A calorie tracker built around honest low/high estimates instead of fake single-number precision. The uncertainty is the product, not a limitation.
Every calorie tracker on the App Store pretends it knows exactly how many calories are in your meal. "A handful of chicken" becomes "184 kcal" and you accept it. The number is wrong, you know it's wrong, and you log it anyway because there's no alternative.
Range refuses to lie to you. You describe the meal in natural language, the AI parses it into food items with low and high estimates for calories and macros, and a short questionnaire narrows the wide ones. The result is honest: a range, not a number.
Before writing a line of native app code, I logged 21 days of real meals through web prototypes to prove the concept actually held up. Those 21 days are now seed data inside the app for testing History and Reports. The product wasn't designed in a vacuum. It was designed against real usage.
Phases 1 through 4 are complete and verified on device: foundation, logging flow with Claude parsing, History & Home with seed loader, theme customisation, and weekly Reports with AI insights. Phase 5 (polish: JSON export, onboarding, app icon, haptics) is in progress. Phase 6 (Vercel API proxy + Supabase multi-user) only happens if the app ships beyond my own phone.