zwitschi
abde390045
A single vision model has been wrong in both directions on this project - it missed real defects and disputed frames that were correct - so agreement between different model families is worth more than one confident opinion. judge_screenshot.py now runs two models by default (openai/gpt-4o-mini + qwen/qwen3.7-flash, both served on OpenRouter; claude-3.5-haiku is gone, which the first attempt discovered with a 404). combine() reconciles them: agreement carries the verdict with the mean confidence, disagreement reports 'uncertain' and keeps both reasons rather than letting the more confident model win. --model still forces a single judgement, and every line prints which model said what. comfyui_generate.py generates the same prompt and seed with two checkpoints, scores both by high-frequency detail (mean absolute Laplacian, which catches blurry or flat results), keeps the loser as <out>__<model>.png with its sidecar, and moves the winner onto --out. --alt-model restores the single run. Tests cover the reconciliation - agreement, disagreement, unknown verdicts, and no verdicts at all - and the suite is at 103.
Tooling tests
Standard-library unittest tests for the host-side tools in Tools/. They never start Blender, Unreal, or the game; anything that needs those processes is verified by the tool itself and recorded in Docs/release-checks.md instead.
Run from the repository root:
python3 -m pip install -r requirements-tools.txt
python3 -m unittest discover -s Tools/tests -t Tools -p 'test_*.py'
The -t Tools top-level directory is required: the tests import the tool modules by their plain names (import input_emulation), which is why Tools/tests/ carries an __init__.py.
Individual test:
python3 -m unittest Tools.tests.test_input_emulation -v
What belongs here
Test the pure parts — argument parsing, key/coordinate encoding, geometry maths, licence gating, report formatting, decision classification. Network, Blender and Unreal calls stay out: a test that needs a paid API, a display, or an engine process is a manual check, not a unit test.
When a tool's logic is buried in main(), extract it into a function so it can be tested directly. That is a tooling improvement, not test scaffolding.
Geo (Geo/tools/) and Blender (Art/blender/*/) builders are tested the same way once their pure parts are extracted; note the module path in the test.