[m-messer]

Summary

• Criteria graphs (app/docs/criteria_graphs/): one example of each shape of criteria graph the contexts build (symbolic equality of expressions, equations and sets; where; written as; other; physical quantity matches and order). Each graph is built the way evaluation_function builds it and drawn twice: with graphviz straight from the graph's evaluation, criterion and output nodes, and with mermaid from CriteriaGraph.to_mermaid.
• Process graphs (app/docs/process_graphs/): example inputs run through evaluation_function (symbolic, physical quantity, and a response that fails to parse). The stages come from the stage_start events _evaluate logs, and the function calls are recorded with sys.setprofile. Each example has two graphviz images:
  • the stages, with the criteria graph the response went through and its path highlighted
  • the app and lambdafeedback functions called in each stage, up to two calls below _evaluate
• Generator scripts: scripts/generate_criteria_graphs.py and scripts/generate_process_graphs.py. No app code changes.
• app/docs/dev.md: new "Criteria graphs" and "Process graphs" sections, with how to regenerate the images. The stray ======= line at the top is removed.

Known issue visible in the images

The physical quantity criteria graph details show QUANTITY: response tags: {}. This comes from f-strings of ExprNodes in app/context/physical_quantity/graphs/match.py and is not changed here.

Test plan

• Both scripts run, and the SVGs were checked visually
• The highlighted path matches the returned tags; the parse-error example stops after parse_reserved_expressions
• flake8 --max-line-length=127 scripts/

🤖 Generated with Claude Code