Devlog #24 ->
Took a shot to the moon and posted on Chief Delphi. Got really specific, really helpful feedback but it made me a little sad, turns out both AI tools have real issues lol.
Icanari actually tested two robot photos and both scans came back wrong. Even flagged a robot’s signal light as an “unused emergency stop button,” which is actually dangerous if a team believed it. Tristan_Lall and philso both pointed out that the rules checker’s whole idea is kinda flawed too, a lot of what actually matters (obscured wiring, full range of motion stuff) just isn’t visible in a photo at all.
Stung a bit to read but honestly this is exactly why I posted there.
What I’m fixing:
Both AI prompts now say “worth double checking” instead of giving a confident verdict
No more guessing what a button/light/switch is, ever
Rules checker only checks stuff actually visible in a photo now (bumpers, frame perimeter), dropped the rest
Updated the hint text to be upfront about limits

P.S. not many commits on my GitHub right now because I remembered I can just run flutter build web locally to test changes instead of pushing every little tweak lol, so there’s gonna be one big commit instead of 100 tiny ones.

Also thinking about renaming since RoboLens is already taken by other stuff…