Repository navigation
v3.10.0 — evidence and disclosure
Every score says what it looked at and what it came from, and a finding needs evidence. Found by re-scanning three projects with 3.9.3 and by an 18-repository precision study; each change lists what it moved in the CHANGELOG.
What moves when you upgrade
- Fewer findings without evidence: declared or local imports are not
phantom_import; placeholder findings respect declared interfaces and overridden hooks; clone clusters leave out protocol methods and per-class overrides; inflation counts claim words only; one finding per bareexcept: pass;mutable_default_argis HIGH unless the default is mutated or returned. - Scores move, mostly down; static data modules no longer dilute a project's weighted score.
scan_coveragediscloses unanalyzed code files, custom-rule exclusions, unevaluated tests, partially measured files and the checks a fallback analyzer could not run; a partially measured scan fails a hard gate.--initkeeps default weights and never writes function exemptions.
Upgrade notes
- Analysis cache recomputed once; history migrates to schema v6;
--self-calibrate(Calibration v2) starts from new scans. analyze_code_string(content)no longer reads the working directory; passproject_root=to analyze it as a project file.- Cold scans are 2.6-3x faster.
- The npm wrapper is not on the npm registry; install it from
npm-wrapper/.
Details: Release notes