Pure Macros

Subscription, not a one-time unlock. Each scan costs me money per photo, so one payment for unlimited scans falls apart quickly. On editing: nothing logs until you tap save. The scan fills in name, calories, protein, carbs and fat, and every one of those is an editable field sitting right there on the confirm screen. There's a note field too, prefilled if the model spots oil or a sauce, and you rewrite or clear it.

Whole plate, one entry. The prompt tells the model to combine whatever's visible into a single estimate. I did try splitting plates into separate items. Every split adds another portion guess, and portion size is already the shakiest part of a photo estimate, so four wobbly numbers replaced one wobbly number. Went back to one. Everything's editable before you save anyway, so if the model lumps things oddly you fix them.

Appreciate it, thanks.

Thanks. Keeping barcode and manual entry was deliberate. Photo scan is quickest when the food is obvious, and the other two cover everything else, a packet with a label, or the same lunch you log every week. On mixed meals and sauces, you get a confidence rating back with the estimate, low, medium or high. A low one throws up a warning and usually a note about oil or dressing. Numbers stay editable either way.

The flagging half is already built. Every scan comes back with a confidence rating, low, medium or high, and the model is told to judge two things separately: whether it knows what the food is, and whether it knows the portion. If either one is shaky, low. A low rating puts a banner up telling you portion size is hard to judge from a photo, check the numbers before saving. You also get a short note when something skews the estimate. Oil, a dressing, a portion bigger than normal. My favourite bit: if the model sees a sauce and doesn't know what the sauce is, it describes the colour instead of naming a type, because reddish tells you nothing about whether you're looking at hot sauce or BBQ. I picked a rating over a range because ranges invite people to log the bottom number. Corrections don't feed anything yet, every scan is independent. I'd like to fix this, and doing so properly on one person's data is harder than it sounds, so, not yet.

Homemade is the hardest case, and honestly the photo scan is the wrong tool for a lot of it. Two better routes. The scan screen has a note field under the numbers where you type whatever the photo missed, oil, sauce, whatever. And manual entry has a meal builder where you add each real ingredient with the amount you used and the totals work themselves out. I built the meal builder after a friend searched smoothie, got a figure for some packaged product with nothing to do with what he'd blended, and ended up doing the maths on paper. Portion size is the harder problem, not identification. A photo has no scale in it. One tablespoon of oil and three look identical.