When I start a project I find an app I like, screenshot six or eight screens, drop them in a folder. That's where it stops. What I saved is pictures, and a picture tells you what something looked like, not how it was put together. I've never once opened that folder again.
So our team tested whether the structure underneath could be pulled out of a store listing automatically. Paste the App Store link, it reads the screenshots and the description, and works out the screen structure how many screens, what sits under what, where the navigation goes.
One run, on a productivity app with 10 store screenshots:
14 screens, 5 paths, 58 components
9 screens were visible in the shots. 5 were worked out from the structure around them
It rated its own confidence 5/10 and said why: the tab bar was never fully visible in any screenshot, so the bottom navigation is a guess
The part I like is that last line. It tells you which pieces it wasn't sure about instead of handing you a clean map and letting you find out later.
The limit I don't think goes away: a store listing is the app on its best day. Empty states, failed payments, logged-out flows are never in the gallery, so they're never in the output. Doing this by hand hits the same wall, but a tidy 14-screen map makes it easier to forget the wall is there.
What I still can't answer is whether anyone wants this. So, two questions:
When you save a reference app, what are you actually trying to keep the visuals, or the structure?
If something handed you the structure instead of the pictures, would you use it? And if not, what's the real reason?
The confidence part is actually more interesting to me than the 14-screen result.
If the system can say “these 5 screens are inferred, and this navigation path is uncertain,” the output becomes much more useful than pretending the reconstructed app is ground truth.
I wonder if the next useful layer would be separating observed structure vs inferred structure explicitly in the output. That would make the map much easier to trust when someone uses it as a reference for building a new product.
Honestly I'd trust that map way more if I knew which parts were guesses.