Maintaining 322 Active Sources

Lists of username sites go out of date quickly. Platforms close, profile addresses change, APIs disappear, and public pages move behind login screens. Keeping every old entry would make PROBE look bigger while making its results less useful.

Two kinds of source

PROBE currently has 272 automatic sources and 50 that must be opened manually, for 322 functional sources in total. In October we converted 7 former manual-only sources to automatic checks after each passed three existing and three missing test accounts through a public profile endpoint whose terms permit this kind of lookup. We removed 44 sources that were consistently blocked from the production server, while retaining 42 blocked sources for targeted proxy testing. Automatic sources give us enough information to make a useful decision. Manual-only sources remain useful links, but without logging in we cannot reliably tell whether a username exists.

Test, do not assume

We test each automatic source with an account known to exist and unusual usernames expected not to exist. A source is fully verified only when it handles the available real account and both missing accounts without returning an unclear answer. The 3 October campaign ran before the catalog was trimmed. Recalculating that run for the 265 automatic sources retained at the time, 190 were fully verified; accuracy among clear decisions was 98.90%, coverage was 82.12%, and there were zero observed false positives. This is a re-analysis of the existing run, not a fresh campaign against the trimmed set.

A subsequent audit identified archive-promoted positive results in the underlying historical campaign. Those observations establish historical captures, not current account existence. The figures above remain historical diagnostic measurements, not an independent validation of live precision. Automatic scans no longer promote archive captures to found results.

The test accounts also need maintenance. During manual review, we found accounts that had been deleted or renamed. A failed test therefore tells us to investigate; it does not automatically mean PROBE is broken.

Keeping the list useful

The next step is to run these checks on a schedule and flag changes in profile addresses, page wording, response codes, or the number of unclear results. Suggestions from users can help expand coverage, but every new source should pass the same real-and-missing account tests before it is added to automatic scans.