Your health comes first. This educational article provides general information. Discuss your individual circumstances, pregnancy, illness or symptoms with your clinician; the article cannot replace an individual assessment.

Estimating critical speed no longer necessarily requires the result of a separate assessment test. Some methods search previous training files for the fastest efforts over different distances. It seems convenient: the watch recorded, the app calculates. Yet the diary does not always capture the limits of your performance. Often it simply contains what was appropriate for you to run during those weeks.

The difference matters. If you have done only easy runs for months, even your best thousand metres may come from an easy run. The file can still be accurate and the line fitting flawless. The next question is what the number describes: the pace of the available training, or a genuinely appropriate assessment of your performance profile.

1. What do we mean by critical speed here?

Critical speed, abbreviated CS, is a model parameter relating speed to the duration for which exertion can be sustained. A common distance–time relationship is: distance = D′ + CS × time. Here, CS is the line’s slope. D′ is the other parameter, expressed in units of distance and appearing as the intercept. Together they form the model, rather than a single highlighted split time.

The physiological interpretation of CS is more complex than “I can hold this pace indefinitely”. Within the model, it helps describe different domains of sustainability, but does not promise you an infinite running duration. The number must not automatically be used as marathon pace or as a threshold that rules out illness.

The original research analysing raw training data estimated CS using, among other things, best efforts over different distances. The procedure was examined in data from many recreational marathon runners. This supports the method’s potential in an appropriate database, rather than showing that a few arbitrary easy runs from any individual suffice for the same conclusion.

2. A fictional example you can calculate by hand

Take three illustrative pairs: one thousand metres in 240 seconds, two thousand in 540 seconds, and four thousand in 1,140 seconds. These data were chosen for the calculation; they are neither a testing programme nor an expected performance. Keeping time in seconds and distance in metres means that the line’s slope will be in metres per second.

Distance Time Average pace
1000 m 240 s 4:00 min/km
2000 m 540 s 4:30 min/km
4000 m 1140 s 4:45 min/km

Between the first two points, distance increases by one thousand metres while time increases by three hundred seconds. The slope is therefore 1000 / 300, or approximately 3.33 m/s. This is 12 km/h, or 5:00 min/km. The third point also lies on the same line. In this example, D′ is two hundred metres, because 1000 − 3.33 × 240 is approximately that amount.

The resulting 5:00 is not identical to the average pace of any effort in the table. This shows that we estimated a parameter rather than simply choosing the mean of the three paces. Real data generally do not fall so neatly on a line; more measurements and a fitting procedure are needed. The example’s perfect fit is a deliberate educational simplification.

3. The diary’s best effort is not necessarily the limit of ability

Now imagine another equally accurate dataset. The runner completes every easy workout at six minutes per kilometre: one thousand metres in 360 seconds, two thousand in 720, and four thousand in 1,440. These points also form a perfect straight line. Its slope is roughly 2.78 m/s and its intercept is zero. It is mathematically neat, but does not establish that the runner cannot perform a faster effort.

Here, the data have plotted the chosen training pace. The runner may have stayed entirely within a comfortable range whose purpose was precisely to avoid exhaustion. The processing program, however, sees only the files supplied. It cannot draw direct conclusions about an unrecorded fast performance.

Therefore, describe the dataset’s nature in a separate sentence: did it include races, targeted intense efforts or only easy runs? Also ask which distances have recent, high-quality data. If the thousand-metre best is recent but the four-thousand-metre effort comes from an old recording, they may no longer describe the same condition.

A runner checks a watch at a safe stopping place among tall urban buildings, where the surroundings can make positioning more difficult.Enlarge image
AI-generated editorial illustration; not a documentary photograph.

4. Why do you need information over different distances?

A line’s slope depends on the relationship between two points. If the efforts cover only very similar distances, the estimate may be sensitive to small time differences between them. A single short distance is compatible with many possible lines. A longer effort can add information that a short, fast segment alone does not contain.

Different distances help only when they can meaningfully be compared. A short, steep downhill sprint and a longer muddy uphill effort are not a good combination, for example. The distance–time series then mixes different environmental difficulties. The resulting parameter does not arise purely from changes in your performance with duration.

Do not interpret the method’s data requirements as a need to complete several maximum efforts immediately. This article gives no such testing programme. First examine what is already available and the question the estimate needs to answer. If a new assessment is justified, discuss its demands and execution conditions with a knowledgeable professional.

5. One incorrect long effort can shift the whole picture

Return to the fictional example and a simple calculation using the two outer points. Keep the thousand metres at 240 seconds, but suppose the four thousand metres is mistakenly entered as 1,020 seconds. The slope then becomes 3000 / 780, or roughly 3.85 m/s. That corresponds to approximately 4:20 per kilometre rather than the previous 5:00.

This is not an estimate of a real algorithm’s expected error. It illustrates how changing a longer effort’s time can substantially rotate the entire fit. A different erroneous point may bias it in another direction, so correcting every apparently fast outlier with the same formula is inappropriate. First understand how the point arose.

A GPS trace jumping off the course, missing time data, automatic pauses or an effort not actually recorded while running can be suspicious. Recording a bicycle journey home under the wrong sport can produce an extraordinary running performance in the diary. Automated searches for best efforts are particularly liable to highlight such extreme values.

6. Gradient correction and stops: which time does the system use?

High downhill speed is not directly comparable with running on the flat. Some models use pace adjusted for uphill and downhill gradients. This may be useful, but the correction is itself a model. It does not necessarily handle traction, technical ground, turns and individual downhill experience in the same way.

Your watch’s similarly named metric does not automatically replace the correction applied in a research method. First examine which data and assumptions were used. If distance or elevation data are incorrect, their correction may also be inaccurate. A smoothly presented curve does not establish that the underlying data were appropriate.

The timing basis is equally important. Total elapsed time including stops may differ from a field summing active time only. If two rests interrupted a long effort, a “continuous performance” calculated from stitched-together moving time describes a different phenomenon. Methodological notes should state how pauses and continuity were handled.

7. A good line is not yet a good marathon prediction

Fit shows how closely the line follows the selected points. Marathon prediction asks another question: how does the estimated parameter relate to the outcome of a much longer race? An excellent measure in the first task does not automatically guarantee success in the second.

The six-minute easy-running example is enough to show this. Its line is flawless, yet we have by no means necessarily described the limits of performance. The points’ substantive meaning is also needed for model validity. If you omit it, elegant mathematics lends authority to input that has not been shown appropriate.

The empirical research examining race times over different distances likewise indicates that transfer to a longer race requires separate validation. Finding a good relationship at one distance is insufficient. Marathon use also requires examining the estimate’s reliability for the target outcome and the population in which it is applied.

A runner reviewing training-diary data adds a separate note about an uncertain measurement.Enlarge image
AI-generated editorial illustration of marking suspicious sensor data in a traceable way; not a documentary photograph.

8. What should you examine in the original file?

The primary documentation for the FIT activity format treats individual records, laps and session summaries separately. Coordinates, speed and cumulative distance do not necessarily arise from the same simple calculation. Keeping the original file is therefore useful, rather than only the peak value displayed in the app.

For a practical check, choose a few efforts influential in the estimate. Examine their start and end times, mapped route, sport, pauses and notes on conditions. You do not need to inspect every second manually. Checking the most influential points may add more value than browsing the entire diary indiscriminately.

Do not treat missing data as zero, or delete a suspicious record without a trace. Note why it was excluded from the estimate. If the app’s result later changes, you can then trace whether this arose from a new performance, corrected data or a different processing rule.

9. A useful information sheet for discussion with your coach

In a short summary, list the distances used and each result’s date and origin. Identify separately whether the best split came from a race, targeted training or an easy run. Also note if the section was downhill, contained pauses or had uncertain positioning data. This sheet supports interpreting the estimate; it does not label someone a good or bad runner.

Use the next line for the purpose. Cautious trend monitoring requires different data quality from supporting a narrow marathon time goal. When comparing new and old numbers, consistent processing may matter. When determining race pace, the background of longer exertion and the course also belong in the discussion. A single metric cannot bear responsibility for every decision.

Finally, state what remains unknown. If there is no suitable longer effort, say so; if there are only easy runs, state that too. There is no need to manufacture an apparently stronger number from weak data. Critical speed becomes useful information when you can place beside it its origin, the conditions of calculation and the decision for which sufficient evidence supports it.

Sources

Helpful? Pass it on.