Preserve the original evidence
Begin with the original recording and a short description of the session’s intent. Was this a controlled threshold session, a tactical race, a recovery ride or an interrupted commute? Export the activity before making corrections. Keep an untouched copy so that a cropped warm-up, a merged recording or a removed power spike does not become the only surviving version. Record the source devices when switching between bicycles or indoor and outdoor setups.
Then inspect duration, timestamps, channel coverage and lap structure. A ride can have a credible average power while missing a decisive interval. Look for long flat lines, implausible spikes, repeated zero values and unexplained gaps. Confirm whether the analysis tool includes zeros in averages and how it handles paused time. Comparing moving-time statistics with elapsed-time statistics can create an apparent performance change without any change in the rider.
Distinguish channels from models
Power, heart rate, cadence and position are recorded observations, although each already depends on a sensing and processing method. Training load, estimated threshold and predicted performance are further interpretations. They can be useful without being direct measurements of the rider’s internal state. A model may be sensitive to the threshold setting, missing activities or the platform’s calculation choices. Treat the assumptions as part of the result.
| Recorded evidence | Possible interpretation | Check before concluding |
|---|---|---|
| Higher heart rate at similar power | Greater physiological strain | Heat, hydration, illness, sensor contact and conditions |
| Lower power late in an event | Reduced sustainable output | Terrain, tactics, fuelling, pacing and missing data |
| New short-duration peak | Improved capacity at that duration | Spike validity, device continuity and comparable opportunities |
| Lower modelled training load | Reduced accumulated workload | Unrecorded sessions, threshold settings and model definition |
Compare the work that was actually prescribed
For an interval session, review the work periods rather than the whole-ride average first. Use laps or carefully checked selections. Ask whether the target duration was completed, whether output was stable and what happened between repetitions. Recovery duration and terrain affect the task. Two workouts with identical average power can place very different demands on the athlete if one contains repeated surges and the other is steady.
Longer averages usually suppress noise, but they can also conceal pacing errors. Look at the trace alongside a small set of summary values. Display smoothing is useful on the bicycle, yet a three-second display value is not the same thing as the raw recording interval. Do not infer rapid physiological changes from a presentation setting. If a peak looks exceptional, inspect the surrounding samples and the relevant sensor history.
Bring context into the file review
Heart rate responds to more than mechanical workload. Temperature, hydration, sleep, stress and illness can influence it. A rising heart rate during a long steady effort may be useful context, but a single ride cannot establish a medical or training diagnosis. Likewise, perceived exertion provides information that a power meter does not measure. A short note about how the session felt often helps explain why a technically completed workout was unusually costly.
Ask the athlete for a small, repeatable set of observations: session intent, perceived effort, fuelling, notable environmental conditions and any symptoms or equipment problems. Keep sensitive health information out of public activity descriptions. For persistent symptoms, chest pain, fainting or concerning exercise responses, seek appropriate medical advice rather than using a dashboard to decide whether training is safe.
Terminology note: threshold is an operational estimate
A threshold value depends on the testing or estimation method and its intended use. A ramp test, a sustained field effort and a modelled estimate are not automatically interchangeable. Document the method and date rather than treating the number as permanent.
End with a decision, not another chart
A useful review produces a limited next step. That might be repeating a controlled session, checking a suspect meter, revising a pacing instruction or discussing recovery with the coach. Avoid changing equipment settings, threshold estimates and training volume simultaneously. If several variables move together, the following week becomes harder to interpret. Preserve uncertainty where the evidence cannot distinguish between explanations.
Platforms such as TrainingPeaks, Intervals.icu and GoldenCheetah offer different ways to examine the same work; Strava adds social and segment context. None removes the need for a clear question. Use the tool that exposes the channels and assumptions relevant to that question. The goal is not a universally impressive score. It is a defensible conversation about what happened, how confidently you know it and what to do next.