See your weight over time

Six weeks into the cut. The scale moves day-to-day in ways that don't tell you anything, but the trend underneath the noise does — and that's what you ask the agent to draw.

What the agent needs to hear

Behind your sentence, the agent works backwards to two pieces: the time window and what to surface. The window defaults to the last three months — long enough to filter daily noise, short enough to reflect what you’re doing right now. What gets surfaced defaults to the smoothed curve plus the direction it’s moving and the rate of change in kg per week.

The window is the only piece worth being explicit about when you want something non-default. « Last month », « since the cut started », « last 6 months » all parse cleanly. Everything else — the smoothing, the direction read, the rate-of-change number — the agent draws by default. You don’t ask for the trend; you ask for the window, and the trend comes with it.

Windows, smoothing, and what the curve filters

The default window is three months. Short enough to show the cycle you’re in, long enough that the underlying signal cuts through the day-to-day bounce. Daily weights swing 1–2 kg from water, sodium, and timing — the curve you see is a moving average, not the raw spikes. That smoothing is automatic; you don’t ask for it.

Override windows when the default doesn’t fit. « Last 6 weeks » zooms in on a recent block. « Since March 1 » bounds it by date. « All time » zooms out to the longitudinal view. If you want to inspect specific days — to find the outlier or the gap — ask for the raw daily readings: « show daily, not smoothed ». The curve and the points are two different reads, and you choose which one answers the question you’re holding.

What the curve actually tells you

The trend reads three things. Direction — climbing, flat, or descending — answers the « is this working » question at a glance. Rate of change in kg per week is the practical number: at 0.5 kg/week down you’re in a sustainable cut, at 1.5 kg/week down you’re probably losing muscle. Inflection points show where the trajectory bent, and they usually correlate with something real — a diet change, a stalled week, a return from holiday.

The curve only matters because the log quality is good. Sparse weeks, mixed-condition readings, or a clothed weigh-in dropped into a fasted-morning series — any of those corrupt the smoothing no matter how clean the math is.

When the agent gets it wrong

If the window doesn’t match what you asked for — « only the last 6 weeks, not 3 months » — name it and the curve re-renders. If the curve looks flat but you know you’ve lost weight, the log series probably has gaps: ask « show me which days are missing in the last month » and the picture gets honest.

If the rate-of-change number reads wrong because one outlier is pulling it — a post-meal evening reading, a sick-day spike — ask for the trend excluding that day. The agent recalculates without it.

What makes the trend worth trusting

Three things decide whether this weight curve tells you something useful: the log series is dense and consistent (sparse weeks read as noise, mixed-condition logs corrupt the curve), the window matches the question you’re asking (3 months for cut progress, 6 weeks for a tighter recent read, all-time for longitudinal context only), and outliers are recognized as outliers (a post-meal reading or a sick-day spike shouldn’t drive the rate-of-change number). A trustworthy trend is what tells you whether the next plan adjustment is « hold the line » or « pull the deficit ».