SomviaHow the prediction is made
The chain is short and each link is visible. A hard sustained effort in a recent session implies an oxygen cost, by the equations behind the VDOT tables. That single fitness figure then implies, for each race distance, exactly one finishing time that would be consistent with it, found by solving the same equations in the other direction. Somvia does that for 5 km, 10 km, the half and the full marathon, and shows how each has moved over the last months.
Because the prediction and the fitness estimate come from one model used forwards and backwards, they cannot contradict each other. A predicted time that improves while the fitness estimate falls would be an artefact of mixing two models, and this arrangement rules it out.
Why not the Riegel formula
The familiar approach to predicting one distance from another is Riegel's power law, where a known time is scaled by the distance ratio raised to an exponent. It is a good formula, but it needs two all-out results at clearly different durations to fit its exponent, and normal training does not provide them. Fall back to the textbook exponent and you have a lookup table with someone else's fatigue profile in it.
Somvia uses the VDOT equations instead, which need one qualifying effort rather than two, and which are the same equations already producing the fitness number on the trends screen.
What the marathon prediction assumes
All four distances are extrapolated from an effort of twenty to thirty minutes, so the further the race is from that duration, the more the prediction assumes. The marathon assumes the most: that you have done the long runs, that fuelling holds up, and that your durability matches your aerobic capacity. Treat the 5 km and 10 km numbers as tight, the half as reasonable, and the marathon as the ceiling a well-trained version of your current fitness would reach.
Predictions describe fitness, not readiness. They say nothing about the taper, the weather, the course or the day. Read them as a moving measure of what the training is producing.
What it needs
- A running workout in Apple Health with both heart rate and speed recorded.
- Inside it, a sustained stretch run genuinely hard, not merely the fastest part of an easy run.
- A maximum heart rate for your account, either measured from your own sessions or entered by hand.
With none of those, Somvia shows no prediction rather than one built from easy running. One hard session brings all four distances back.
Where the method comes from
The two equations are Daniels and Gilbert's, published in Oxygen Power in 1979 and familiar as the VDOT tables in Daniels' Running Formula. Somvia solves them numerically for each distance; the results agree with the printed tables to within a few seconds. Sources for every model are on how the scores work.
Questions
How accurate is a race time prediction?
It is an estimate of what your current aerobic fitness implies, not a forecast of a race. Shorter distances are tighter than longer ones because they sit closer in duration to the effort the estimate came from. Weather, course, pacing and taper are not modelled at all.
Why did my prediction disappear?
Predictions come from a qualifying hard effort within the recent window. If none of the sessions in that window contained one, Somvia stops showing a number rather than extrapolating from easy running.
Does it work for cycling?
No. The equations behind it are running equations. Cycling fitness is expressed as critical power instead, on the threshold page.
Can my AI Agent read these numbers?
Yes. Connected over MCP, an Agent can read the predictions along with the fitness estimate and the sessions behind them, so you can ask what a target time would require rather than reading the four numbers alone.