Most AI food log demos are built around the easiest case. Good lighting. Clean plate. Full attention. Phone already in hand. Real usage is usually the opposite. You are standing in the kitchen. Or walking back to your desk. Or halfway through the meal before you remember to log it. The app gets most of it right, but misses the sauce, the second serving, or the drink. That is the moment that decides whether the habit survives. If fixing the meal feels annoying, people skip it. If they skip it a few times, they stop trusting the app. If they stop trusting the app, the feature list does not matter. That is the design constraint behind MetricSync. I am building it as an iPhone nutrition tracker that keeps logging flexible: photo when that is fastest barcode when the meal is packaged text when the camera path is slower than just typing it But the more important part is correction speed. The goal is not pretending AI will nail every mixed plate perfectly.…