Health information
Sources & methodology
How Kern turns health and training signals into informational scores and recommendations.
Important context
Kern scores and recommendations are informational estimates, not diagnoses or medical advice. Wearable measurements can be incomplete or inaccurate. Use trends as context, listen to your body, and consult a qualified professional before making medical decisions.
Readiness and recovery: which model applies
The current mobile app defaults to readiness v2. A release can still use the legacy model, or calculate v2 in the background while displaying legacy scores. Accounts selected for the v2 rollout use the v2 result. The methods below distinguish those models and the separate sleep and training metrics.
Readiness is a 0–100 informational score. Neither model has been clinically validated. The weights, score curves, cutoffs, and training adjustments are Kern heuristics, not a formula validated by the sources below. Neither score measures disease, guarantees exercise safety, predicts injury, or establishes biological age. The references provide background on measuring signals and interpreting training load.
Readiness v2: recovery and personal baselines
With all usable signals, recovery combines sleep (50%), HRV (35%), and resting heart rate (15%). The sleep contribution is required. Missing or unusable HRV and resting-heart-rate contributions are excluded and the remaining weights are rebalanced. Sleep alone can therefore produce a provisional score with limited data quality. Without sleep recorded for that day, both recovery and readiness are unavailable; check-in answers or training records alone do not create a score.
The v2 sleep contribution compares recorded sleep minutes with your saved sleep goal, or 480 minutes when no goal is saved. It combines 65% of that day’s duration score with 35% of the mean-duration score from the preceding seven calendar days, provided at least three nights are recorded. Otherwise it uses that day’s duration alone. Duration scores are capped at 100. With at least three recorded wakefulness values in that history, an additional deduction grows linearly to a maximum of 5 points at an average of 60 awake minutes. Missing nights are not counted as zero sleep. This contribution does not use sleep stages or a device’s sleep score.
HRV and resting heart rate use the latest 28 earlier daily records as their candidate history, excluding the current day. Gaps can make this span longer than 28 calendar days. Only readings with the same source, metric type, and observation type as the current reading are comparable. Unknown legacy measurement types are not scored, and RMSSD and SDNN are not pooled. After outlier removal, at least 14 comparable readings are required: 0–6 is unusable, 7–13 provisional, 14–20 usable, and 21 or more established.
HRV values are log-transformed; resting heart rate remains in bpm. Each baseline uses the median and a spread of 1.4826 times the median absolute deviation, with minimum spreads of 0.08 log units for HRV and 2 bpm for resting heart rate. Historical values beyond five spreads are excluded before recalculating the baseline; an equally extreme current reading is also excluded from scoring. A reading at the baseline centre scores 72. HRV 2.5 spreads below baseline, or resting heart rate 2.5 spreads above it, scores 20; the favourable direction reaches a cap of 80 at one spread. Values between these points are interpolated. These rules are product choices, not clinical normal ranges.
Readiness v2: check-in and training load
When a morning check-in is available, the base readiness score is 80% recovery plus 20% check-in. Check-in uses energy (35%), fatigue (45%), and stress (20%) on 1–5 scales, mapped to 0–100 with fatigue and stress reversed. Without a check-in, the base score equals recovery. Feeling unwell and affected body areas can restrict training separately from the number; feeling unwell marked “yes” requests rest. The numerical guidance thresholds are below 35 for rest and below 60 for reduced intensity. These are not medical decision rules.
Recorded training load can then subtract up to 10 points; it never raises the score. The current app uses saved endurance session-RPE outcomes linked to identified completed activities. RPE is the reported effort on a 0–10 scale; internal load is recorded elapsed minutes multiplied by RPE. Only valid positive recorded internal loads are included, and duplicate activity identities count once. Imported sessions without a usable outcome and strength workouts are not automatically assigned a v2 load.
Recent load is the sum of eligible loads in the last seven days, exponentially reduced with a 36-hour half-life. The personal reference is the total eligible load from more than seven to 28 days earlier divided by three weeks, requiring at least three sessions. The app evaluates each date at the end of that UTC day using the activity’s recorded start time. The deduction starts above 15% excess over that reference, increases linearly, and reaches 10 points at 50% excess. Final scores are rounded and bounded to 0–100. This comparison and its time-decay rules are experimental heuristics, not a validated injury-risk ratio.
Missing recent sessions, an insufficient load history, or a missing load reference means no load deduction and an unavailable-load flag. It does not establish that you are fully recovered. Pending data or an unavailable selected model result does not silently switch to another model. Gaps in v2 history remain gaps. Limited or partial data should reduce confidence in the result, including when the displayed number is high.
Legacy readiness and separate training recovery
The legacy physiological score combines available sleep (50%), HRV (30%), and resting heart rate (20%), rebalancing weights when signals are missing. It uses a stored sleep score when available, otherwise the separate sleep-score calculation below. If that day’s sleep is missing but at least two sleep records exist, it can use the mean of up to seven recent recorded sleep scores. In the current mobile Today and Health views, at least one recovery signal must be recorded for the selected day; history alone does not produce today’s legacy score.
Legacy HRV and resting heart rate use the arithmetic mean of up to 30 previous available readings once at least four exist. With less history, they use generic fallback curves: HRV / (HRV + 50), and (100 − resting heart rate) / 50 respectively. With a baseline, the curves start at 0.72 and add relative HRV deviation / 0.6, or subtract relative resting-heart-rate deviation / 0.3. These contributions are bounded to 0–1 before weighting. This older method does not enforce v2’s matching measurement streams or robust outlier checks.
Current mobile Today, Health, and Training readiness displays use the selected model’s daily score. Under legacy selection, this is the physiological score, without the v2 check-in or load deduction. A separate legacy coaching calculation still combines 70% physiology with 30% estimated recovery from recent training, or training recovery alone when physiology is unavailable. That coaching value is not the current mobile Today readiness number. Older app or web views may use this combined method. Training recovery is itself a heuristic based on recorded activity and time since exercise, not measured recovery.
Separate sleep score
Sleep cards use an available stored or corrected sleep score, otherwise Kern calculates one from the recorded sleep data. A score supplied by a device can use that provider’s own method. Kern’s calculation uses sleep duration, estimated sleep-stage composition, sleep efficiency, and recorded awakenings when available. Duration is compared with a target of 465 minutes (7 hours 45 minutes). This is a different calculation from the v2 sleep contribution above. Sleep stages reported by consumer devices are estimates and are not equivalent to a clinical sleep study.
With stage data, weights are 50% duration, 35% deep-plus-REM composition and 15% efficiency. Recorded awakening counts change these to 45%, 30%, 10% and 15% restlessness. Without stages, duration/efficiency use 80%/20%, or 70%/15% plus 15% restlessness when awakenings are available. The 465-minute target, 45% deep-plus-REM reference and these weights are app heuristics, not individual sleep requirements or clinical cutoffs.
Training guidance
Training suggestions combine the goals and availability you enter with recent activity, planned sessions, and available recovery signals. Kern reduces suggested load when recovery signals are low. Recommendations are starting points and should be adjusted for pain, illness, injury, experience, and professional guidance.
The Training metrics view’s weekly Training load points are a separate activity-volume heuristic: rounded cardio moving minutes plus 10 points per recorded strength set, with a minimum of 10 points per strength workout. This card does not use heart rate or RPE and is not the v2 load deduction. Its Recovery trend is the weekly average of available sleep scores, not the full daily readiness calculation. Consistency counts recorded cardio activities and strength workouts.
Running oxygen cost — not VO₂max
Estimated running oxygen cost uses 3.5 + 0.2 × running speed in metres per minute, expressed in ml/kg/min. Kern displays it only for running speeds of at least 134 m/min. This level-ground equation assumes steady running and does not account for hills, terrain, wind, running economy, or individual physiology. Session-average speed does not establish that these assumptions hold.
This is not measured oxygen consumption, a VO₂max test, or a fitness score. A faster pace produces a higher estimate by definition; a higher value does not demonstrate improved aerobic fitness. Consumer-device VO₂max estimates, when supplied separately, are not laboratory measurements either.
Energy and macronutrient estimates
Estimated resting energy expenditure uses the Mifflin–St Jeor equation and is multiplied by the activity level selected in Kern. Goal rate changes the daily estimate using 7,700 kcal per kilogram. Protein targets range from 1.8 to 2.2 g/kg depending on the selected goal; remaining energy is divided between fat and carbohydrate. These are estimates and are not suitable for pregnancy, eating-disorder treatment, clinical nutrition, or medication decisions.
The 7,700 kcal rule is a simplified adjustment, not a prediction of weight change. Kern rejects calculated changes exceeding 25% of estimated daily energy, loss rates above 0.9 kg/week and gain rates above 0.5 kg/week. Maintenance requires a zero rate. Calculated and manual targets must be within 1,500–6,000 kcal/day, have positive protein and fat, nonnegative carbohydrate, and matching macro energy within a 20 kcal rounding tolerance. Unsupported values are rejected, not silently clamped.
The calculator supports ages 18–100, weight 30–300 kg, height 120–230 cm, activity multipliers of 1.2–1.9, and BMI of at least 18.5; a goal weight must also stay above that BMI threshold. Before previewing or saving targets, users confirm that pregnancy, breastfeeding, eating disorders, underweight, and medical-diet needs do not apply. These are product support limits, not proof that a target is safe or appropriate for an individual. Ask a registered dietitian for individual advice.
Questions or corrections
For questions about these methods or to report a misleading recommendation, email support@joinkern.com.