How we make methodology decisions

Every number in TriCoo has an origin: a study, a consensus from a professional body, a technical convention, or a cautious decision of our own. We think it matters not to mix these up and not to pass off a coaching habit as a law of physiology. This article is about how methodology decisions are made and why the platform is immovable in some places and offers you a choice in others.

Three levels of confidence#

Every rule carries a label saying how firmly it stands.

Hard-wired. Robust findings that cannot be adjusted: the taper before your main race, overtraining as a drawn-out state, loss of fitness during a confirmed break, fever and warning cardiac signs as an unconditional stop, the ban on automating clinical decisions, and the rule that no data does not mean rest.

Adjustable guardrail, labelled as such. Guardrails that exist to slow you down: the share of easy work, the weekly ceiling on load growth, the rhythm of recovery weeks, automatic threshold estimates, the way signals vote in the readiness traffic light. These are conservative values with a version number, not laws of nature: they change by configuration, and we say honestly that this is exactly what they are.

Optional and experimental. The cycle log as a personal diary, DFA alpha 1, durability from the power curve, personal fitting of the model's memory, a composite finish forecast. Features like these do not affect the plan until they are confirmed: they live off to the side and move nothing in your schedule.

The level label is not bureaucracy, it is a promise. If a rule is called an adjustable guardrail, it means we know it is debatable and are not passing it off as physiological truth. And a debatable rule is always set on the cautious side.

How we read the sources#

Every number carries a grade: strong — a consensus statement, a meta-analysis or several good-quality studies; moderate — the effect is there but transfers only so far; standard — a technical convention rather than physiology; heuristic — our own coaching default; disputed — there is substantial scientific criticism; weak — low-quality data or too much variation between people.

Only something that rests on a verified primary source or an official document can be hard-wired. A retelling with no checkable bibliography never becomes a hard rule — at most a labelled heuristic.

The whole bibliography is open: the list of sources — 123 entries, each with its level, from the consensus statements of professional bodies to our own heuristics.

We lean towards under-training#

There is no coach beside you to stop you on a Wednesday. So when in doubt the platform picks the easier option and backs off sooner rather than later. The cost of one extra rest day is trivial next to the cost of overtraining or training through illness.

A second consequence follows: uncertainty makes the plan more cautious, not bolder. While data is scarce you will see “calibrating”, a grey traffic light and a label next to your threshold — and in that state load is not increased. The asymmetry here is deliberate: an error towards rest is cheaper than an error towards load.

We propose rather than rewrite in silence#

An ordinary change arrives as a proposal showing “before — after” with “accept” and “reject” buttons. The decision is yours, and a rejected proposal does not come back through the back door.

The only things that apply by themselves are hard safety rules — for example, stopping the plan when you have flagged an illness. The platform tells you about that directly rather than pretending it was intended all along.

Planning and medicine are different things#

Fitness, fatigue and form are planning tools, not health indicators. A chart cannot tell you whether you are healthy, and we do not let it pretend that it can.

Everything to do with illness, injury and warning signs lives in a separate layer that routes you to a doctor. That layer always takes priority over the status of the day, and no setting can bypass it. More in “Safety and where medicine begins”.

Explainability#

Every decision keeps its reason: which rule fired and on which numbers. This is not interface decoration but the condition that makes it possible to argue with the platform: a rule you cannot understand is a rule you cannot check, and so cannot trust. If a decision cannot be explained, it should not reach the plan.

What we have decided not to do#

• We do not use the acute-to-chronic workload ratio as an injury predictor: the evidence around it is disputed.
• We do not hold the “10% rule” as a hard ceiling on growth: the data does not support it in that role.
• We do not fit the model's memory to the individual: on amateur data such fitting gives a beautiful match with the past and no predictive power.
• We do not show finish times, including by converting from another distance.
• We do not automate heat or altitude protocols, or training with low stores.
• We do not feed proprietary readiness scores from devices into the calculation: a closed formula, numbers that cannot be compared with one another, an unexplainable status.

Who signs this off#

Every decision has an owner and a date. Training defaults and the safety layer are approved by the platform's methodology function and signed off by the founder; contentious forks go for a second independent opinion, and its verdict is recorded next to the decision.

We do not write that these texts have been reviewed by a doctor: there is no external medical reviewer in the loop. Clinical decisions — diagnoses, contraindications, treatment — are not covered by that sign-off and are not produced by the product.