own commit attribution

There is no GitHost-style user manifest. GitHub stores one JSON blob per (repo, user, push).

CACHE_VERSION = v2
gh:attr:v2:{full_name}:{username}:{version_token}

version_token is repo.pushed_at, else updated_at, else "head". A new push changes the key. Old blobs die by TTL (default 7 days, ATTRIBUTION_CACHE_TTL_SECONDS).

A truncated measurement is returned for this response and not cached. Only a finished repo walk is written.

Deadline gates starting more HTTP, not in-flight work. /{username}/repos never walks. It reads the cache with cache_only=True and budget 0.

get_user_contributions lists repos (sort=pushed, type=all), skips archived (and forks if include_forks=false), newest first, then analyze_repo_contribution:

1. Cache hit. WalkProgress.resolved++.
2. cache_only or deadline expired. deferred++, return None.
3. List commit SHAs, maybe count via Link rel="last".
4. Fetch files per SHA. Skip vendored paths (node_modules, dist, lockfiles, minified).
5. Contributor totals for exact user vs repo additions.

RepoContribution.method:

• commits: full own-commit diffs
• estimated: sample mix scaled by contributor-stat additions
• contributor_stats: exact user line totals, empty language list

Whole-repo GET /languages is never used to invent a mix. tests/test_language_fallback.py fails if attributed mode touches get_language_stats.

status on the breakdown: complete, deadline, rate_limited, cache_disabled. coverage = resolved / (resolved + deferred). partial is true when deferred > 0.

python scripts/warm_attribution.py tashifkhan --passes 10 --deadline 60

Needs GITHUB_TOKEN and Redis. Default deadline is 60s, not the inline 3.5s. Or hit /tashifkhan/contributions/breakdown until coverage is 1.