Reachability

A registry can tell you what exists. It cannot tell you what answers. This page is the second number: of everything Wellknown has indexed, how much responds when something actually calls it. Every figure is a count of records over the same population, so any two of them can be divided by each other. Nothing here is declared by a publisher; all of it is measured.

Where the gap is

Two different things get called unreachable. Some records were never callable to begin with — they are libraries you install, and there is no address to try. Others had an address, answered once, and stopped. Only the second group is rot.

By source

Where a record came from, and how much of what that source lists is actually reachable. A record listed in more than one place appears in each of its sources, so these rows add up to more than the index.

By protocol

A record can declare more than one protocol, so these rows also overlap.

What the words mean

checkable

The record has at least one endpoint we are willing to probe. A record distributed only as an npm or PyPI package has no address, so it can never be live; counting it as unreachable would say something about packaging, not about reachability.

live

The most recent check succeeded within the last 24 hours.

recently observed

A check succeeded within 7 days, but the latest one failed or is older than 24 hours. 198 records.

unavailable

Two or more consecutive checks failed and none succeeded in 7 days. 4.5k records.

unknown

There is nothing to check, or nothing has been checked yet. 34k records. For most of these, the first is true.

Checking this yourself

Every record carries its own evidence: the checks behind its status, when each ran, and what changed. Open any row from the reliability report and then its history, or read the same facts as JSON through the API. The sources page lists where records come from and when each source was last read.

Measured continuously; this page is computed when you load it. As of 2026-09-19 10:10 UTC.