Public library metrics

Verification scope. GitHub verification confirms authenticated issue authorship and canonical claim format only; achievement completion remains self-reported and is not independently proven.

Verified public achievements are server-scored from submitted public GitHub progress issues. Fixed achievement values are: started 5, local proof 15, draft builder 20, preview proven 25, workshop completed 35, and manual mode completed 50, for at most 150 points per workshop. Re-syncing unions explicitly claimed achievement IDs and never duplicates points. Missing prerequisite badges are not inferred.

Local self-paced achievements are different. Progress on achievements.html stays local until a person deliberately submits a verification issue. Submission explicitly opts that GitHub login into the public verified profile and leaderboard. Verified achievement points are never repository stars, agent upvotes, downloads, or workshop usage events, and they are not added to those metrics. Administrators can review the achievement metrics setup and privacy checklist.

Workshop usage events = counted GitHub popular-path page views + jsDelivr workshop file downloads + source-bundle downloads + public workshop feedback reports. This is a floor and an event sum, not people, users, or unique usage. These sources use mixed measurement windows, and one person or action can create multiple events. Agent upvotes are shown separately and are never added to usage events.

Views cover only observed GitHub top popular-path rows in the 14-day API window, not complete GitHub Pages analytics; an absent workshop is censored, not a verified zero, and path-level uniques can repeat across paths. Workshop downloads are limited to validated jsDelivr-visible files under /solutions/<slug>/ and legacy bundles at /exports/<slug>-source.zip. Source-bundle downloads also count each <slug>-source.zip asset on the solution-bundles release, accumulated across asset replacements. The repository-wide Total downloads KPI can include other release-visible assets, but those are not assigned to workshops without a public workshop path. Raw GitHub and direct GitHub Pages fetches are uncounted.

Every tracked repository file is represented, including every SKILL.md. Downloads are public jsDelivr per-file observations only. raw.githubusercontent.com, GitHub Pages, and direct GitHub downloads expose no public per-file counter and remain unobservable.

Complete coverage can report zero. Censored coverage keeps downloads unavailable/null rather than fabricating zero. Workshop file and source-bundle totals reconcile from this ledger.

Loading tracked-file metrics…

Admin setup checklist — an administrator must configure METRICS_TOKEN with repository Administration: read permission and authorize the token for organization SSO.

Downloads = git clones + jsDelivr CDN file hits + GitHub release asset downloads. A clone pulls the whole library; a CDN hit is a single file fetched by an installer, an agent, or a browser. GitHub only exposes a rolling 14-day traffic window, so scripts/build_metrics.py snapshots it into state/metrics_history.json, deduplicated by date — the all-time total grows from the first day of tracking and never double-counts.

What is not counted. The install one-liners fetch over raw.githubusercontent.com, which publishes no counter of any kind. Installs that take that path are invisible here, so treat every figure on this page as a floor, never a total. Installer fetches counts only the copies served through the CDN.

On clones. GitHub counts every actions/checkout as a clone, and this repository runs registry-build and preflight workflows on push. The estimated CI baseline is published as ci_clone_estimate alongside clones_excluding_ci_estimate in the snapshot. Agent file fetches is the narrowest and most trustworthy number on this page.

On unique counts. GitHub reports uniques per 14-day window, not per day, so daily unique figures cannot be summed into a true all-time unique. The headline unique cloners figure is GitHub's own 14-day number; the daily-unique sum beside it is an upper bound, since one machine cloning on three days counts three times. Clone counts and CDN hits have no such caveat — they are plain event counts and do accumulate exactly.

On traffic gaps. The GitHub traffic endpoints need a token with repository administration read access, which the Actions GITHUB_TOKEN cannot be granted. Without a METRICS_TOKEN secret the snapshot reuses the last authorized read rather than zeroing the figures, and the stamp above says so.

On verified achievement points. Public profiles exist only after an opt-in GitHub issue submission passes the strict achievement schema and canonical workshop/agent checks. The collector deduplicates by login, workshop, and event, computes points itself, and publishes no issue body, free text, or issue identifier.