Backspace repository data

Every figure the archive holds, as plain tables. This is the same data the charted dashboard is drawn from, rendered without JavaScript so it can be read, cited and crawled directly.

coverage 2026-02-15 to 2026-10-22 · generated 2026-10-22T19:15:11.067Z · resolution daily

Page views

Daily views of the repository on GitHub, with the number of distinct visitors.

Clones

Daily git clone operations, with the number of distinct cloners. GitHub counts this repository’s own actions/checkout steps here too, so a day of heavy continuous integration inflates the figure well above human traffic; clones far above one per unique cloner is the signature.

CI activity

Workflow runs this repository started each day, published beside the clone table above because they are the reason it moves. A 0 here is a measured zero, not a gap: the Actions API is asked for a date range and answers it completely, where the traffic endpoints omit a day they recorded nothing for. History before daily collection began is reconstructed only as far back as GitHub still retains runs, because a day whose runs have been deleted cannot be told apart from a day that had none.

Stars

Cumulative star count, one row per day. A day on which nobody starred carries the previous day's total, because that total is known rather than missing. History before daily collection began is reconstructed from each star's permanent timestamp, which counts only stars that still exist, so those early totals are a lower bound on what the counter read at the time.

Forks

Cumulative fork count, one row per day, on the same basis as stars above: quiet days carry the previous total, and pre-collection history is reconstructed from forks that still exist.

Contributors

Cumulative count of people with at least one commit, dated by each contributor's first commit week.

Repository counters

Watchers, open issues, and release asset downloads. Downloads are split because GitHub counts an updater feed file the same as an installer: app is installers and archives, update checks is the metadata every installed client fetches when it checks for a new version.

Referring sites

Where visitors arrived from. GitHub reports this only as a trailing 14-day total per snapshot and lists at most 10 rows, so these are fortnight aggregates rather than daily figures.

Popular paths

The most-visited paths in the repository, on the same trailing 14-day basis.

Usage pings

These pings are opt-in. A self-hosted instance can choose to send one small ping a day; most instances do not, and an instance that never opts in is invisible here, so every figure in this section is a lower bound on the real network rather than a total. Counts are rounded to two significant digits on the instance and again on arrival, so they are the right size but not exact. An instance is counted only once it has reported on two separate days within the last thirty, which keeps a single curious request to the endpoint out of the totals. A version, country or client kind held by fewer than three instances is folded into other before anything is written here, so no row can point at one instance. A ping carries no domain, no instance or account name, no message content, no file names and no IP address.

Each row describes the fleet on its date, built from the most recent ping per instance in the seven days ending on that date, so a quiet instance keeps its last reported figures for up to a week. Three columns sit outside that window: instances reporting today and active today count only what reported that day rather than over the week, and within 30 days counts every instance seen in the thirty days ending on that date. An instance whose ping was late is absent from the first two and present in the rest; one silent for longer than a week, but seen inside the last thirty, is absent from all of them except within 30 days. Every column is a reading taken on that date, not a daily total: rows are comparable to each other but must not be added together.

Server versions

How many reporting instances run each release, at the latest snapshot.

Countries

Where reporting instances are hosted, as a two-letter country code the receiver derives from the connecting address. The address itself is neither stored nor published; a code it cannot resolve is recorded as ZZ.

Client kinds

How people reach their instance. This one counts users rather than instances: the fold threshold still protects the instance, so a kind is published only when at least three instances report it.

Releases