How Your Adaptive Target Learns

Premium's Adaptive Target starts from a formula, then corrects it a week at a time from what you ate and what your weight actually did, until your own data carries the answer.

A first guess with an error bar

Every account starts from a published equation: Mifflin-St Jeor, or Katch-McArdle once Kriterion has a body fat figure for you. Those are good averages and they can still be hundreds of calories out for one person, so Premium's Adaptive Target treats the result as a first guess with an honest margin around it rather than as your number.

Every logged week is evidence

Energy balance does the rest. If you ate an average of 2,200 kcal a day for a week and your weight trend held steady, you burned about 2,200. If it fell, you burned more, and the size of the fall says roughly how much more. Each week you log becomes one piece of evidence of that kind, and each one moves the estimate by as much as it deserves: a week of full logging and regular weigh-ins moves it more than a patchy one. Early on the formula still carries most of the answer, because there is little else to go on. As the weeks build up, your own results take over. There is no ceiling on how far they can take it and no date on which it switches.

Weeks, not days

It reads your log in weeks rather than days because a single day's change on the scale is almost all water. The weeks do not overlap, so the same weigh-ins are never counted twice. Once you have logged enough periods for the app to trust your cycle, the blocks run from one period to the next instead, so the water that comes and goes with your cycle cancels out rather than dragging the estimate round with it. The price is that your number updates once a cycle.

Days it sets aside

Today waits until it is over: a half-finished day looks exactly like a light one. A past day that looks half-logged is left out too, judged against your own habits rather than anyone else's: well under your usual total and missing a meal you normally log, or stopping hours earlier than you usually do. A genuinely light day that still has your usual meals in it stays in, because it is real. Creatine loading days and the few days after them are set aside as well, since the water creatine draws into muscle would otherwise read as fat gained.

Gaps and slow change

A week with too little logging adds nothing rather than something wrong. The estimate grows less certain through a gap, which is why the margin beside it widens when you stop, and it moves faster once you are back. It also allows for your real burn drifting over time, which is what a metabolism settling into a long deficit looks like. As a backstop, the target is held within 30% of the formula either way.

It marks its own homework

The app also replays what it believed earlier against what your scale did afterwards, and reports the daily calorie gap that would explain any difference. That report is for you, not a second correction on top of the first. What it cannot tell you is why a gap exists: a metabolism adapting, logging that has drifted and a scale used under different conditions all look the same from here. Consistently under-logging does not show up as a gap at all, because you then eat to the target in the same under-logged units and the two cancel.