Smarter Autopilot: why we check two signals before moving your calories
Autopilot used to read your weight trend alone. It now checks that trend against your real first-to-last weight change before it suggests anything. Fewer false alarms, more patience.
The old rule, and where it broke
When Autopilot first shipped it called every result from one signal, the trend slope across your comparison window. Take the last 5 days, fit a line, read where it points. Flat means you held. Down means you lost. Up means you gained.
Clean idea. It worked most of the time. It also had a way of getting things wrong.
Weight is noisy. Water, salt, sleep, the bathroom you visited or didn't, the day of your cycle. All of it moves the scale by 0.5 to 1.5 kg with no bearing on your actual fat mass. When that noise lined up with the slope just right, Autopilot saw a trend that wasn't there and recommended a calorie change you didn't need.
The worst version went like this. You are in a Cut, two weeks of good adherence, and one bad night of sleep spikes a single weigh-in. The trend turns down. Autopilot suggests another -200 kcal. You are already in a deficit, and now it is deeper for no reason.
That costs you trust in the app, so we fixed it.
The new rule: two signals, and both have to agree
Autopilot now reads two things before it suggests anything.
1. The trend slope. Same as before. Fit a line across the comparison window and read where it points.
2. The endpoint delta. Your actual weight today against your actual weight on day 0. One number against another, nothing fitted.
Autopilot acts only when both point the same way. Trend says down but your first-to-last weight is flat, nothing happens. The endpoint dropped but the trend is a mess, nothing happens. Both, or neither.
When they do agree, the call holds up. The trend gives you the direction and the endpoint proves the move was real, rather than an artefact of where the line happened to land.
This is patience, not hesitation
It would be easy to read this as the app doing less and therefore being wrong less often. That is not what changed.
When both signals agree clearly, a clean cut, a steady gain, a flat plateau, Autopilot moves exactly as fast as it used to. The full 200 kcal step, on the same comparison window.
What changed is the edge. The cases where the slope was the only thing arguing for a change, and one weigh-in was carrying that slope on its own. Those used to trigger an adjustment. They no longer do.
In testing, noisy weeks stopped producing recommendations. Travel, cycle, illness. Clean weeks produced them at the same pace as before.
The mental model
The old rule: the trend is what it is.
The new rule: the trend has to mean something.
A line through 5 noisy points can point anywhere. A line that points down and ends below where it started is a signal. That is what Autopilot waits for now before it touches your calories.
You will see fewer Briefs that arrive too early. The ones you do get, you can act on.
Haven't tried Autopilot yet? Peak Maintenance is free. Read the full Autopilot guide