Every Publifye server runs one small daemon, walle. It measures its own disks, memory and pressure every fifteen minutes, cleans up by rule, and reports what it did. These are the fleet's figures, aggregated and rounded; nothing here identifies a server.
Data cleaned: moved to walle's trash, deleted outright by a rule, or freed by the emergency ladder, summed; emptying the trash is not counted again.
Bytes walle moved to its trash, deleted outright by a rule, or freed in an emergency, summed per month across every server; emptying the trash is not counted again, because those bytes were counted when they were moved. A sum, never an average. These are calendar months; the 365-day tile is a rolling window, so their totals differ. October is drawn hollow: the month is not over.
walle cleans only what a written rule names, such as build caches that have grown past their cap. Some cache rules delete directly, where the cache rebuilds itself. The rest go to walle's trash first and are purged for good only once they have sat there long enough to be missed, so a mistake there can be undone for days.
If a disk heads for full faster than the daily clean can keep up, walle climbs a ladder of increasingly firm, pre-written steps, so the server does not reach the point where nothing can write. Every step is recorded.
When something needs a person, walle says so once. It speaks again only if the problem gets worse or comes back after it was fixed. An alert that repeats every few minutes teaches people to stop reading alerts.
A backup on the disk it protects is lost with it. Copies go to another disk, and the history behind these figures is also kept offsite.
One walle per server, owning that server's disks. Two cleaners on one disk would fight, and nobody could say which one did what.
Figures here are sums and medians, never averages: an average hides the one bad day. Coverage is the share of time the servers reported in, not a claim about uptime. A gap is shown as a gap.