The science behind how OnWeight turns a noisy daily scale reading into a trustworthy, early signal — and never an alarm. Prepared for sports-dietitian review and sign-off.
Every weight-cut app faces the same tension. Daily weight swings so much that reacting to a single reading is meaningless and stressful. But a combat athlete on a tight weigh-in deadline can't afford to find out a week too late that they're off track — lost runway can't be recovered.
The instinct is to pick a side: trust the noisy daily number (fast but wrong), or smooth over a week (reliable but slow). We reject that binary. This brief shows how we get an early signal and a reliable one — by separating the two jobs into different layers, each tuned to what the science actually supports.
"A 7-day average is a long time for an athlete on a tight deadline. By day 7, isn't it too late?" — the question this methodology is built to answer.
This number sets every threshold in the system, so we anchored it to published data rather than intuition.
The consequence is purely statistical. Averaging reduces random noise only by the square root of the number of days — so a short window barely dents the noise, while the real signal we're chasing (a safe loss of ~0.5–1% of bodyweight per week) is small. For a ~78 kg fighter:
| Window | Residual noise (1 SD) | True loss in window @0.5–1%/wk | Verdict |
|---|---|---|---|
| 1 day | ~0.4–0.8 kg | 0.06–0.11 kg | Noise ≫ signal |
| 3-day average | ~0.23–0.46 kg | 0.17–0.34 kg | Noise ≈ signal |
| 7-day rate | ~0.15–0.30 kg | 0.39–0.78 kg | Signal clears noise |
This is exactly why a "within 200 g of plan" pass/fail on a 3-day average would misclassify often — its own noise is as large as the thing it's measuring. The reliable verdict only emerges at the 7-day rate. That finding shapes the whole design below.
The weight the athlete sees as their "verdict" is a smoothed trend, computed with an exponentially weighted moving average (EWMA) — the same approach popularised for body-weight tracking in The Hacker's Diet, where the trend line is used instead of focusing on daily fluctuations.3 Crucially, an EWMA weights recent days most heavily, so it bends toward a genuine change within a couple of days rather than lagging a full week.3,4
A fighter's true-weight target shouldn't be a straight diagonal, because hormonal water retention predictably lifts the scale late in the cycle. The best controlled evidence is unambiguous about both the effect and its size: a 2023 study tracking 42 women across the cycle found bodyweight was higher during menstruation by 0.45 kg, attributable almost entirely to a 0.474 kg rise in extracellular water — with no other body-composition change.5,6 Population summaries put the typical gain at roughly 0.5 kg, ranging from none up to ~2.3 kg in some individuals.7
So we default the water amplitude to a modest 0.5 kg and individualise it from the athlete's own logged cycles — never assuming a large swing for everyone. Every tracking comparison runs against this undulating line, not a raw one.
We never judge a cut on one reading. Each morning's weigh-in steps through a cascade — and every step compares against the cycle-adjusted line. Steps 1–2 are silent; only the rate produces a verdict.
Two facts dissolve the deadline worry. First, the 7-day window is rolling — it recomputes every single day, not at a weekly checkpoint, so nobody waits for "day 7 to arrive." Second, we add a faster early-warning layer on top: a drift-watch.
The drift-watch is a CUSUM — a cumulative-sum control method designed for sequential change detection.8 Instead of looking at each day alone, it adds up small same-direction deviations from the cycle-adjusted plan. A built-in slack term absorbs ordinary water noise, but a consistent drift accumulates and becomes visible far sooner than a windowed average would admit it.8,9 This is its whole reason for existing: cumulative methods make small, persistent shifts visible much faster than traditional charts.9
The only weight shown. Recent days weighted heaviest, so it turns quickly without chasing noise.
CUSUM of deviation vs plan. A quiet, non-red note. Soft only — it never launches a plan change.
Rolling rate + daily projection. This is the verdict, and the only thing that offers a plan change.
A fighter's costs are asymmetric: a missed "you're behind" is far more damaging than one extra soft nudge, because late runway is irreplaceable. So the drift-watch's decision limit shrinks as the deadline approaches (more sensitive at ≤4 weeks, more sensitive again at ≤2 weeks) — buying faster detection precisely when it matters most, at the price of a few more (still non-alarming) watches.
The soft drift-watch never changes anything. A Change-your-plan flow is offered only on a confirmed, hysteresis-protected trigger — never one noisy window:
| Trigger | Condition | Why it's separate |
|---|---|---|
| A · Projection miss hard | Projected scale weight (trend + weigh-in-day cycle water) misses the limit, confirmed across 2 reads. | Protects against arriving over the limit on the day. |
| B · Safety ceiling hard | Measured rate exceeds ~1%/wk of bodyweight. | Fires independently of making weight — protecting performance and health is its own goal. |
Both directions are handled with the same care but opposite advice: drifting behind → tighten intake gently; cutting too fast → add food back. The flow itself offers only safe routes — move up a division, move the date, or hold at the capped safe rate — and nothing changes until the athlete chooses.
Every statistic above is deliberately invisible. What reaches the athlete is three plain things — a calm number, a one-line verdict, and a living projection — delivered in the corner-coach voice, never as a dashboard of metrics.
| Behind the scenes | What the athlete sees | How it's said |
|---|---|---|
| EWMA trend §03 | The single weight on the dashboard. | "Your morning trend" — calm, singular, never the raw scale. |
| Cycle-adjusted projection §04 · §07 | "You'll come in at 70.2 kg" — to the day. | A living forecast, refreshed every morning; the wet number, not the dry one. |
| 7-day rate verdict §05 | A tone-coloured pace line: on track / behind / too fast. | One sentence, in order: fact → why → action. |
| Drift-watch §06 | A soft "I'm watching this" note. | Amber, never red; flags a trend, asks for nothing. |
| Confirmed trigger §07 | A Change-your-plan offer. | Only safe routes — move up, move the date, hold safe; nothing moves until they choose. |
One number to trust, one sentence to act on. Everything else — the filters, the CUSUM, the regression — stays under the hood, because an athlete cutting weight needs a corner, not a control chart.
Detection theory is clear that there is an unavoidable balance between how fast you detect a change and how often you raise a false alarm.10 You can't escape that curve — you can only choose where to sit on it. Lower the threshold and you detect faster but cry wolf more; raise it and you're calmer but slower.
Our context makes the choice for us. Because a fighter's downside is asymmetric, we deliberately sit toward early and sensitive — but we put that sensitivity on a soft, non-alarming channel (the drift-watch), so being early never means being stressful. The high-confidence, low-false-alarm verdict stays reserved for the one thing that actually changes the plan. Two further protections matter more than raw speed:
The fighter sees a calm trend and, when needed, a quiet "I'm watching this" days early — while the actual plan only ever changes on a confirmed, evidence-grade signal.
Clinical figures (daily variance, cycle water) and the detection methods (EWMA, CUSUM, the delay/false-alarm trade-off) are cited below for independent review.
Note on sources: where a primary study was available (cycle water retention), it is cited directly; consumer-health pages are used only for figures they attribute to named clinical bodies (e.g. Cleveland Clinic). This brief describes OnWeight's weight-tracking logic only and is not medical advice.