Ai-Q Labs

Small Actors that answer one question well.

Every Actor here does two things. It pulls records out of a public source — app stores, package registries, government registers, podcast feeds, web archives — and then it checks whether what those records claim is still true.

That second half is the part almost nobody does. A listing says "contact us" and the domain lapsed two years ago. A registry says a package is MIT and the repository relicensed after you installed it. A feed's build date says this week and the newest episode is from 2019. The record still looks fine. It just isn't.

How I try to keep these honest

• Base rates first. Before a finding gets a severity, I measure how often it occurs across a sample, and I publish that number in the README. A warning that fires on 30% of healthy inputs is not a warning.
• Declared is not inferred. "Archived" is the owner's own statement. "No push in a year" is my guess. They never share a severity.
• No means no. No logins, no proxies, no CAPTCHA solving, and nothing a site's robots.txt disallows. If a server refuses an automated request, that is the answer — not a problem to route around.
• Unknown stays unknown. If a source cannot be read, the row says so instead of quietly reading ok.

Feedback and feature requests are welcome on any Actor's Issues tab. If a check is missing, ask for it.