raea judge

Judge a 3D scene's review sheets blind, by a model on a frozen rubric or a fresh agent session, with the pass decided in code.

View as Markdown

raea judge reads the sheets raea render --review makes (colour.png, clay.png) and has them judged without the code or the conversation that made them: by clef-omni on a frozen rubric (the default), or by a fresh Kimi or Claude session on the judge prompt. The scores come back as data, and whether the scene passes is decided in code from them. clef only advises until it has been calibrated against a person's or an agent's eye.

Terminal
raea render scene.tsx --review review/
raea judge review/ --brief brief.ts
raea judge review/ --by claude --target reference.png --json
raea judge eye review/ composition=4 light=3 --pass yes      # record your own scores, for calibration
raea judge calibrate

Usage

Terminal
raea judge DIR [--by clef|kimi|claude] [--brief brief.ts] [--target T.png] [--json] [--dry] | judge eye DIR (CATEGORY=SCORE ... [--pass yes|no] | --by kimi|claude) | judge calibrate [--json]   (a --review folder's sheets judged blind: clef-omni on the frozen rubric, or a fresh kimi or claude session on the judge prompt; the pass decided in code; clef advises until calibrated against your eye or an agent's)

Headless: it needs nothing running and opens no window. Takes any number of arguments besides its flags.

Flags

FlagTakesWhat it does
--briefbrief.tsThe project's brief, so the judge knows what the scene is for.
--byclef|kimi|claudeWho judges: clef-omni on the rubric (default), or a fresh Kimi or Claude session. With judge eye, the agent whose eye is recorded.
--dry–Print what the judge would be asked, with its images, without asking.
--json–Print the verdict as JSON: each score, the defects found and the pass.
--passyes|noWith judge eye: whether you'd pass it, recorded with your scores.
--targetT.pngA reference image the colour sheet is judged against.