Discovery data

The sources, coverage, and limits behind Popular and New packages.

Downloads

Popular uses completed transfers reported by Asm’s delivery service. The artifact’s storage provider does not supply the ranking counter.

Each event identifies the exact package, version, target, format, byte length, and SHA-256 digest selected for delivery.

A full response counts only after the service writes and flushes all declared bytes. This does not prove that a client saved or verified them.

Head requests, redirects, incomplete transfers, range responses, conditional responses, known bots, and internal validation do not enter Popular.

Repeated full requests count as separate transfers. Replayed event identifiers count once; conflicting event contents are rejected.

Unidentified automation can still contribute. Transfers do not measure people, installations, suitability, or safety.

The rolling window covers 30 days through the reporting watermark. The ≥ symbol marks an incomplete observation window.

New collection, server downtime, and reporting failures create coverage gaps. Missing reports never become zero downloads.

Only positive observed totals enter Popular. Equal totals use the canonical package name as a stable tie-breaker.

Downloads outside Asm’s observed delivery sources are outside this metric. Missing mirror reports do not establish global volume.

The local service records anonymous delivery events without cookies, fingerprints, IP addresses, user-agent strings, or search terms.

Reporting failures leave artifact delivery available and make the metric stale or incomplete. Search and exact package access remain separate.

New packages currently uses the first matching public GitHub release across each legacy package’s registered history.

These historical dates are separate from download measurement. GitHub download counters never enter Popular.

Recently updated uses the latest active release when its availability is later than the package’s first release.

Registry admission events require the separate authenticated publication service. Historical imports do not invent those events.

Yanking does not reset a package’s first date. Packages without active releases leave discovery while their version records remain available.

Sponsorship does not influence ranking.