T_246
The work as shown. Every value below comes from vae.db, the master database of the Crucible Year.
Recorded
- Date
- 2026.04.21, day 114
- Size
- 70 × 180 cm
- Paper
- S3, Tanxi Jing Pi, raw
- Inks
- MB4, MB5, MB6
- Phase
- P2 The Switch
- Shown on the site
- yes (V23 rating of 5.0 or higher)
- Early storage triage
- SA: save, archive (Peter's early system, since retired)
Measured · V23 (v23.8)
- V23 rating
- 5.7
- Form
- 2
- Tonal depth
- 2
- Layering
- 2
- Composition
- 2
- Material and surface
- 1.8
- Ink cover
- 63%
- Dark share
- 36%
- Tonal steps
- 4
- Flaws named by the reads
- water stain 6, bloom 5
Peter's own rating
- Rating
- 5.5
“There is always something that can be done better.”Peter Yuill · 2026.07.02
Older engine
- V7
- 7
How the number was made
Read by the Visual Analysis Engine, version 23.8, on 2026.10.06: measured by code, then read 5 times by Claude Opus 5.5 against three reference paintings (R1 to R3), each read independent of the others. How the engine works.
Three views of one sheet.
Left, the photograph. Centre, the tone stretch the model is shown beside it: the darkest 45% of tones spread over the whole range, so flat black, banding and stains show. Right, every pixel as the measuring code classes it: cyan black ink, deep teal translucent grey, stone the palest ink, near black bare paper.
4 families of tone.
Every inked pixel's brightness in 64 bins, smoothed; each circled peak holds at least 4% of the ink. Life inside the black: 0.050 (under 0.045 reads as flat).
Five reads, agreeing on 4 of 5 qualities.
Each quality placed on its ladder by each read, and the mean the engine keeps. Where reads differ, they differ by one rung; close calls resolve down.
READ 1, IN ITS OWN WORDS
“A gestural crescent sweep with dark masses at both ends, much like R2. The arc and the open paper above it work, the left mass is blobby and cloudy, and the sheet is ordinary throughout.”
The crescent sweep holds as a coherent, nameable form and follows the brush's travel. The lower-left dark mass is a soft, lobed blob with dissolving walls.
WEAKEST Bulky blobby dark mass at lower left
STRONGEST Right crescent horn sweeping up to top right
Dark modulation of 0.050 is only just above flat. The right mass has stepped density and dry-brush streaks. The left dark mass is mostly a cloudy, even grey inside.
WEAKEST Cloudy, near-uniform interior of the left dark mass
STRONGEST Graded greys under the right crescent mass
A pale grey sweep, a mid-grey band and a dark top layer are all readable, especially along the crescent. At the left the passes run together.
WEAKEST Left mass, where the passes merge into a cloudy grey
STRONGEST Translucent grey sweep under the dark right mass
The crescent carries the eye from left to right, and the empty paper in the upper centre is shaped by it. The lower centre is busy with scattered spatter and drops.
WEAKEST Crowded spatter across the bottom centre
STRONGEST Crescent arc framing the empty upper centre
The ink is mostly controlled, with dry-brush and spatter used on purpose. There are faint pale water drops in the upper centre and a soft bloom edge at the lower left, all minor.
WEAKEST Bloomed, soft edges of the left dark mass
STRONGEST Dry-brush streaking at the right crescent end
4 distinct places, 0 major.
Each read must name every flaw and where it is, in a ninth of the sheet. Here the reads named flaws in 4 distinct places, 11 mentions in all across 5 reads; the number in each ninth is how many times a flaw was named there. Reads agreeing on a place strengthen the evidence for one flaw; they do not multiply it.
Measured from the 16-bit master.
The previous engine (V22) measured this sheet's master file: 2 principal ink masses traced (cyan), 9 enclosed voids at true size in their grid cells (amber), the main gesture from start to end (red). Of 17 edge segments, 41% crisp, 24% soft or bloomed, 35% broken.
Tonal depth held at rung 1.
4 steps: not appliedMaterial held at rung 1. Mild waves in the unmounted paper are never a flaw.
none namedThe rating cannot exceed 4.
lowest 1.8A read that breaks the reply format is discarded.
5 of 5 validV23 rating 5.7.
Engine v23.8, 5 reads. Fixed arithmetic, the same for every painting; the constants were set on 24 randomly chosen paintings so the median sheet scores 5. The scale is relative to this archive, not to painting in general, and a difference of less than half a point between two paintings is noise.