How RunDNA works

We read the last 26 weeks of activities labelled as runs on your Strava account: duration, distance, average pace, heart rate, and — when available — splits, laps, and best efforts. We never see GPS coordinates, photos, comments, or your social graph.

Before we generate anything we check there's enough data to say something meaningful. A report is only produced when all three are true:

• At least 6 weeks of running history
• At least 12 runs total
• At least 3 runs with usable heart-rate data

Below the gate we tell you exactly what's missing rather than guessing.

Pace and heart-rate zones are not pulled from a textbook formula — they come from your own training density across the last 16 weeks. We find the clusters your easy, moderate, and hard work actually live in, then project six zones (Recovery, Easy, Aerobic, Steady, Threshold, VO₂) from those clusters. If your data is sparse or noisy we widen the zones and lower the confidence score rather than fabricate precision.

The limiter engine scores four physiological systems from 0–100: aerobic base, threshold, intensity distribution, and recovery. The lowest-scoring system is your current limiter. Each score is a pure function of your metrics — there is no random component and no AI involved in deciding the limiter.

Your archetype is a decision-tree classification on top of the four metrics. It describes the shape of your training, not a verdict on you as a runner. Two people with the same archetype can be at very different fitness levels.

Every metric returns one of four confidence levels: high, medium, low, or insufficient. Low and insufficient results are excluded from the AI narrative so we never weave a confident story around shaky numbers.

The narrative is written by a language model, but it is on a tight leash. It receives only the structured metrics we computed — never your raw activities — and its output is passed through a numeric-fabrication filter that strips any number not present in the source metrics. We also block claims about VO₂max, race-time predictions, injury, and medical topics.

• It is not a lab test. Wrist HR is noisy and we treat it that way.
• It does not predict race times.
• It is not medical advice. Talk to a clinician about pain or injury.
• It is not a coach. It's a second opinion on the patterns in your data.

• 2026.06.1 — 2026-06-04
  Initial public methodology stamp. Quality gate: ≥6 weeks, ≥12 runs, ≥3 HR runs. Dynamic zones from 16-week density. Goal-aware recommendations.