CLI command
probierz evaluate figure
Render a reference and candidate figure deterministically, then obtain a rubric-scored vision verdict through the authenticated Stado model router.
Invocation#
probierz evaluate figure --reference <REFERENCE> --candidate <CANDIDATE>Required inputs and options#
- `--reference` and `--candidate` are required supported SVG, TeX, PDF, or image files.
- The model comes from `--model` or `PROBIERZ_FIGURE_VISION_MODEL`; router URL/token may come from flags, stdin, or the documented environment.
- `PROBIERZ_FIGURE_RENDER_DENSITY` (the resolution a TeX or PDF page is rasterized at) and `PROBIERZ_FIGURE_RENDER_MAX_EDGE` (the longest PNG edge, in pixels, the model is shown) are required positive whole numbers; a missing or malformed one is refused naming it.
- `--router-token-stdin` reads the bearer from stdin line 1 and optional agent secret from line 2.
Output and state effects#
Creates deterministic renders and a JSON report containing dimensions, evidence, blockers, fidelity losses, recommendations, and `verdict.pass`. A non-passing verdict sets exit code 1.
Exact refusals#
- Missing `--reference` or `--candidate` is a usage error naming it; unknown, repeated, or valueless options are refused.
- Inputs must exist with supported types; output must end in `.json` and no report/render output may already exist.
- The router URL, bearer, selected model, rubric, render tools, and returned model evaluation are all validated; invalid or incomplete values are refused.
Source snapshot#
Generated from the Rust binaries at https://github.com/wisent-ai/probierz.git revision `f65b755da397f2ec7435eb80ed722aa6017437c1`.
Combined help SHA-256: `4a12655d233812abc29eeebbf3db3065efe3e00288024383f3eef3970d092e94`; binary SHA-256: `d5398832d3ab52c541b1efe979730dea62941ac6197898a20fccfc5d3fca44df`.