What Six Hours of Cross-Training Is Worth

A coach looking at an injured athlete gets two numbers that will not talk to each other: a mileage figure, which has collapsed, and a pile of cross-training time, which has not. Six hours of something. There is no way to weigh one against the other.

The problem is that six hours is not one thing. A kid who walks to school and back every day and a kid who swims three mornings a week both show up as “lots of cross-training.” One of them is training. The other is commuting.

We wanted a single number that could tell those two weeks apart. Here is what the data said about how to build it.

Walking Is Most of It

The first thing worth measuring was the shape of the roster’s cross-training, and it was not what we assumed.

Walking and hiking come to 825 hours across the roster over twelve months — more than cycling and swimming combined. Cross-training is not mostly pool sessions and stationary bikes. It is mostly people moving around.

That single fact rules out the obvious approach. Any flat conversion — every minute of cross-training is worth the same fraction of a mile — is dominated by walking, because walking is where the hours are. Whatever else such a number measures, it mostly measures how much an athlete walks.

The Inversion

The clearest way to show a rate is doing real work is to find two athletes it has to rank correctly, and check.

One athlete had 2.1 times the cross-training hours of another. Their hours were 86% walking and hiking. The comparison athlete’s were 92% swimming, cycling and elliptical.

Under a flat rate, the walker scored nearly double. That is the wrong answer, and obviously so: one athlete was maintaining fitness through an injury and the other was walking a lot.

Under per-modality rates — swimming at 8 minutes per point, machine cardio at 10, cycling at 15, hiking at 25, walking at 30 — the walker scores below the swimmer, on 2.1 times the hours.

The part that mattered more to us was the other athlete’s number. It barely moved. Genuine cross-training scores about what it scored before; only the walking came down. A model that deflated everything uniformly would have been useless — it would just be the same ranking in smaller numbers. This one changes the order.

The Trap We Nearly Walked Into

Watches label activities inconsistently, and one of those inconsistencies almost got encoded into the rates permanently.

Two labels kept appearing in the data — generic and fitness_equipment — and they looked like different things. They are not. They are the same activity on different hardware:

They are near mirror images of each other. One brand calls an indoor session one thing, the other calls it something else. Rating those two labels differently — which we very nearly did, since one sounds more specific than the other — would have scored the identical elliptical session differently depending on which watch the athlete happened to own. About half our roster is on COROS.

They are rated the same. So is the third generic bucket, training, for the same reason.

Why Not Heart Rate

This is the first question anyone asks, and it has two separate answers.

As an intensity multiplier, the data will not support it. Roughly 1,041 activities carry a sentinel heart rate of 255 bpm — the value a device writes when the strap drops out. Any average built on top of those is wrong, and silently so.

As a classifier, it does not separate. This is the more interesting failure. We checked whether heart rate could distinguish real machine work from ambient activity, scaled against each athlete’s own running heart rate to control for fitness. Real machine sessions sit at 0.58–0.75 of an athlete’s running heart rate. Non-training sits at 0.50–0.53. There is no usable margin between those, and the fitter the athlete the narrower it gets — a fit athlete’s easy effort and a fit athlete’s real effort are closer together, not further apart.

Average heart rate cannot see the difference between a 45-minute lift averaging 120 bpm and a 45-minute elliptical averaging 120 bpm. It survives in the model in exactly one narrow place: an activity over two hours whose average heart rate is under 95 bpm is a watch someone forgot to turn off.

Elevation, Which Actually Works

Elevation turned out to be the one auxiliary signal clean enough to use, and for a reason worth stating: bad elevation data is identifiable, not merely extreme.

Genuine mountain days in our data measure 6–14% grade. Barometric drift produces figures like 74% and 123% — implied grades that are not physically possible over the distance recorded. There is a wide, empty gap between those two populations, so a single threshold separates them cleanly.

So walking, hiking and cycling get a climb term: about 1,000 ft of climb is worth one extra point, capped so climbing can never out-earn the time spent, and thrown away entirely when the grade is not credible. Across twelve months it adds 15% to hiking, 10% to cycling and 5% to walking. It refines the number; it does not distort it.

What We Deliberately Did Not Do

We did not exclude the ambiguous buckets. The generic indoor labels could be strength work, which is not aerobic volume. They could also be elliptical, which is. In this roster they are predominantly elliptical, and excluding them by default would zero out exactly the injured athletes the number exists to serve. Ambiguity resolves in the athlete’s favour. Where a watch explicitly reports strength training, that is excluded — but only when it actually says so.

We did not calibrate per athlete. The number is meant to be read within one athlete over time. Per-athlete rates would move under an athlete as their fitness changed, which is the one thing a trend line must not do.

We did not make runs more complicated. A run contributes its mileage, unchanged — no climb bonus, no time conversion. That keeps one property that turns out to be worth more than internal consistency: a pure runner’s volume equals their mileage, so any gap between the two numbers is exactly the cross-training. A coach can read that off the page without knowing anything else on this list.

What It Is Not

It is worth being blunt about the limits, because a weights table invites more confidence than the number deserves.

It is not a training-load model. There is no intensity, no fatigue, no TSS. It is not calibrated per athlete, and it is not for ranking athletes against each other — two athletes with the same volume did not do the same training. It is not precise enough to prescribe from.

It is for noticing, within one athlete over time, whether the week was real work. That is a lower bar than it sounds like, and it is the thing a mileage column genuinely could not do.

The full rate table and the exact arithmetic are in the Aerobic Volume documentation, including a worked week you can recompute by hand.