Cortex install stats

How the estimate works. Every install fetches checksums.txt exactly once, so counting those fetches counts installs — on every platform, without double-counting the two binaries each install downloads. The release smoke test installs exactly as a real user would (curl | sh from GitHub Releases), so it lands in that number too and cannot be filtered out by behaviour. It can be subtracted by count: one job, no matrix, and one install per run lands in these figures — it installs three times, but on a push to main the other two land on main-latest, whose counts reset on every push and are excluded here.

What it still cannot tell you. Cortex Installs means "not CI" — it still includes maintainers and contributors, and at this scale a few of us on a few machines could account for most of it. Nothing here separates us from new users. And installs cannot show whether anyone kept using Cortex afterwards: it collects nothing from the workstation by design, so retention is deliberately unmeasurable from here.