measure-complexity

Run radon cc with the -s flag to show numeric complexity scores per function, for example `uvx radon cc -a -s path/to/file.py`. The -a flag adds an average at the end, and -j emits JSON for machine processing.

Use the -n flag to set the minimum displayed complexity, so a threshold of 10 means `-n 11`, and add `-o SCORE` to sort by complexity descending. For machine-readable output, filter the JSON from `radon cc -j` where complexity exceeds your threshold.

Complexity of 1–5 is simple and fine as-is, 6–10 is moderate and worth watching, 11–20 marks refactor candidates, and 21+ indicates priority debt where bugs concentrate. The threshold is a judgment dial since parsers and state machines can legitimately score higher.

No, radon is not listed as a dev dependency in pyproject.toml. The recommended approach is running it via `uvx radon`, which downloads it into the uv cache without changing the repo; alternatively install it into the venv or add it with `uv add --dev radon`.

The maintainability index is a 0–100 per-module score computed from Halstead volume, cyclomatic complexity, lines of code, and comment ratio, obtained via `radon mi -s`. Rank A is above 19, B is 9–19, and C is 9 or below, with low MI signaling modules that are structurally hard to maintain.

High complexity is acceptable when the logic is inherently branched, such as a parser, state machine, or decision table. The scan is a locator rather than a gate, so read the flagged function first to decide whether it needs extraction, guard clauses, or a dispatch table.