🔬Reading a session in detail

The three kinds of number, and why it changes everything

• •Measured: a sensor recorded it. Heart rate from a chest strap, power from a crank power meter, altitude from a barometer. This is as reliable as it gets, sensor quality aside.
• •Computed: derived from the GPS track by a formula with no hidden assumption. Distance, elevation gain, average speed, the second-half drop. Reliable if the track is — hence the importance of the “GPS quality” block.
• •Estimated: a model with constants chosen for you. Pulling power, calories, effort score. Useful to compare yourself with yourself, never to compare yourself with someone else. These blocks carry an “estimate” badge.

What each line measures

• •The Good / Fair / Poor badge combines three things: accuracy at the 95th percentile (20 m and 35 m are the two thresholds), the largest gap between two points (30 s and 90 s), and the share of the session on a weak signal (5 % and 25 %).
• •Average accuracy ±X m · p95 ±Y m: the second number is the one that counts. It says that one point in twenty is worse than that. Under 8 m on average, you had open sky; beyond 35 m at p95, you were in dense woods or an urban canyon.
• •Worst gap: the longest silence. Beyond 90 s, this is no longer the signal slowing down, it is the operating system suspending the app. Turn off battery saver.
• •Weak signal: the cumulative time spent with accuracy beyond 25 m, rounded to the minute. So “0 min” can hide up to 29 seconds.

Four reading traps, all real

• •The vertical axis is not anchored at zero: it fits the session's values with 8 % of margin. A 3 bpm variation fills the whole frame and looks dramatic. Always read the numbers, not the shape.
• •The average shown on the chart is a PER-SAMPLE average, stops included. The one in the top tiles is computed over active time. On speed, the chart will therefore always sit lower than the tile — both are right, they answer different questions.
• •The trace is decimated to 250 points to stay smooth, but min and max are computed over the complete series. A peak announced in the “max” number can be invisible in the drawing.
• •On a Strava import, speed is already smoothed at the source: real peaks are shaved off and the chart is flatter than reality.

How the traction number is built, step by step

1. For each small interval of the track, the app adds up three forces: rolling (proportional to mass), gradient (mass × gravity × gradient), and air (proportional to the square of speed).
2. If the sum is negative — downhill, when gravity is enough — it is brought back to zero. The model then records 0 watts.
3. Power = force × speed, on each interval.
4. The result is averaged weighted by TIME, over only the intervals where you were really moving.
5. “Per dog” is that average divided by the number of tagged dogs. Nothing more.

The assumptions, written out plainly (you cannot change any of them)

• •System mass, rig + human + dog: 90 kg in bikejoring, 85 on a scooter, 110 in sledding, 75 on skis. Whether you weigh 55 kg or 110 kg, you read the same number.
• •Rolling resistance: 0.015 in bikejoring, 0.018 on a scooter, 0.050 in sledding and skijoring. The bikejoring value matches a tire rolling on hard ground; on soft trail or sand, the real value is three to four times higher, so power is largely underestimated.
• •Air penetration: 0.45 m² in bikejoring, 0.50 on scooter and sled, 0.40 on skis.
• •In canicross there is NO number at all — and that is deliberate: a rolling and drag model makes no sense for a running human.

What this number does not say

• •“System power” is NOT what your dog pulls. It is the power needed to move the whole thing forward, whatever its source. In bikejoring, everything you produce by pedaling is counted in there.
• •“Per dog” is a division, not a measurement. No split by size, by position in the team or by actual engagement. A passenger dog in a team of six “produces” as much as the lead dog.
• •There is no wind. Drag is computed on your speed relative to the GROUND, not relative to the air. Twenty kilometers per hour of headwind nearly doubles the real drag without changing the displayed number at all.
• •There is no acceleration. Surges — the start, a sprint, the exit of a turn — are exactly the moments when a dog works hardest, and the model does not see them.
• •Descents are worth zero watts. A very downhill course therefore drags the average down even though the dogs ran the whole way.
• •Changing the session's discipline changes the watts RETROACTIVELY, since the discipline picks the mass and the coefficients.
• •These watts feed no load or recovery calculation in the app. Do not use them as a basis for dosing training.

Expected orders of magnitude, on the flat, at cruising speed

• •Bikejoring at 22 km/h: about 140 W and 23 N. At 25 km/h: about 180 W.
• •Scootering at 18 km/h: about 110 W and 22 N.
• •Sledding at 18 km/h: about 310 W and 62 N — the rolling resistance of snow is what dominates.
• •Skijoring at 16 km/h: about 185 W and 42 N.
• •A 5 % climb in bikejoring at 22 km/h takes it to about 410 W: gradient weighs far more than speed.
• •Beyond 700 W in bikejoring, or an average force above 200 N, suspect noisy altitude rather than a feat.

The three rules, and their weak point

• •Speed down: triggered below 93 % of usual, alert below 88 %. The reference is an UNweighted average of the sessions: four 800 m warm-ups drag the usual down and can mask a real drop. It also mixes every course profile — a mountain run triggers the alert with no fatigue whatsoever.
• •HR high for this pace: triggered above 5 % over usual, alert above 8 %, and only if the pace is comparable. This is YOUR heart rate, not the dog's — despite sitting right next to the previous line, which is about the dog.
• •Heavy second-half drop: from 10 % up, alert at 15 %. No history required. On an interval session or a descent-then-climb course, it fires every time.

What each item covers

• •The intensity badge (light, moderate, sustained, hard, very intense) reads off the internal “load”: under 25, under 50, under 80, under 120, then beyond.
• •“Load” is a unitless score, on a scale of roughly 0 to 250. Its graduation is not displayed: remember that the tip into “hard” is at 80.
• •“Avg pace” is displayed in km/h, despite its name. And in comparisons, “+18 % pace” means “faster” — the opposite of what the word suggests.
• •The “vs 30-day avg” comparisons fire at ±15 %, over the sessions of the last thirty days containing the first dog, ACROSS ALL DISCIPLINES. A canicross compared with an average dominated by bikejoring will always come out far below.
• •“Estimated recovery” is a table of fixed steps tied to intensity and distance, not an individualized calculation.

Base — WHICH sessions go into the comparison

• •Same course: runs starting from the same spot within a hundred meters or so, over an equivalent distance within 500 m, in the same discipline. This is not track recognition: two completely different loops that start from the same parking lot and cover the same length are grouped together.
• •Similar distance: same discipline, distance within ±25 %. Looser, so terrain becomes a factor that muddies the reading.
• •Auto: takes “same course” as soon as there are at least two passes, and falls back to “similar distance” otherwise.
• •In all three cases, the window is 180 days and only EARLIER sessions count. A twelve-month-old puppy is no longer the same athlete six months later; comparing with later runs would also make the number unstable retroactively.
• •A greyed-out button means no session meets that criterion. The reason is written just below the button group.

Reference — WHAT you are compared with inside that set

• •Usual: the MEDIAN, never the average. A single exceptional race in the set would be enough to pull an average, but not a median.
• •Best: ONE real session, picked on average speed (distance divided by time), among those that have a usable kilometer breakdown.
• •It is never a point-by-point maximum. Taking the best of each slice would manufacture a session nobody ever ran, and would put you in deficit everywhere, including on an excellent run.
• •In “Best” mode, ONE single session feeds everything: the charts, the diagnostic's reference column, the raw gap. Plural phrasings of the “vs 1 passes” kind are a wording flaw, not a wrong calculation.

How to read the three

• •The chart: the solid line is you; the dashed line is the reference; the area between the two is filled red when you are below, green when above. It is the area that reads at a glance, not the two lines. The vertical red band marks the costliest segment.
• •The delta bars: the same thing, slice by slice, around a central zero.
• •The map: your track colored by the same delta, tied to the GPS points by distance covered. Hovering a slice gives its percentage and its delta as a number.

The order of the nine charts is not alphabetical — we go from symptom down to cause

• •Speed, in km/h — the symptom. Higher is better, so green is above.
• •Heart rate, in bpm — LOWER is better, so here green is BELOW. That is deliberate, not an accidental flip.
• •Power, in watts — the power meter's. Absent if there is no meter, rather than a flat line at zero that would read as information.
• •Cadence — in crank revolutions or in strides depending on the discipline.
• •Cardiac cost, in bpm per km/h — how many beats each unit of speed costs you. This is THE chart that tells “I slowed down” apart from “I slowed down AND it was costing me more”. The second is a problem, the first only a pace.
• •Power cost, in watts per km/h — the same reasoning with the power meter.
• •Altitude, Gradient, Temperature — the context. Always grey: they explain the first six, they are not judged.

The four numbers under each chart

• •Min and Max: the extremes of the SLICES, not instantaneous measurements. The “Max” of speed is therefore not your top speed: it is the fastest of your forty slices.
• •Slice avg: the average of the slices, which gives the same weight to each piece of distance. Your average speed, on the other hand, gives the same weight to each second — and since you spend more time in the slow slices, it is always lower. A concrete example: 4 km at 24 km/h then 1 km at 6 km/h give 20.4 as a slice average and 15.0 as a real average speed. Both are right, they do not measure the same thing.
• •vs ref: the delta with the reference, computed ONLY on the slices where both curves have a measurement. Without that precaution, a two-minute stop was enough to invent a cardiac-cost delta colored red.

Three things to know before concluding from a chart

• •The vertical scales are NOT shared between charts. Each one fits its own values. Two dips that look identical on two graphs bear no relation in amplitude.
• •A perfectly flat plateau can mean “no measurement here”: an empty slice takes its neighbor's value, because showing zero would dig a false chasm.
• •The two cost charts cut off below 4 km/h. A long stop or a walked section silently disappears from them — that is deliberate (140 bpm at 0.4 km/h would give 350 and crush the whole scale), but it means those two charts do not necessarily cover the whole session.

How the model is built

• •It does not use a classic regression but a MEDIAN-of-slopes method: a real training log contains races, truncated sessions and GPS blips, and a single outlier is enough to tip a classic regression over. Not a median.
• •Two variables are estimated together, temperature and elapsed time. Without that, a season that warms up would make the dog's progress look like heat sensitivity.
• •The slope is CAPPED at zero or less: a dog does not go faster because it is hotter. An estimated positive slope is an artifact, and is brought back to zero.
• •The reference speed is the one the model predicts at 12 °C, at your fitness in the MIDDLE of the analyzed period. Twelve degrees is an anchoring convention, not a physiological norm; the middle of the period keeps every number from moving each time a new session is imported.
• •It takes at least 8 sessions and 6 °C of spread between the coolest and the hottest. Below that there is no model at all, and the “Due to heat” line shows “—”.

The limits, said plainly

• •This is not causality. It is a regression on your training log: everything that varies alongside temperature — the start time, the season, the surface, your motivation — is absorbed into the slope attributed to heat.
• •The model is fitted on a set of sessions DIFFERENT from the one your session is compared with: 365 days for the model, 180 for the comparison, with filters that are not the same.
• •A heat share of 0.00 is ambiguous: it can mean “this dog is not very heat-sensitive” or “the estimate was absurd and was neutralized”. The screen does not tell the two apart.
• •It is a linear model: a constant loss per degree, with no threshold and no tipping point. Humidity enters no calculation at all, even though it is a major factor in a dog's heat risk.
• •If a session carries a watch-probe temperature rather than a weather temperature, it enters the model with that bias.

Six common mix-ups, to settle once and for all

• •All heart rate, calorie and measured-power data is YOURS. The app measures nothing cardiac on the dog. The 👤 dot marks it on the tiles.
• •The watts in the tiles (sensor, human) and the traction watts (model, whole system) are two unrelated things. It is normal for them to show very different numbers.
• •A kilometer's “Δ elev.” is a net difference; “Elevation gain” is a cumulative sum of climbs. Same screen, two quantities.
• •“Course segments” (automatic split by gradient) and “Segments crossed” (Strava public segments) are two different blocks despite the similar names.
• •The colors of the speed map and those of the loss map are RELATIVE to your session. They do not compare from one run to another.
• •Missing data shows as “—”. A displayed “0” is a real zero value.

To make these numbers mean something

1. Tag the dogs that actually pulled. That single piece of information drives power per dog, team records, the things to watch and the whole comparison block.
2. Check the discipline. It picks the traction model, the cruising speeds, the cadence unit and the comparison set.
3. Wait for the GPS fix before starting, and turn off battery saver on long runs.
4. Do the same loop again. Almost everything this page describes becomes far more reliable as soon as there are two comparable passes on the same loop.
5. If you have a heart rate strap, wear it: it unlocks the zones, the cardiac cost, the drift and a good part of the diagnostic.