From Workout Data to Training Insight

Raw activity records become useful when they are cleaned, organized into indicators, and read in context.

By Run Briefing

Raw activity records become useful when they are cleaned, organized into indicators, and read in context.

Your last training week is easy to remember. You can still feel the long run, the hard session, the missed day, the fatigue.

That immediate feeling has value. If your legs hurt and you feel worn down, it may be enough to make a short-term call: rest today, keep tomorrow’s run easy, or avoid adding stress.

But it does not explain how you got there. The same tired legs can follow one demanding session, four weeks of rising load, a harder intensity mix, poor recovery, heat, hills, or life outside training.

Training insight usually comes from patterns. A sore day matters, but it means something different after a normal week than after several weeks of rising load, fewer easy days, and harder sessions.

That is where workout-data analysis helps. It does not replace how you feel. It gives that feeling a structure: what was recorded, how the data was cleaned, which indicators changed, and whether the change matters across time.

Raw workout data is only the starting point

Your watch or app records streams of timestamps, GPS coordinates, elevation, heart rate, pace, splits, pauses, route, and sometimes temperature, power, or gait dynamics.

Those records are called raw data for a reason. At one sensor reading per second, one hour of running across seven basic streams produces 25,200 recorded values.

That volume of data is useful, but it is also noisy. Heart-rate values from a watch are sensor observations of a physiological signal. GPS pace is an estimate of movement across changing terrain. Elevation can be imperfect. Pauses and interruptions can change what the activity means.

Raw values do not answer the basic training question by themselves: is your aerobic endurance improving, stable, or getting worse? They first need to be cleaned, summarized, and compared.

Deriving signals from workout data

The first layer is cleaning and normalization. Terrain adjustment is a simple example.

Grade-adjusted pace

Pace alone can punish uphill running and flatter downhill running. Grade-adjusted pace tries to reduce that terrain noise so hilly and flat runs become easier to compare.1

The point is not to make the data look more precise than it is. The point is to remove obvious noise before the workout is summarized.

A simplified version of the idea looks like this:

Grade-adjusted pace = raw pace adjusted for terrain cost

That is only one layer. Later layers can combine normalized pace with heart-rate behavior to study whether the same external output required more or less internal cost. That belongs to durability and aerobic-efficiency interpretation, not to the first read of raw data.

For this article, the important point is simpler: before workout data becomes training insight, it passes through interpretation layers.

flowchart LR
    A[Raw records] --> B[Cleaned data]
    B --> C[Normalized values]
    C --> D[Indicators]
    D --> E[Time context]

Indicators separate different kinds of change

Once workout records are cleaner, they can be summarized into training indicators. Each indicator gives you a specific angle on the week.

Indicator What it helps you see
Load How much running work the week represents.
Intensity How hard the work was, separate from volume.
Frequency Whether your running rhythm held together.
Performance Whether output quality appears to be moving.
Confidence Whether the evidence is clean enough to emphasize.

This separation matters. More distance is not the same as more intensity. A stable load with more hard running is different from a stable load made mostly of easy running. A performance signal with low confidence should be read differently from the same signal with clean data behind it.

A simple weekly load signal might start with a summary like this:

Weekly load = sum(running duration or distance across the week)

But the summary is still incomplete until it is compared with the surrounding pattern:

Load change = this week - recent baseline

The first useful question is not “was the week good or bad?” It is: what changed, which category changed, and how clearly does the data support that signal?

Time horizon gives the signal meaning

An indicator becomes useful when it has a comparison window.

This Week vs Last Week is the first comparison layer. It shows immediate movement: load up or down, intensity rising or easing, frequency holding or breaking, performance moving with or against the rest of the week.

A longer baseline adds context. Workload research often compares the most recent week with a longer preparation window, commonly a rolling 3–6 week average.2 The exact model matters less than the principle: recent training is easier to interpret when it is compared with the training that came before it.

That does not mean one ratio or one trend line can explain everything. Training-load monitoring can help show whether an athlete may be adapting to training, but research is clear that there is no single definitive marker for fatigue or readiness.3

That boundary is useful. It defines the role of a weekly training signal: show what changed recently, where the change appeared, and which time horizon gives that change meaning.

Where this appears in your weekly report

In your weekly Run Briefing email, the same pipeline runs underneath the report: raw workout tracking data is cleaned, normalized, grouped into indicators, and compared across time horizons.

The This Week vs Last Week view is the first comparison layer. It helps you notice recent movement before the report widens the view to longer baselines and trends.

Read the week by layer: what was measured, how it was cleaned, what signal it became, and which time horizon gives that signal meaning.

References

Footnotes

  1. Minetti AE, Moia C, Roi GS, Susta D, Ferretti G. “Energy cost of walking and running at extreme uphill and downhill slopes.” Journal of Applied Physiology. 2002;93:1039–1046. PMID: 12183501. The study measured running cost across uphill and downhill slopes and provides one physiological basis for terrain-adjusted pace models.

  2. Maupin D, Schram B, Canetti E, Orr R. “The Relationship Between Acute: Chronic Workload Ratios and Injury Risk in Sports: A Systematic Review.” Open Access Journal of Sports Medicine. 2020;11:51–75. PMCID: PMC7047972.

  3. Halson SL. “Monitoring Training Load to Understand Fatigue in Athletes.” Sports Medicine. 2014;44 Suppl 2:S139–S147. PMCID: PMC4213373.

Published