Freshness is explicit
Data refreshes once a day. Every report shows the exact time it was generated.
Every number should be explainable.
Definitions, freshness, transformations, and known limits for repository analytics. No invented precision and no visitor-level claims.
Bundled methodologyGaps stay visible
A missing day means no trustworthy observation was stored. Missing values are not silently converted into zero activity.
Aggregate, not identity
Reports contain repository-level aggregates only. They cannot identify individual visitors, cloners, or README viewers.
Metric dictionary
What each signal means
Metrics retain the vocabulary of their upstream source. Derived values are labeled as derived.
Views
Repository page views reported by GitHub for each UTC day.
Unique views
GitHub's daily estimate of distinct repository viewers.
Clones
Repository clone events reported by GitHub for each UTC day.
Unique clones
GitHub's daily estimate of distinct cloners.
Stars
Net star gains derived from daily repository snapshots. Trends begin when you start tracking a repository, because GitHub no longer shares historical star dates with apps.
Forks
Net fork gains derived from cumulative repository snapshots.
Genuine subscribers
The repository watcher count from GitHub, excluding stars and social followers.
Release downloads
Downloads reported for release assets during the selected interval.
Package downloads
Downloads reported by the supported package registry during the selected interval.
Chart transformations
Same facts, different lenses
interval
Shows the value recorded in each reporting interval.
cumulative
Adds each interval to the running total.
normalized
Rebases each selected repository to 100 at the beginning of the window for relative comparison.
rolling
Shows a trailing average across the selected number of days.
Interpretation guardrails
Unique counts are upstream estimates and should not be summed into a person-level total. A README view cannot be identified. Correlation between a release and a traffic change is context, not proof of conversion. Anomaly flags are deterministic observations, not forecasts.