risk-synthesis-scan

Run a Sokrates analysis first, then this scan ranks files by complexity, recent churn, and size using the select_hotspots.py script. The AI then reads each shortlisted file and explains what it does, why the metrics matter, and what fix would help.

The script scores each file as (1 + max McCabe) × (1 + commits in 90 days) × log2(lines of code), so risk grows with both change difficulty and change frequency. With --src-root it also excludes embedded Rust test modules so test code does not inflate scores.

Yes, but degraded. The script warns on stderr, marks history_data as absent, and falls back to a complexity-only ranking. Churn, ownership, and coupling findings collapse into a single info finding stating those signals are unmeasured.

The script extracts top-contributor commit shares and lists single-owner production files over 200 LOC, including each owner's last commit anywhere in the repo. An owner who has gone quiet strengthens the finding; active owners lower the severity.

Pairs where a file lives in a same-named directory (Rust mod.rs or #[path] style) are flagged same_module_hint, meaning they are one logical module split across files. These are reported as info observations rather than boundary violations.