The full record.

The first registered painting, T_001, is dated 29 December 2025. The idea was set down on 20 December; Peter dates the practical start to 2 January, the move into the studio.

All three dates are in the record. This site counts from the first painting and says so.

Within each rating system later paintings rate higher. The monthly average of studio-photographed paintings rises into June, the year's highest month, then settles.

The early months rest on V23 lite, read once from phone photographs, about 0.8 of a point rough per painting. The two systems are never averaged.

On 10 paintings held back from development and rated blind by Peter, rank agreement 0.95 and an average miss of 0.76 of a point. On the 13 development paintings, 0.86.

Ten paintings is a small test, and 0.76 just misses the 0.75 bar set before it. The engine reads three strong sheets lower than Peter, consistently. And Peter has ruled that the engine is the reference over his own eye.

The arithmetic is fixed and shown; the scale was set on randomly chosen unrated paintings; no constant was fitted to Peter's ratings; five reads are independent; a repeat pass correlates 0.985; every rating Peter gives is registered blind, before the engine reads that painting, so disagreements cannot be chosen after the fact.

Every rung is still a model's judgement of a photograph of the painting. Scale, sheen and presence do not reach the camera.

14,140 tests corrected for false discoveries, time held constant, batches tested as units, every source tested the same way (the sky included), and a placebo run alongside as a chance check.

0 placebo tests passed against about 16 expected by chance, inside the allowance: a reminder that some signals will be chance. Read a single finding as a lead.