Play counts told us something uncomfortable: 57 percent of all historical engagement was on games that no longer work. The obvious move is to find those games at a new address. We tried it properly on the 37 most-played games that died from link rot rather than from advertising.
The hosts are gone, not the paths
Fourteen of the seventeen GitHub Pages organisations our catalogue relied on now return 404 at the root. Not a moved file — the whole account. These were mirror repositories and the platform removed them.
That is worth checking first, because it changes the strategy entirely. If the host is alive you can hunt for a moved path. If the host is gone there is nothing to repair and the only option is finding the game somewhere else.
Sitemaps beat crawling when you have a list of titles
Fetching sitemap.xml from seven portals located 24 of the 37 titles in seconds. A breadth-first crawl of the same portals would have taken hours and found fewer, because the game pages are buried behind category navigation.
One detail cost us a match: our first matcher was strict, and missed a game whose slug carried a leading article. Match loosely, then verify.
The popular games are popular because they are monetised
Of thirteen candidates that survived to a real browser screen, ten were serving advertising. One portal was seven for seven. This is not bad luck. A game with five thousand plays is exactly the game a portal puts a video pre-roll on, because that is where the money is.
The games worth recovering are the games somebody else already decided were worth monetising.
Two traps in the extraction
First, do not take the first iframe on a page. Two portals we tested carry fifteen to twenty iframes per game page and not one of them is the game — they are advertising exchange frames, with the game itself drawn to a canvas. A naive selector imports an ad frame as a game.
Second, screen with software graphics rendering or your verdict is meaningless. Three candidates came back clean with almost no network activity. All three had rendered a browser not supported fallback. The game never started, so neither did its ad stack. Re-tested properly, one of them showed an error message and was correctly rejected.
What we would do differently
Spend the effort on new games rather than lost ones. The recovery rate was five percent, and the same time spent harvesting and screening fresh sources produced hundreds of working games. Sentiment about a title with good historical numbers is not a strategy.