How InsightTrail builds its Insights

The relationships InsightTrail draws on are documented across sports science and exercise physiology literature – HRV suppression following consecutive high-load sessions, REM compression under accumulated fatigue, the divergence between subjective readiness and objective markers after travel, illness, or stress, and the lag between training stimulus and measurable adaptation. These are named, studied phenomena, not proprietary claims.

Where research gives a direction, we use it. Where the evidence is more equivocal, the language in the brief reflects that uncertainty rather than overstating confidence. InsightTrail never anchors to population averages – every signal is read against your own history and trend, because that is where the meaningful information lives.

InsightTrail is a coaching tool. It reads the patterns in your data and translates them into plain language relevant to your day or your training. It is not a medical device, a diagnostic tool, or a substitute for professional health advice.

When InsightTrail notes a pattern worth attention – a run of suppressed HRV, a skin temperature that has been creeping up, a Body Battery that is not recovering as it normally would – it is flagging something worth being aware of, not making a clinical determination. The brief is designed to prompt useful self-awareness and better daily decisions, not to replace the judgment of a doctor, physiotherapist, or registered coach.

If something in your data concerns you beyond the scope of training and daily energy management, speak to a qualified professional.

InsightTrail connects to Garmin Connect and, where you choose to link it, Oura – both via their official APIs, with your explicit authorisation. You connect your account, grant the specific permissions required, and can revoke access at any time through your device platform’s connected apps settings.

We read what we need: sleep metrics, HRV, Body Battery, resting heart rate, training load, and activity data. We do not read financial data, contacts, location history, or anything outside the health and activity scope required to generate your coaching brief.

We do not sell your data. We do not share it with third parties. We do not use it for advertising or profiling. Anonymised brief outputs and aggregated feedback scores may be used to improve the coaching layer over time – this is opt-in and presented clearly at onboarding. Raw Garmin or Oura data is never used for this purpose.

You can request deletion of all your data at any time by contacting hello@insighttrail.app. For any other questions about data handling, the same address reaches us.

Heart rate variability is not a single standardised measurement. Different devices capture it differently and the numbers are not directly comparable across platforms.

Garmin measures HRV during a short window in the early morning hours when the body is in its most restful state. The value reflects that window, not continuous overnight tracking.

Oura measures HRV continuously across the full sleep period and derives its figure from the entire night. This produces a different number from Garmin even on the same night from the same body.

When InsightTrail combines both signals, it treats them as complementary perspectives on the same night. Garmin anchors the early-morning autonomic snapshot. Oura extends that picture across the full sleep period, adding continuous HRV, sleep staging depth, and skin temperature deviation. Together they surface things neither device sees alone – a strong Garmin reading alongside a fragmented Oura sleep architecture tells a more complete recovery story than either number in isolation.

InsightTrail’s coaching narrative is generated by a large language model, but the model does not interpret your data. That work is done upstream, in code.

Python logic reads your Garmin and Oura data each morning, computes your baselines, identifies what has shifted and by how much, flags what is significant and what is noise, and structures that into a set of coaching inputs. The language model receives those inputs and translates them into the plain-language brief you read. It generates prose. It does not decide what your HRV means.

The prompts that shape that translation have been refined over months of real daily use and testing across genuine Garmin and Oura data. The coaching voice, the hedging, the things InsightTrail will and will not say, the distinction between a point-in-time observation and a trend statement – all of that is deliberate and tested, not a default model output. Development is ongoing; the coaching layer improves as real-world use surfaces new edge cases and patterns.

Two distinct prompt architectures serve the two modes. Debrief speaks in training terms. Clearday speaks in daily life terms. Neither invents thresholds, neither overstates certainty, and neither uses language the research does not support.