Every section of the answer
Nothing here is generated prose. Each section is an engine’s output, and the sections that say “nothing” are as load-bearing as the ones that don’t.
| Section | What it answers |
|---|---|
| What changed | Whether it stepped on a dateable day or has been sliding, and by how much. A slide is never given a false onset date. |
| Drawn evidence | The readings the verdict was reached on, with the onset marked when one honestly exists. |
| Moved alongside | What else changed in the same period, with timing distance and change size in its own units. |
| You recorded | Injuries, illness, travel you logged yourself — the only thing Brin will put beside a change, because you said it happened — never as a cause. |
| Checked, nothing found | Named explicitly. Ruling things out is a result. |
| Couldn’t check | Measures with too few readings, listed rather than silently dropped. |
| Chance note | How many measures were swept, and what that means for coincidence. |
Falling across this period.
A trend describes the overall direction; its first date is not necessarily when the change began.
Windows you choose
90, 120, 180 or 365 days. A change that is real at one window and absent at another is telling you something, and Brin would rather you saw that than picked a window for you.
Personal range
The band is the 2.5th to 97.5th percentile of your trailing 60 days. It moves as you do, which is why a slow drift can stay inside it for months.
Onset, only when real
A step gets a date. A trend does not, and the chart refuses to draw one, because a dashed vertical line is a claim.
Change in real units
−38 min of sleep, +5.0 bpm, +0.6 br/min. Never a normalised score.
Your own words
Event titles stay private. Only the category ever reaches a model, and only if you use the assistant.
Why this is the whole product
The example on this page is a slide: heart rate variability drifting down with no single day it began, so nothing here is dated and the measures that moved alongside it are reported as a separate trend rather than a coincidence in time. The home page shows the other kind — a step, on a day Brin can name, where a recorded event can sit beside it.
Most health apps answer “how am I doing?” with a number. Brin answers what moved alongside — or says plainly that it cannot.