The Science of Running Projections: Week One Round-Up

The science of running race prediction is clear: one generic guess cannot learn your body. Every runner is different. Your aerobic base, your threshold, your stride, and your training patterns are your own. One formula for all will get some runners right by luck. It will miss most by a wide margin. Learning the individual runner wins every time. That is the idea at the heart of this week's posts.

This week covered five big myths about race prediction. Each one harms runners who believe it. Below is a quick recap of each myth and the real answer.

Week 1 myth	The PIRX correction
Your smartwatch gives you a real race time	Watches estimate VO2 max from pulse and pace, then apply a hidden formula. Error rises near 10 percent in trained runners.
A single fixed formula works for every distance	It handles short races, but overpredicts the marathon by 10 or more minutes for about half of all runners.
More mileage always fixes a flat time	Five drivers shape your race time. Flat times and tired legs each point to a different driver to fix.
One recent run is enough to predict your race	A single run is noise. Six to twelve months of history narrows the Supported Range to a reliable band.
VO2 max tells you your marathon time	Among trained runners, running economy and lactate threshold predict the marathon far better than VO2 max alone.

Why do watch race predictions often get it wrong?

Watch race times often miss because they start with a guess. Your watch never measures your breath. It reads your pulse and your pace. It runs both through a hidden formula. It calls the result your VO2 max. A study found the error reached about 10 percent in fit runners (Engel et al., 2025). Fit runners care most about a goal time. They also get the shakiest number.

Even a perfect VO2 max is not enough. Running economy is how much fuel you burn at race pace. It can vary by up to 30 percent among runners with the same VO2 max (Barnes & Kilding, 2015). Your watch cannot see that. It also cannot see the heat at the start line. It cannot see how deep your last long run went. Day 1 broke this down in full detail. The takeaway is this. A wrist score is a trend tool. It is not a race-day number.

30%

how much running economy can differ between runners with the same VO2 max (Barnes & Kilding, 2015)

Are race calculators more reliable than watch scores?

Race calculators do better for short races, but they fail at the marathon. The most common type applies one fixed fade rate to every runner. The rate says how much your pace drops as distance grows. It is the same rate for all. Pick two runners with the same 5K time. Feed both into a calculator. It gives both the same marathon number. But one may log 55 miles a week. The other may log 25. Their marathons will not match.

A study of over 2,000 runners showed this clearly (Vickers & Vertosick, 2016). The fixed formula gave marathon times at least 10 minutes too fast for about half of all runners. Adding one personal number, weekly mileage, cut that error by about 45 percent. One real input did what the formula could not.

Research across 25,000 marathon seasons found that each runner has their own endurance trait. This trait shows how well they hold pace over long miles (Kirsch et al., 2020). It varies widely. A fixed formula gives everyone the same value. Day 2 explored why calculators break down at the marathon.

What do flat times and tired legs actually tell you?

They tell you which of your five drivers needs work. Flat times and tired legs are not random. Each one points to a specific driver.

A flat race time for two seasons points to Threshold Density. This driver tracks time spent near your lactate threshold each week. Without that work, your top race pace does not grow. Runners who always feel tired but never feel fitter have a Load Consistency problem. Sudden mileage spikes raise injury risk and block the body from adapting (Gabbett, 2016). Easy runs that keep getting harder point to Running Economy slipping. The middle-zone trap hurts Aerobic Base. Runs at a 6 or 7 out of 10 effort are too fast to build the base and too slow to push the threshold. Athletes who ran easy on easy days and truly hard on hard days outperformed those stuck in the middle (Stoggl & Sperlich, 2014). Day 3 maps all five signs and their fixes in detail.

How much training data does a race predictor really need?

It needs at least 6 to 8 weeks to give a stable answer. A single run is a snapshot of one day. It reflects sleep and heat just as much as real fitness. A study of 15 million activities from 300,000 runners found that a one-week gap in training cost 5 to 8 percent of finish time (Feely et al., 2023). One skipped week shifts the result by that much. One run is far too thin.

Six to twelve months is the sweet spot. That window covers a full build and a peak. It holds the mix of runs needed to score all five drivers. More data also narrows the Supported Range. With one run, the range may span 10 or more minutes. With months of steady history, it tightens to a small, honest band. Larger samples give more precise estimates in any field (Hazra, 2017). Training history is no different. Day 4 walked through how each added week sharpens the picture.

Is VO2 max really not enough for marathon prediction?

VO2 max is useful for new runners. It gets weaker the more trained you become. In a group of fit runners, nearly all have a large oxygen ceiling. So ceiling size stops deciding the race. What matters is how well you use that oxygen. And how long you stay below the pace where you start to fade.

In one study, all runners had the same VO2 max. Running economy still explained 65 percent of the gap in 10K times (Conley & Krahenbuhl, 1980). That finding has held up for decades. Two runners can share the same ceiling and burn fuel at very different rates. Over 26.2 miles, that gap adds up to many minutes.

Lactate threshold adds more to the picture. It is the fastest pace you can hold before you start to fade. The marathon is run near this threshold. A higher threshold means you can push harder for the full race (Faude, Kindermann & Meyer, 2009). Two runners with the same VO2 max can have thresholds a full minute per mile apart. Day 5 built the case for using all five drivers instead of one score.

What does the case study from Day 6 show us?

Day 6 followed a real runner's data from training to 5K race day. Watch estimates and fixed calculators both gave numbers built on averages. The runner's own data told a different story. Their Load Consistency was high. Their Running Economy had grown over six weeks. Their Threshold Density was rising. All five drivers pointed to a faster time than the generic tools called. The actual 5K matched the data-driven projection. The lesson is simple. The gap between what generic tools see and what your own training shows can be large. Real data closes that gap.

Why does PIRX approach race prediction differently?

PIRX reads your actual training. It is a proprietary machine-learning prediction engine. It looks at your real runs from your Garmin, COROS, or Strava. It breaks your fitness into five drivers. Those are Aerobic Base, Threshold Density, Speed Exposure, Load Consistency, and Running Economy. Each driver adds or costs you seconds. Those seconds sum to your total change from your Baseline Race. You see exactly where your time comes from.

Your Projected Time recalculates after every synced activity. A visible change shows only when the shift is at least 2 seconds. So you see real signal, not noise from a single hard day. You also get a Supported Range. It is the band of likely finish times based on your own profile. It narrows as your data grows.

98%

validated prediction accuracy (PIRX users)

This accuracy is not possible from a wrist estimate or a fixed formula. It comes from reading all five drivers across months of real training data.

What should you do after this week's science of running race prediction?

The pattern across all five posts is the same. Generic tools fail because they treat every runner alike. The fixes are also the same. Use your training history. Score all five drivers. Track changes in seconds across weeks, not just your finish time at one race.

A flat year does not mean you have hit your limit. It likely means one driver has stalled. Tired legs do not always mean more miles. They may mean your Load Consistency is off. Each problem has a clear fix. Finding it starts with the right data.

Get your real number from your own data

Stop guessing what your training is worth. Do not rely on generic watch algorithms that treat every runner the same. Connect your Garmin, COROS, or Apple Watch to PIRX in under 30 seconds. See your 5 performance drivers. See exactly where your seconds are coming from. Get the most accurate race-time projection on the market. It is free. You do not need a new watch. Your existing data is enough to get started.

Sources

• Engel et al., 2025. Validity of VO2max estimates from the Forerunner 245 smartwatch. European Journal of Applied Physiology. (Watch under-reads VO2 max in trained runners; error rises near 10 percent in fit athletes.)
• Barnes & Kilding, 2015. Running economy: measurement, norms, and determining factors. Sports Medicine. (Running economy can vary up to 30 percent among runners who share the same VO2 max.)
• Vickers & Vertosick, 2016. An empirical study of race times in recreational endurance runners. BMC Sports Science. (Fixed formula predicts marathon times 10 or more minutes too fast for about half of recreational runners; adding weekly mileage cuts error by 45 percent.)
• Kirsch et al., 2020. Human running performance from real-world big data. Nature Communications. (Each runner has their own personal endurance trait; it varies widely across 25,000 marathon seasons.)
• Gabbett, 2016. The training-injury prevention paradox. Br J Sports Med. (Steady training load reduces injury risk and keeps fitness gains coming.)
• Stoggl & Sperlich, 2014. Polarized training has greater impact on key endurance variables. Frontiers in Physiology. (Hard/easy split outperforms middle-zone training; athletes in the middle zone get flat race times.)
• Feely et al., 2023. Estimating the cost of training disruptions on marathon performance. Frontiers in Sports and Active Living. (15 million activities from 300,000 runners showed training gaps of 7 or more days cost 5 to 8 percent of finish time.)
• Hazra, 2017. Statistical estimation and sample size. Journal of Thoracic Disease. (Larger samples produce better, narrower estimates; the same logic applies to training-history prediction.)
• Conley & Krahenbuhl, 1980. Running economy and distance running performance of highly trained athletes. Med Sci Sports Exerc. (Among trained runners with similar VO2 max, running economy explained 65 percent of the variance in 10K time.)
• Faude, Kindermann & Meyer, 2009. Lactate threshold concepts: how valid are they? Sports Medicine. (Lactate threshold predicts endurance race results closely, often better than VO2 max.)