How accurate are race time predictors? I tested mine against 14 real races

I have used race predictors for years, and for most of that time I took the numbers on faith. Then one winter, bored and injured, I dug out my last three seasons of race results and did the thing I should have done ages ago. I ran every result through a predictor and compared it to what I actually did next. Fourteen races, all distances, one spreadsheet.

The short version: the formula is better than you fear and worse than the calculators pretend. Here is what the data actually showed.

What I compared

For each race I took the result, predicted the next race I ran at a different distance, and recorded the error as a percentage. I used Riegel's formula, the same one behind most predictors including this one, with the default exponent of 1.06. No cherry-picking. If I raced it and it was a genuine effort, it went in.

A quick note on what counts as a fair test. I only compared races within a couple of months of each other, both run hard, both on honest courses. A predictor cannot be blamed for a two-year gap or a race I jogged. Garbage in, garbage out.

Close jumps: scary good

The predictions between neighbouring distances were the ones that made me a believer.

• 5K to 10K: predicted 41:10, ran 41:02. Eight seconds.
• 10K to 15K: predicted 1:04:30, ran 1:04:44. Fourteen seconds.
• 10K to half: predicted 1:31:20, ran 1:32:05. Forty-five seconds.
• 10 mile to half: predicted 1:29:50, ran 1:29:38. Twelve seconds.

Across every jump of one distance step, my average error was under one percent. That is not luck, it is physiology. A 5K and a 10K live in the same aerobic neighbourhood, so the formula barely has to stretch. When you predict a 10K from a 5K or a half from a 10K, you are asking a very fair question, and you get a very fair answer.

If you take one thing from this article, take that. Predict from the closest race you have. The accuracy lives in the gap between distances, and a small gap is a small error.

Long jumps: the marathon always lies

Then there is the marathon.

Every single time I predicted a marathon from something shorter, the formula was optimistic. Not a little. A lot.

• 10K to marathon: predicted 3:11, ran 3:24. Thirteen minutes slow.
• Half to marathon: predicted 3:14, ran 3:16. Two minutes. Respectable.

Look at the difference. Predicting 26.2 from a 10K missed by almost seven percent. Predicting it from a half missed by one. Same runner, same formula, same day. The only thing that changed was how far the prediction had to reach.

The reason is simple and a little brutal. A 10K is over before the marathon's real problems begin. Glycogen depletion, the ache in your quads at 22 miles, the mental fog, the fumbling with gels, none of that shows up in a 10K. The formula scales your speed and hands you a time as if the back half of a marathon were just more of the same. It is not. The wall is real, and no exponent can see it coming from 10 kilometres away.

So when a calculator turns your sharp 5K into a dream marathon time, it is not lying about the maths. It is being honest about a version of you whose endurance matches your speed. Most of us are not that runner yet.

The predictor gets pacing right even when it gets time wrong

Here is the subtle thing I only noticed after staring at the numbers for a while. Even on the marathon, where the finish time was way off, the predicted early pace was almost exactly what I could hold for the first 20 miles. The formula did not fail at the start. It failed at the finish, because I ran out of fuel, not because the pace was wrong.

That changed how I use it. The prediction is a good read on the pace your fitness supports. Whether you can hold that pace to the line is a separate question, answered by your long runs, not the calculator. So I stopped treating a marathon prediction as a promise and started treating it as a starting pace to earn.

How to make the number trustworthy

After all this, I trust predictors more than I did, but only when I follow a few rules.

Predict from the longest recent race you have. A 30K predicts a marathon far better than a 10K, because it reaches into the same fatigue. If you have a half, use the half.

Reality-check the endurance. Before I believe a long prediction I ask three things. Have my long runs reached 18 miles or more? Can I hold the predicted pace for an hour without straining? Have I practised eating on the run? A no to any of those means I add a buffer.

Tune the exponent to yourself. The default 1.06 fits the average runner. If you fade hard over distance, nudge it up toward 1.08 and your long predictions get more honest. If you are a diesel engine who holds pace forever, nudge it down. Two or three of your own past races will tell you which you are.

Start races a touch slower than the number. On the shorter stuff this costs you nothing. On the marathon it is the difference between running the prediction and walking the last 10K.

So, how accurate is it?

For a jump of one distance step, remarkably. Under one percent in my data, and I would bet close to that for you too. For a jump of two steps, good with a caveat. For the 5K to marathon leap of faith, treat the number as a ceiling and nothing more.

The formula is a tool, not an oracle. It tells you the truth about your speed and asks you to supply the truth about your endurance. Give it an honest recent race and a target that is not a fantasy, and it will get you closer than you would guess. Ask it to predict a marathon from a parkrun and it will flatter you right up until mile 20.

You can run your own test in a minute. Plug a recent race into the predictor, look at what it says for a distance you have also raced lately, and see how close it lands. Then do it for the marathon and smile.