How the counting works

The single source of truth for these rules is analyzer/rules.go; this page is generated from those same constants.

What counts

• Author = your verified GitHub emails. Commits are deduped by SHA across every repo and fork (--all --no-merges).
• Lines = added lines from git log --numstat, code languages only — 56 recognised file extensions.

What is excluded

• Non-code: JSON, Markdown, Notebook, TOML, XML, YAML are not counted as code.
• Per file: a single file adding more than 5000 lines is skipped.
• Per commit: a commit touching more than 300 files or adding more than 25000 lines is dropped whole (bulk import) — and every dropped commit is listed on your card.
• Paths: dependency, build, output, vendored, generated and lock/minified/asset files are skipped by this pattern:

  (^|/)(node_modules|vendor|dist|build|target|\.next|coverage|bower_components|out|__pycache__|Pods|\.gradle|venv|\.venv)/|(package-lock\.json|yarn\.lock|pnpm-lock\.yaml|go\.sum|Gemfile\.lock|Cargo\.lock|poetry\.lock|composer\.lock|bun\.lockb?)$|\.min\.(js|css)$|\.(map|snap|svg|lock|csv|tsv|txt|log|pb\.go)$|_pb2\.py$|\.g\.dart$|generated|(^|/)[a-z]*aot/|vendor

What we can't see

Time in a chat or an agent session never reaches git, and unsigned agent code looks like human code — so the card says at least N% agent-signed, never "N% AI".

AI release timeline

Ticks on the monthly chart. These dates are unverified drafts — each still needs a vendor-announcement source before launch.

What we keep