How we check
A directory is easy to copy. A record of having actually watched is not. Every listing here is scheduled for a fetch every 6 hours, every result is written to an append-only log, and everything below is read straight out of that log. How many sweeps actually finished is shown too, because a schedule is a promise and the count is the measurement.
The last two figures are measured, not promised. If a sweep is ever late, they say so before anyone has to point it out.
Those fetches identify as ItStillWorksBot/1.0 and point here. Allowlist that name if you want our checks to reach you. We still send a browser-shaped user-agent so a listing is not marked dead for blocking a bare bot string.
Not every listing is a page we can load. 233 listings are software installed on your own machine and 2 are repositories. We re-check installed software through its project site and a repository through GitHub, so for those a check says the site answered — not that the software ran. Cards, badges and the API say which was checked.
- Checks logged
- 104,801
- Web apps working at last check
- 467 of 486
- Last full sweep
- Sweeps, last 7 days
- 28 of 28
- Oldest check
- 25 days ago
Every listing, fastest first
Uptime and median here cover each listing’s last 14 checks. The listing page covers its last 40, so the two can differ.
Browser render check
An HTTP 200 says the server answered. It says nothing about whether the page rendered. A blank page that loads instantly looks identical to a working one from the network layer.
The render check loads each listing in a real Chromium browser, waits 6 seconds for client-side rendering to finish, and reads the resulting DOM. It records whether the page had meaningful text or interactive elements, and logs any JavaScript errors thrown during load.
What it confirms
The page loaded and rendered in a real browser, on the date shown. That is the claim and nothing more.
What it does not tell you
Quality, speed, security, or whether the app's features work. A page can render fast and still have a button that does nothing.
Route discovery
We collect links from the landing page and follow up to 4 of them. If the navigation is behind a menu or loads lazily, we may only check 1 page — we say so explicitly and never imply coverage we did not achieve.
Platform-embedded apps
Some apps run inside an iframe on a platform page (itch.io games, browser extension stores). We cannot read inside the iframe, so we record this as a coverage limit rather than a failure.
What we refuse to show you
Popularity rankings
We have no traffic data we would trust. Trending among what, exactly?
Usefulness scores
A 0–100 score built on a handful of saves is fabricated precision.
Guessed build attribution
An app only says it was built with a given AI tool if its creator said so. A wrong attribution destroys the trust we are selling.
Unconfirmed facts dressed as confirmed
Pricing and signup requirements we inferred are labelled as detected, everywhere they appear.