How these rides are measured

Where the numbers come from

Everything on a ride page is worked out from the GPS trace I recorded — position, altitude and time. There is no power meter and no heart-rate strap in the data, so anything resembling power is modelled rather than measured, and it is worth knowing how.

Distance, climbing and speed

Distance is the sum of the great-circle hops between fixes. Climbing uses hysteresis rather than a running total: a barometer wanders by a metre or two while you stand still, and adding up every small rise turns that wandering into hundreds of metres that were never climbed. Height is only banked once you are decisively above the last reference point.

Barometers also fail outright, and they fail by reporting a number rather than a gap. Day three of the Burundi tour records an altitude of exactly zero for two stretches in the middle of a ride at 1,700 m. Readings that would require an impossible rate of climb are discarded, as are any that sit implausibly far from the ride's own median height.

Top speed is the fastest five-second average, not the fastest pair of fixes. A single dropped fix between two good ones implies a speed no bicycle reaches, and taking that as a record would be flattering but false.

Stops, and the marks on the route

Three things are marked on the map. The fastest point is the middle of the quickest five seconds; the steepest is the sharpest gradient sustained over two hundred metres, so one abrupt pair of readings cannot claim it.

A stop is harder to spot than it sounds, because the recording does not stop when you do — the watch keeps logging while you stand at a junction. So a stop is found as movement of less than twenty-five metres over more than two minutes. Anything shorter is a junction rather than a stop, stops within a hundred metres of each other are counted as one, and only the longest eight are drawn: a map with forty dots on it tells you nothing.

Effort

Power is estimated from the physics of riding a bicycle, using the model in Martin, Milliken, Cobb, McFadden and Coggan (1998), Validation of a Mathematical Model for Road Cycling Power, Journal of Applied Biomechanics 14(3), 276–291. It adds up what you are working against — air resistance, rolling resistance, gravity, and the cost of accelerating your own mass — and multiplies by speed. Against a real power meter the model correlates at R² = 0.97.

Air thins with altitude, so air density is computed from the height the trace recorded rather than assumed. At 1,700 m the air is 15% thinner than at sea level, and ignoring that would overstate the cost of the hardest rides here.

That power is then turned into an effort score the way a coach would: normalised power by Coggan's method, which weights hard efforts more heavily because the body does; divided by a threshold power of about 206 W, estimated as 95% of my best twenty minutes anywhere in this collection; and scaled so that one hour at threshold scores 100.

Heart rate, where I was wearing a monitor

Modelled power is a good guess, but it is still a guess. Heart rate is a measurement, and on the rides where I was wearing a watch it replaces the estimate entirely. It is scored as Banister's TRIMP: each second is weighted by where the heart rate sat between resting and maximum, on an exponential curve, because ten minutes near the top is far more than twice the cost of ten minutes near the bottom.

My resting rate is 47 bpm, a twelve-month average from the watch, and my maximum 191 bpm, recorded on a climb in May 2026 and held for half a minute — long enough that it cannot be the optical sensor misreading.

The great advantage over the modelled figure is wind. My heart already paid for the headwind, at the strength it actually blew in that lane — which is finer than the regional average the weather line reports, and finer than anything the physics is given. Comparing the two across the rides that have both found a real fault in the model, which is the argument for keeping both.

Rides from before I wore a monitor keep the modelled figure, converted onto the same scale using the median ratio between the two measures across the rides that carry both. It is approximate — the typical error is about a fifth — but a five-band rating is coarse enough to absorb that, and the alternative was leaving thirteen years of riding unrated. Those figures are marked with an asterisk in the log.

The heart-rate chart on those rides is banded by training zone, set as fractions of heart-rate reserve rather than of maximum. Reserve is the better basis when a resting rate is unusual, and 47 bpm is unusual: measured against maximum alone, an easy hour would land in the wrong zone entirely.

The weather line

The italic line under the figures is not something I measured. It comes from ERA5, a reanalysis of past weather, read through Open-Meteo and averaged over the hours the ride took. The grid squares are about twenty-five kilometres across, so it describes the region and the afternoon rather than the lane and the moment — enough to say the wind was on the nose, nowhere near enough to say what it was doing round any particular corner.

Where a wind is called a headwind or a tailwind, that is worked out leg by leg along the route and weighted by distance, not taken once from the ride's overall direction. On a loop the two cancel, which is honest: the wind cost that ride nothing overall. That is why a circuit says the route met the wind from every side rather than picking one, and why the useful figure for an out-and-back is the share of the distance ridden into it.

It deliberately feeds nothing else. A twenty-five kilometre average cannot correct a power estimate for one rider on one road, and folding it in would trade an honest unknown for a confident wrong number.

The rating panel

Four measures, each scored one to five against the rest of these rides rather than against some absolute nobody shares. A five means "among the hardest, longest or fastest rides I have recorded", which is the only comparison that means much without a peer group.

Spice: Annie's word, from riding Burundi together: the point where the fun and the danger start to overlap. Three things make it — how fast the ride got, how steep it got in either direction, and how hard the corners were taken. Each is capped, so no single one can carry the score, and each is scored against the rest of these rides rather than against anything absolute.

A corner counts for more the faster it was taken and the tighter it was: the sideways pull through a bend goes up with the square of the speed, so the same corner ridden twice as fast is worth four times as much. That is deliberate. A lane full of bends ridden slowly is not spicy, and neither is a fast straight road.
Grind: Metres climbed per kilometre ridden. A short steep ride and a long flat one can share a total; this separates them.
Distance: How far. The plainest of the four.
Effort: The training-stress figure described above.

What these are not

Wind is the single largest source of error in the modelled figure. A 20 km/h headwind can double the power needed to hold a speed. The weather line now tells you roughly what the wind was doing, but that is a 25 km regional average and the model needs the wind in a particular lane at a particular minute, which no reanalysis can give — so the figures still cannot tell a hard day into a westerly from an easy day with it behind, and feeding the average in would trade an honest gap for a confident wrong number. Drag area and total mass are assumptions (87 kg of rider, bike, kit and bottles), and the effort figures scale with both. Drafting is invisible.

So these numbers are consistent with each other and useful for ranking one ride against another. They are not measurements, and I would not present them as such.