Contrast

How Contrast rates its own evidence, in four words

Every metric in Contrast carries one of four labels for how strong the evidence behind it is: Established (4 metrics), Borrowed (5), Theoretical (2) and Your data (3). Each label says what kind of claim the number can support, and five of the fourteen are saying the app invented the measure or is only describing your own log.

Fourteen metrics, forty citations, and a label on every one saying how much to trust it. The labels are the most useful thing in the Analysis tab and the easiest to skip past.

The four

Established: measured the same way in published research, with results you can compare against. Four metrics: sauna sessions, hot baths, cold shock, resting heart rate.

Borrowed: based on research into a closely related measure, applied here to sauna and cold. Five metrics: the weekly goal, weekly cold and heat, heart work in the heat, cold settle, heat strain.

Theoretical: Contrast’s own measure. It rests on known physiology but has not been tested in a study, so read it as a personal trend. Two metrics: heat adaptation, and heart rate by water temperature.

Your data: a description of your own sessions and Apple Health data. It shows what goes together for you, not what causes what. Three metrics: sessions and your nights, when you do it, over the months.

Five of fourteen are in the bottom two tiers. That is a third of the tab declining to claim scientific backing it does not have.

What each tier licenses

The label is not a confidence score. It tells you what kind of sentence the number can appear in.

TierYou can sayYou cannot say
Established“Four to seven sessions a week was associated with a hazard ratio of 0.37 in that cohort”“This will lower your risk by 63%”
Borrowed“ACSM defines moderate as 64–76% of max heart rate; Contrast applies that to heat”“You did 20 minutes of moderate exercise”
Theoretical“My heat adaptation estimate says adapted”“I am heat adapted”
Your data“My deep sleep is higher after session days”“Sauna improves my deep sleep”

The right-hand column is where most wellness software lives, and the labels are what keep the app out of it.

Why “borrowed” is the interesting category

Established and Your data are easy to understand. Theoretical is honest and obvious. Borrowed is the one doing the subtle work, and it is the largest group.

Take heart work in the heat. The ACSM intensity bands, moderate at 64% and vigorous at 77% of maximum heart rate, are thoroughly established as exercise thresholds. Applying them to sitting in a sauna is the borrowing, and it is not a small step: heat raises heart rate by moving blood to the skin rather than by working muscle, so the same percentage means a different thing.

Or cold settle. The finding that the early response to cold water gets shorter with repetition is from the lab. Nobody has validated the particular measurement Contrast makes, which is the seconds from the peak to the halfway point.

In both cases the ruler is real and it is being held against something it was not made for. The label says so, and the metric’s limits paragraph says exactly what was carried across.

Why not just hide the weak ones

Two reasons, and the first is the honest one.

An unvalidated measure of your own trend is still informative. Your cold settle has no population range and cannot be compared with anyone else’s, and it can still tell you whether the first surge is passing sooner than it did in February. That is worth knowing. The label stops it being mistaken for something more.

The second reason is that the alternative is not “no weak metrics”. It is weak metrics without labels, which is what most of this category ships. A composite readiness score is built from exactly these kinds of borrowed and invented measures; compositing them into one number does not make them stronger, it makes the weakness invisible.

What the app does with a disagreement

Heat adaptation is the clearest case, because there are two metrics measuring the same thing at different tiers.

Heat adaptation (theoretical) is a model: it counts your recent heat days and predicts. Heat strain (borrowed) is a measurement: it reads your heart rate at the same heat.

When they disagree, the metric page gives an instruction rather than an average: if it says adapted but your heat strain is not easing, trust your heart rate.

Tiering the metrics is what makes that instruction possible. Without the labels there would be two numbers and no principled reason to prefer either.

The three with no citations at all

Sessions and your nights, when you do it, and over the months carry the Your data label and an empty source list. There is nothing to cite because nothing is being compared against. They describe your own log and your own Health data.

That is the strongest and the weakest kind of evidence at once. Strongest because it is about you specifically rather than about a cohort of Finnish men. Weakest because a description of one person with no control condition cannot establish cause, which is why all three carry the same sentence: it shows what goes together for you, not what causes what.

Common questions

What are the four evidence tiers in Contrast?

Established: measured the same way in published research. Borrowed: based on research into a closely related measure. Theoretical: Contrast's own measure, resting on known physiology but untested. Your data: a description of your own sessions and Apple Health data.

Which metrics are unvalidated?

Two carry the Theoretical tier: heat adaptation and heart rate by water temperature. Five carry Borrowed, which means the research behind them studied something adjacent.

Why show a metric that is not validated?

Because it is still informative about your own trend, and because the alternative is either not showing it or showing it without the label. The label is what lets you weight it correctly.

How many citations does the app carry?

Forty, across fourteen metrics, each with the population and the finding written out. Three metrics carry no citations, because they describe your own data rather than setting it against published research.

Sources