Running · Every run

How these runs are measured

Where the numbers come from

Everything on a run page comes out of the GPS trace and, where the watch recorded one, a heart rate. There is nothing else in the data: no power, no cadence, no lactate. So the figures below are either measured directly, or worked out from my own running rather than from a table of somebody else's.

Critical speed, and why almost everything hangs off it

The one number the rest depends on is critical speed: the quickest pace I can hold without drifting steadily towards exhaustion. It is not a guess. Take the furthest I have covered in any two minutes anywhere in this collection, then any three minutes, five, eight, ten, fifteen and twenty, and those distances fall on a straight line against time. The slope of that line is critical speed and the intercept is the small reserve available above it.

For me that comes out at 12.9 km/h, about 4:39 a kilometre, with a reserve of 138 m. The line fits at R² = 0.9989, which is tight enough to trust and would be suspicious if it were much tighter.

The four speed bands

The colours on a route, on the bar beneath it and behind the speed chart are all the same four bands, and each boundary is a measured thing rather than a round number I liked.

Walking — slower than 7.4 km/h

The walk-to-run transition. People switch from walking to running at about 2.06 m/s regardless of how trained they are — 7.23 km/h in non-runners and 7.42 in trained runners, which is close enough to call it 7.4. The better way to set it would be by leg length, since the transition happens at a fixed Froude number across body sizes, but my leg length is not in a GPS trace and inventing one to three significant figures would dress a guess up as a measurement. It does move with the ground: the transition falls about 1% for every 1% of gradient, so the threshold on a climb is lower than on the flat.

Jogging — up to 10.6 km/h (5:39)

Below the aerobic threshold: conversational, and sustainable more or less indefinitely. The threshold sits at 82.3% of critical speed, pooled across 26 studies and 527 runners, with a confidence interval of 81.1 to 83.6 — tight enough to use directly.

Running — up to 16 km/h (3:45)

Above the aerobic threshold. Hard work, and it includes everything from a steady effort up through my five-kilometre race pace.

Sprinting — faster than 3:45

The one chosen number here rather than a derived one, and chosen twice. It sat at critical speed to begin with — the fastest pace I can hold, 12.9 km/h — which is my five-kilometre race pace, so the band was calling a hard sustained effort a sprint and nearly every run collected a mark for it. Sixteen is what sprinting actually looks like in my own data: across sixty runs I spend 0.19% of the time above it and my median run tops out at 14.2. Five of those sixty runs reach it at all. A tier most runs never enter is the point rather than the flaw — it is the one that means something when it appears.

Speed is a nine-second rolling average rather than an instant reading, because consumer GPS is not trustworthy under acceleration, and a band edge is a statistical line rather than something a body crosses in one stride. Stretches shorter than twenty seconds are merged into their neighbours before the route is drawn, so the map shows a few broad stretches rather than confetti — but the percentages in the key come from the unsmoothed data, because a twelve-second wait at a crossing is real even when it is not worth a colour change.

What the coaching is comparing against

Every run on a route I have run at least four times is cut into twenty equal slices of distance, and each slice is compared with the same slice on every other run of that route. That is the whole method, and the reason for it is that it needs no model: the same slice is the same hill, so the gradient cancels without anybody having to correct for it.

The obvious alternative is grade-adjusted pace, and I have avoided it deliberately. It has no validation literature — searching the biomedical databases returns a handful of records, none of them comparing it against measured energy expenditure. The cost-of-running curve underneath it is real, but it was measured on mountain runners, and individual responses to slope do not generalise: being economical on the flat predicts nothing about what a steep climb costs you.

A slice is only mentioned when it falls outside the ordinary spread of my own runs on it, so a route run forty times says a great deal more than one run four times. Where the notes say there is time available, that is the difference between this run and my own best on that same stretch — not a target invented for me.

What the notes say also depends on what kind of run it was, which is decided by how much of it was spent above critical speed. Fading over the last third of a long run is the shape of the event rather than a fault, and the tool is not allowed to tell me to push harder there.

How much of it was off-road

Each GPS point is snapped to the nearest way in OpenStreetMap, and the run is scored on how much of it was on made ground rather than laid ground. Where a way records its surface that is used; where it does not — and it is missing on more than half of them — the road class stands in, which predicts well enough in this part of the world.

There is no ground truth to check it against, so the test is repeatability: the same route scores within about three percentage points from one morning to the next, against a spread of seventeen between different routes. That is a real signal, but it is precision rather than accuracy. It rests entirely on OpenStreetMap being right about ground nobody has verified.

Two things it cannot do. A pavement beside a road reads as paved, which is true about the surface and useless for telling a town centre from a carriageway. And a hard gravel byway and a muddy field trail are both simply off-road.

Why the maps stop short

Almost every run here starts and finishes at my front door, and that is a problem the individual maps do not have but the collection does: one route is a line, and sixty lines radiating from the same point are an address, fixed more precisely than any single one of them could manage. So the drawn geometry is clipped inside a disc a few hundred metres across, and the routes simply begin and end at its edge.

Three details make that worth doing rather than merely reassuring. The disc is not centred on the house — if it were, the hole in the middle of sixty converging routes would be a better pointer than the routes ever were. It never moves, because a zone that varies run by run can be intersected across repeats and recovered. And it removes points rather than nudging them, since a moved coordinate is both guessable and a small lie about where somebody ran.

It costs about a tenth of each drawn route and nothing else. Every figure on the page — distance, climbing, pace, heart rate, the bands, the coaching — is calculated from the whole recording, which never leaves my machine. Only the picture is trimmed.

When something was not recorded

A missing measurement is not a measurement of zero, and the figures here try hard not to confuse the two. Two of the runs in this collection reached Strava with no heart rate at all — the watch recorded the run and the heart rate never left it — and an earlier version of the spice score read that as "this run took no effort", which made a five-kilometre personal best the mildest thing on the shelf. Ingredients with no data now drop out of the blend rather than scoring nought, so a run is judged on what was actually measured.

Where a run says no heart rate recorded, that is what happened: there is no heart-rate chart, and the effort side of its spice comes from pace alone.

What these are not

The bands are honest about pace and silent about effort. Heart rate lags a change of pace by half a minute or so, so nothing here reads intensity from a short burst. On a descent the two disagree sharply and in a way worth knowing: the heart rate falls while the impact through the legs rises by about half again at a moderate gradient, so a downhill stretch that looks easy on every chart on this page is the part of a run doing the most damage.

Nothing here is a calorie count, a fitness score, or a training load I would defend. The figures are consistent with each other and useful for ranking one of my runs against another. They are not measurements of me, and I would not present them as such.