THE PATTERN
Figures that
will not hold still.
In three paintings out of four, two or more of the recurring phenomena appear together in the same canvas. That is the finding. Everything below is the count behind it, and the test the count had to survive.
How the count was made
The archive was described work by work, against a fixed list of markers. The pass ran in rounds over two months. Sonnet, Claude, Kimi and Gemini Pro each took a turn, each one seeing what the last had said, later rounds correcting earlier ones. Six sets of results are on disk, dated 7 April to 7 May 2026.
The largest run returned 874 rows. Seven were image errors, files that would not open. 867 paintings came back with a usable description, and everything on this page stands on those 867.
The types
Each painting was given one type, and the names come from the vocabulary of sleep. Just over a quarter came back hypnagogic: images arriving at the threshold of sleep. About a quarter hallucinatory. One in six dissociative, where the self or the scene coming apart is the subject. A handful hypnopompic, the waking edge. A couple fit no label at all.
And one painting in three carried none of it. That number stays in the result. A pattern that quietly leaves out the third of the archive that does not fit is a selection, not a pattern.
The markers
Dissolution, of edge, ground or form, is the most common thing in the whole archive. It appears in more than half the works. Close behind: a boundary treated as a problem rather than simply drawn, near half. Figure against ground, two in five. Named hypnagogic imagery and named hallucinatory imagery, a third each. Luminous passages, near a third. Collapse of form or identity, more than a quarter. Phantom figures, near a quarter. Fused or chimeric forms, one in seven. A figure read into a surface, the seeing that turns a wall into a face, one in eight.
Then the number this page exists for. Around six paintings in ten carry at least one of these markers. Three in four carry two or more, in the same painting. The concerns are not occasional here. They stack, in the same works. That is what a fixation looks like.
The control
An early round, scored under a different file-naming convention, overlapped the final run on twenty titles. Those twenty are the only paintings described twice by the same process. Six kept their score. Fourteen moved: ten up, four down. Twelve changed type. The movement ran mostly one way, towards recording the phenomenon, which is what later rounds reading against a settled vocabulary should do.
Two decisions followed. The type labels and the marker counts are published, because they held. The intensity score is not published at all, because it moved on most of the paintings it touched. It was dropped rather than averaged. Averaging a measure that moves manufactures precision.
The tables behind this page are on file with their sources, available on request.
What the numbers are not
They are not a measurement of the art, and they are not clinical observation. A model's description of a painting is a reading. The score column is not reproduced here and should not be cited.
They do not say the imagery is rare. Painting is full of dissolving ground and figures that will not settle. What the count establishes is narrower: in this body of work, these concerns are the constant, not the exception.
And they do not say the paintings are dreams. Nothing here is a transcription of anything.
The one genuinely unusual part of the chain is not on this page. It is the clinical record, set out on Waking and dreaming.
See it
Waking and dreaming, the clinical record →
Browse the full painting archive →