How MushMate builds your plan — and why you stay in charge

Your MushMate by Skimate.eu training plan does not come out of a hat, and it does not come out of a veterinary study either. It comes out of a model called CAMT — Canine Adaptive Multi-discipline Training. Here is how it works, what you can set, and above all where it stops.

This is a long read because it hides nothing. By the end you will know what to adjust, what to leave alone, and when your judgement should outrank the app's.

Read this first

Not one number in this model has been clinically validated. That is the most important sentence in this article, and it is written in those words in our own technical documentation.

The constants that drive your plan come from three places, and nowhere else:

1. Human endurance physiology. The core fitness-and-fatigue calculation was designed for cyclists and runners. We transposed it to sled dogs. Nobody has shown that the transposition holds.
2. Standard periodisation templates. Splitting a season into four phases, and the size of each phase, comes from common training practice — not from a study on racing dogs.
3. Deliberate design decisions. Where neither the literature nor a vet settled the question, we picked a value and wrote it down somewhere marked "provisional".

None of these values has been fitted to a real season's data. They are what statisticians call priors: starting guesses. The model is designed to personalise them dog by dog as it watches you — that learning loop is wired but dormant, and will stay dormant until a veterinary dossier signs it off.

Exactly one reference in this model is genuinely science-backed: the 9-point BCS scale (Laflamme), whose relevance to long-distance sled dogs was established by Maffi et al. (2025). Everything else is our best estimate.

What that means in practice. CAMT is a numerate notebook, not an authority. It is good at spotting a progression that climbs too fast, a dog falling behind, a week that doesn't add up. It is not qualified to tell you that your four-year-old can do 48 km tomorrow. That is you, and your vet.

---

1. The chain: from one run to the plan

The whole model is four steps. Each one only knows the previous one.

Step 1 — what you do

A run is a distance, an elevation gain, a towed weight, an average speed, a temperature, and the list of dogs in harness. Nothing else goes in.

Step 2 — what it costs: the load

The model turns each run into a single number per dog: the load, or TLI. It is not an empirical formula but a force balance — the mechanical work actually done:

• the friction of the rig on the ground, which depends on the surface;
• the slope, which adds to friction;
• air drag, which is the only place speed enters at all.

That last point is worth pausing on, because it is counter-intuitive: work per kilometre barely depends on pace. Towing 400 kg over 20 km costs the same at 12 km/h as at 16 km/h, air resistance aside. An earlier version of the model multiplied everything by speed squared, with no physical derivation behind it. That is fixed.

That work is then shared among the dogs, converted into energy expenditure per dog, corrected for heat and rest, and divided by a reference energy. That reference energy is a constant we chose. It does not change the ratios between sessions, but it sets the scale on which every safety threshold is calibrated.

Step 3 — what it leaves: fitness and fatigue

Every dog carries three numbers, recomputed after each run:

Fitness rises slowly and fades slowly. Fatigue spikes fast and clears fast. String sessions together and ATL climbs faster than CTL, TSB dives, and the model brakes.

Both constants were changed by hand on 20 June 2026: fitness from 42 days to 21, fatigue from 7 to 4. They are borrowed from human endurance work and recalibrated on intuition, not measurement.

When a dog's TSB drops below −25, it is declared in the red zone: that is the safety floor, and the whole team's volume ceiling keys off the lowest dog. The −25 is a prior — set too negative, a dog banks more fatigue before the brake bites.

Step 4 — what we suggest

Out of that come the week's volume, the pace ceiling, the training days, rest and alerts.

And the loop closes: what your dogs actually did feeds back into step 1. When you regenerate a plan, weeks already in the past are copied over verbatim, and only future weeks are recomputed — on what really happened, not on what was scheduled.

---

2. The season: four phases

The runway between your season start and your objective is split into four phases. Default shares are 50 / 25 / 15 / 10 %, tunable, with a one-week floor per phase.

Each phase carries two things: a volume multiplier applied to the discipline's ceiling (× 0.7 in base, × 0.9 in intensity, × 1.0 at peak, × 0.4 in recovery), and a pace ceiling.

"Up to 16 km/h" is not an instruction to do 16

This is the product's most expensive misunderstanding, and a musher reported it to us. The sled pace ladder is 16 / 18 / 20 / 14 km/h by phase. Those are ceilings, not targets. Our documentation has always said so: "pace is permission, not a target — slower is always safe."

In base work the whole point of the phase is volume at low pace. If your dogs are comfortable at 10 or 12 km/h, run at 10 or 12. Running at 16 because the screen says 16 is precisely the opposite of what the phase asks for — and the screen didn't say so. It does now.

The only direct lever you have on that number is your team's VMAX, in the objective's advanced settings.

Volume ceilings by discipline

An objective classed as mid-distance uses half its discipline's ceiling. The threshold between mid and long is tunable.

All four rows are marked "provisional" in our documentation. The sled one is even explicitly suspected of sitting a touch high.

---

3. The gates: what runs before the maths

Six causes can stop or cap a dog whatever the maths says. They do not negotiate with the model: they apply after it.

Five force full rest — load zero:

• the bench, when you have explicitly put the dog on rest;
• an illness logged at "severe" severity;
• an active restricting medication;
• a BCS of 1–2 or 8–9, too thin or too heavy;
• a weight drop of more than 8 % below baseline.

A sixth caps without stopping: a BCS of 3 or 6–7 puts the dog in the soft band, where it trains at half its load.

And the whole team moves at the pace of its most restricted dog.

Three things worth knowing about these gates:

• The bench is the only one of the six that is a decision rather than a measurement. It has no setting, and never will.
• The 8 % weight-drop threshold is doubly flagged as provisional in our documentation: "no validated vet figure exists yet." And it compares against a baseline weight you enter by hand. The real failure mode is not "the dog is losing weight" but "the baseline is stale". Keep it current.
• The illness threshold is set to "severe". A moderate injury therefore blocks nothing on its own.

---

4. Young dogs

The model sorts each dog into an age band, and each band carries a capacity ceiling: under 12 months, then 15, 18, 30, 96, and beyond.

Three regimes exist — cautious, standard, sport — and your kennel picks one. It is probably the most consequential setting in the whole app, and also the one whose values are the least substantiated.

Two honest clarifications:

• Age is re-evaluated every week of the plan. A dog that crosses a threshold mid-season changes band mid-plan.
• Since August 2026, maturity no longer shrinks the whole team's volume. Previously the youngest pup throttled the entire pack. That was judged a design error: the pack keeps its adult volume, and the young dog simply drops out of the sessions that exceed its own ceiling.

That last change created a defect we fixed the same day we measured it: a team with no adult at all — three young females and nothing else — was getting sessions nobody could hold, plus 227 warnings that nothing enforced. A session no dog can hold is now brought back down to the team's own ceiling.

---

5. What influences what

Two consequences are worth remembering, because they surprise people:

• Closing a day does not cut your volume. Declare Sunday unavailable and the kilometres spread over the days that remain. You are not training less; you are training differently. Rest, on the other hand, is counted in open days — a closed day mid-week lengthens the real rest.
• VMAX is the only direct lever on the displayed pace. Neither days, nor equipment, nor starting fitness moves it.

---

6. The priors, one by one

Here are the values that drive your plan, and where they come from. "Borrowed from humans" means: transposed from a literature that is not about dogs.

---

7. How to set it up, concretely

What you must get right — everything else rests on it

• Each dog's baseline weight, kept current. It arms the weight-drop gate. Stale, it protects nothing.
• BCS, scored regularly. It is the one measurement in the model that rests on established science, and it triggers two gates out of six.
• Dates of birth. They decide the age bands, and therefore what the model will refuse to ask of a young dog.
• Your season dates and your objective. The phase split is not a fixed calendar: it is a division of the time you have left.

What you can adjust without worry

• Training days per week, and unavailable days. If you hunt on Sundays, declare it: the plan will stop putting a session there on every regeneration.
• Team VMAX, if the suggested pace doesn't match your dogs.
• The young-dog regime, knowingly: cautious, standard or sport.
• Maximum weekly growth (+10 % by default). Lowering it makes the plan more conservative.

What you should leave alone without a reason

The volume ceilings and heat cutoffs. Not because they are right — they are as provisional as the rest — but because they are the only automatic brakes on volume, and a ceiling set too high stops triggering anything at all.

What to read every week

• The limiting dog. If there is one, it dictates everyone's volume.
• The alerts. They almost never block; they flag.
• The displayed pace, remembering that it is a ceiling.

---

8. What CAMT does not know

This list matters as much as the rest.

Heat acclimatisation. This is the most serious hole, and we measured it in September 2026. The field literature and Dr Janice Baker give as the number one heat-risk factor not the absolute temperature but the gap against what the dog has experienced over the last fourteen days: a dog used to −5 °C is at risk at +8 °C, a dog used to +15 °C is not. That factor entered our model in no form at all. It exists today in the day's weather verdict, and nowhere else: not in the heat cutoffs, not in the load calculation. Every thermal threshold we have remains absolute.

Your ground. Deep snow, glare ice, a groomed trail: the model knows three surface profiles, not your trail.

Your dogs specifically. The loop that should personalise the constants dog by dog is written, but switched off. Today all your dogs share the same population priors.

Individual nutrition. The ration thresholds are calibrated on a history that, by our own admission, has never seen a winter.

A vehicle's speed. There is no ATV or kart speed parameter in the model at all. An ATV or kart only changes the towed weight.

---

9. So what do you do with it?

The plan proposes. You decide. That is not a courtesy — it is an exact description of the product: the model has no way to stop you doing what you judge right, and it does not seek one.

Use it for what it does well:

• keeping the books on what each dog has absorbed, which none of us does well in our heads across twenty dogs and six months;
• spotting a slope that is too steep — the progression climbing 30 % week over week;
• naming the dog that is holding the team back, and asking you the question at the right moment;
• not forgetting a dog on rest, under medication, or whose weight has dropped.

Don't ask it what it cannot know: whether your dogs are right today, whether the trail is good, whether the wind is turning. Look at your dogs. They, at least, have been validated by a few thousand years of selection.

---

A question, a disagreement, a figure that looks wrong to you? Write to us. The priors in this model are meant to move, and one well-argued report from the field beats a constant picked by default.