viewers an experience they may not have. The animation was removed and the question was handed back to the user, who is now asked to report how the image arrived. The feature was a loss; the honesty was not.
The wall asks you to view a plate of apple slices from underneath. The reasoning was that spatial transformation is preserved in people with no visual imagery, so all six panels should complete the move. Checking found that the studies supporting that claim measured rotation of familiar shapes, and that rotating an object and taking up a new viewpoint separate in factor analysis. Nothing had measured generating a view nobody has ever seen. The geometry half of the claim survived; the optics half turned out to be an empty literature. The interaction stayed, its justification narrowed, and the gap is labelled on the page.
The site used a figure for how fast a mental image fades. Tracing it found the sentence verbatim in the introduction of an unrelated 2019 paper, attributed to a 1994 book with no page number; the book's own passage gives no figure at all. The value belongs to a different phenomenon: the persistence of something just seen, not of something generated. The number was overturned and the decay timings on the wall are now labelled as design choices carrying no citation. One link in that chain is still unverified and is recorded as open.
A note on the wall told readers, in both languages, that nothing had been checked against original sources yet. Four rounds of checking had happened by then and several statements on the wall had been corrected as a result. The note was wrong in the direction of understating the work, and for an English reader who could not open the Chinese ledger it was the only thing they learned about the evidence state. It was replaced, and a sentence neither version had ever carried was added: none of this has been peer reviewed.
The record keeps a list of unsettled questions rather than resolving them silently. Some are small, such as which of two wordings to use for a button that has been frozen since it started collecting data, where changing it means the data before and after are not the same instrument. Others are not small:
Per-claim evidence, confidence and read depth live on the evidence page, not here.
If you find an error, reporting it is not a favour. Write to lining.wang@niaur.com. Mail only, there is no form. Corrections are published alongside what they correct, credited to whoever found them if they wish.
Text and this record are licensed CC BY 4.0. Code is licensed MIT.