I’m in the prerelease program of Photoshop and I test my plugins against the Beta anyway, so I usually know what’s broken before a release ships. The Beta is open to everyone, but nobody should do real work in it, and for the release version the only signal you get is the Community board, in pieces, spread over dozens of threads. So I wrote something that reads that board every night and turns it into one answer.
Photoshop and Lightroom Classic so far, with a verdict per system: update, check known issues, don’t update, or too early.
The scraping is boring PHP on cheap shared hosting. Cron there can only call a URL, so the run is an endpoint that does a slice of work within a time budget, saves its position and returns. The next call continues from there.
The messy part is that bug reports have almost no structure. People write “27.10”, “latest”, “the new one”, or a version of something else entirely, like their macOS version, so every extracted version carries a confidence level. And half the reports never say whether it’s Windows or macOS, but a reply three posts down often does, as in “same here on Mac”. So the replies get read too.
Every issue then gets a severity and a guess at how many people it hits, done by an AI following a written rubric, and the worst open issue decides the verdict. The rubric matters more than the model. What counts at all is a separate question: an issue appears on the page once Adobe tracks it, or once three people have voted for it, but an issue Adobe hasn’t touched only counts toward the verdict from ten votes. Otherwise a single annoyed person with a broken preference file turns the whole release red.
Front end is static Next.js on Cloudflare Pages.
If you have five minutes, I’d like to know whether the verdicts match what you see in your own work. The ratings are the weakest link, and I only find the bad ones when someone who knows the app looks at them.