Methodology

How every number on this dashboard is collected, stored and calculated. Last reviewed with data through 2026-11-17.

Data source

All traffic numbers come from the official GitHub REST API repository traffic endpoints: views, clones, popular referrers and popular paths. Repository metadata (stars, forks, topics, language, license, description, dates) comes from the repositories endpoint. Profile links come from the public GitHub profile, its social accounts and the profile README. Nothing is estimated, purchased or generated.

Why history has to be stored

GitHub keeps only the most recent 14 days of views and clones, and only the top 10 referrers and paths. This project saves what GitHub returns and merges it by date, so days that later fall out of GitHub's window remain available here. History starts on 2026-10-30, the first day of collection; earlier traffic cannot be recovered.

Update schedule and rate limits

A GitHub Actions workflow runs daily, plus a full refresh of every repository once a week. Each API call returns a rolling 14-day window, so repositories with views in the last 14 days are fetched every day, others every 3 days and dormant ones (no views for 45+ days) weekly, staggered by a stable hash of the repository name. Calls are sequential with a short delay, with retry and back-off, and the run stops and saves its progress if the remaining hourly budget falls below a safety floor. See GitHub's rate limit documentation. Visitors of this site never call the GitHub API: the page reads static JSON files.

Definitions

Score and tiers

score = views(30d) + 0.5 × clones(30d) + 2 × unique visitors(14d) + 3 × stars + 4 × forks. Clones have a lower weight because they include automation. Among public, non-fork, non-archived repositories the top 20% are labelled top and the bottom 20% low. The weights are a transparent heuristic, not an official metric.

Metadata audit

Each repository receives a 0-100 checklist score: description present (25) and 40-160 characters (10), topics (2.5 points each up to 20; 8+ recommended), license (10), homepage URL (10) and a push within 90 days (15). It highlights gaps in public metadata; it does not predict search ranking.

Insights

Insight sentences are produced by fixed rules from the stored numbers (for example momentum, concentration, busiest weekday, spikes above 2.5 standard deviations). A rule stays silent when there is not enough data.

Limitations

Optimizer

An automatic optimizer improves repository metadata. It is described on the optimizer page: what it may change, how AI output is validated and fact-checked, and how changes are rolled back.

Monitoring

Every run is logged on the status page. A failed or partial run, a large traffic spike and each finished weekly report create an alert issue in the repository, and a daily watchdog checks that the published data is fresh.

Integrity and changes

The code is open source under the MIT license at github.com/musman550/github-traffic-analytics-dashboard. Only the owner can change it; every data update is a visible commit. Raw data is available as JSON.