Your weight moves 0.5 to 1 kg between two mornings on an ordinary day. Your fat moves 0.1 to 0.15 kg a day if you are in a solid deficit. Every weight tracking app shows you the first number. Good ones smooth it into a trend. AiTrainingPlan goes one step further: it tells you the evening before what tomorrow should read, so a heavy morning after a carb refill is expected rather than alarming.
What you see every day
- Tonight: tomorrow’s predicted fasted weight and a range, with the reasons. For example: “87.3 kg (86.8–87.9). Trend 87.5, carbs +0.0, energy balance −0.2.”
- Tomorrow morning: your weigh-in, the error, and whether it landed inside the range.
- Over time: your average miss, how often you land inside the range, and the four numbers the model has learned about you.
How the prediction is built
| Part | What it uses | Starting value |
|---|---|---|
| Trend | Fasted weigh-ins over the last 10 days, exponentially weighted (each new reading counts 25%). Flights and flagged outliers excluded. | – |
| Carbs | Yesterday’s carbs vs your 7-day average (glycogen is stored with water) | 0.3 kg per 100 g, capped at ±0.5 kg |
| Salt | Yesterday’s sodium above your 7-day average; skipped if any food that day has no sodium value | 0.1 kg per 1,000 mg, capped at +0.3 kg |
| Energy balance | Calories eaten minus your adaptive, training-adjusted TDEE | 1 kg per 7,700 kcal, capped at ±0.5 kg |
| Bowel movement | A yes/no flag on each weigh-in | 0.3 kg |
| Range | Your average miss over the last 14 scored mornings | at least ±0.3 kg |
It learns your body, not the textbook
The starting values are textbook averages. After every scored morning the model refits all four for you, using ridge regression pulled back towards those defaults, so a single odd morning cannot flip a coefficient. Some athletes hold a lot of water after carbs; others barely show it. In our founder’s first week, the carb effect fell to almost zero while the calorie balance of the day before counted about three times more than the 7,700 kcal rule suggests.
Rules that keep the data honest
- Conditions travel with the number. Each weigh-in carries a fasted flag and a bowel-movement flag. Readings that are not fasted are stored but never scored.
- The raw error stays on the record. When a known factor explains a miss, you see the error with and without it. The model never quietly rewrites its own score.
- Missing data is not zero. If a logged food has no sodium value, the salt factor is skipped for that day instead of reading as zero.
Why it matters for athletes
Endurance athletes swing harder than dieters. A long ride, a 300 g carb day before a key session, salt tabs and a recovery shake can put 1 kg on the morning scale with no fat involved. Cut calories because of that number and you under-fuel the next session. A prediction that already expected the jump stops that reaction. For the method behind it, see the hybridathlete.guide strategy guides on water weight for athletes and telling a plateau from water.
Comparing apps? Read MacroFactor vs Happy Scale for how the two best-known trend apps handle the same problem.
FAQ
How accurate is the weight prediction?
It shows its own accuracy. The range around each prediction is the average miss over your last 14 scored mornings, never narrower than ±0.3 kg. Under consistent conditions most misses come from things the evening before cannot know, like whether you go to the toilet before weighing, and the log labels those.
Is this a moving average?
The trend is an exponentially weighted average of your fasted weigh-ins over 10 days, where each new reading counts 25%. That reacts faster than a simple 7-day average and ignores single spikes. The prediction then adds what you ate and burned the day before on top of the trend.
Do I need a smart scale?
No. Any scale works if you use the same one every time. Weights come in from Apple Health or a connected smart scale, or you enter them by hand with the fasted and bowel-movement flags.
Does it tell me when I will reach my goal weight?
The trend line shows your direction, and your planned calorie balance shows the rate. The prediction itself is built around the next morning, because that is the number that makes people change their plan for the wrong reasons.