The science
Every session has a reason. These are the rules the engine runs on — the same rules the app shows you, in plain language, while you train.
Readiness is measured, then it gates the day
Each morning the app weighs heart-rate variability (35%), sleep (25%), resting heart rate (20%) and your own wellness answers (20%) into a 0–100 score — if a signal is missing, the weights renormalise over what exists rather than guessing. Two rules are hard: feeling good cannot override suppressed HRV, and two red mornings in a row force a rest day.
The plan has guardrails you cannot talk it out of
Weekly training load may grow at most 8%. Every fourth week is a recovery week at roughly 40% less load. The taper is engineered, and race week drops about 60%. Beta describes the product's maturity — it does not suspend the rules the app applies to your training.
Zones come from your tests, not your age
Power, pace and swim zones are anchored to your FTP, CSS and run test results, and the plan schedules test weeks to re-anchor them as you get fitter. Where a measurement exists, no age formula is allowed to overrule it.
Race projection is physics, not a promise
The bike projection prices aerodynamic drag and rolling resistance against your power — the same formula printed on the page — and the app is honest about what that is: a projection. No plan can guarantee a time, a result or a physical change.
Fuelling is personalised where you measured, generic where you did not
A measured sweat rate drives your fluid targets on every session and on race morning; unmeasured athletes get the defensible generic band, labelled as such. Carb rates respect gut training. Supplements are capped and evidence-graded — caffeine is dosed at 3 mg per kg and held at 200 mg.
Low energy availability is screened, not diagnosed
The app checks that the food in your plan can cover the training it prescribes, and says so plainly when it cannot. That is a statement about the plan, not about you: there is no food logging, the app cannot see what you actually ate, and it never asks for medical history. A screening prompt, not a diagnosis.
The AI layer proposes; you decide
Multi-week replans arrive as proposals you accept or decline; the morning decision adjusts only today, and it always shows its reasoning. Every AI decision has a deterministic fallback, so a model outage never blocks your day. AI coaching is off by default, runs only with your explicit versioned consent, and what leaves the server is documented in the privacy policy.