About your scores

This app turns the data your phone and wearables already collect — sleep, heart rate, and exercise — into four daily numbers that try to answer one question: how is your body doing today, and what should you do with that information?

We try to be honest about what these numbers can and can’t tell you. They are decision aids, not diagnoses. If something feels off in your body, trust your body over the score.

Your profile matters

When you first open the app, you select a Physiological Profile — Athlete, Active, or Sedentary — that tunes how we interpret your data. Different activity profiles show different recovery patterns, so the thresholds we use to score you are customized.

• Athlete — you train regularly (3+ days/week structured exercise). Your metrics will be interpreted with tighter circadian consistency targets (±20 min) and more nuanced HRV sensitivity.
• Active — you exercise 1–3 times per week or have an active job. Moderate thresholds (±30 min circadian, balanced HRV sensitivity).
• Sedentary — you don’t exercise regularly but may have daily movement. More relaxed thresholds (±45 min circadian) so normal life variation doesn’t penalise you.

Why profiles exist. An athlete’s HRV is normally more stable than a sedentary person’s, so a genuine departure means more for an athlete — we don’t compare them to the same bar. Likewise, someone training to peak benefits from tighter circadian consistency targets than someone with a relaxed schedule. Profiles let us keep scores fair across different lifestyles.

A note on measurement

Wearables estimate sleep stages, HRV, and nocturnal physiology indirectly using probabilistic algorithms. These estimates may contain significant measurement error compared to clinical systems like polysomnography or ECG. The scores shown here are fitness-focused estimates, not clinical measurements. Profile settings optimize estimated recovery signals, which are influenced by many factors (stress, sleep environment, caffeine, hydration, illness, etc.). Scores are readiness indicators, not diagnoses.

The four scores at a glance

You’ll see the scores on your dashboard once enough data has been collected. Training Readiness is hidden by default until you enable its card. Until then, we’ll show you what we have and explain what’s missing.

Sleep Score

A 100-point summary of last night’s sleep, made of four parts (under the default Balanced profile):

• Duration (40%) — how much sleep you got, compared to your goal (default 8 hours, configurable). Scoring follows a smooth continuous logistic curve below your goal. A configurable oversleep dead zone (default 125% of goal) prevents penalizing modest sleep-ins while gently decaying scores for excessive hypersomnia. If you sleep in multiple segments (biphasic sleep or naps), all sleep periods count toward your total duration (naps add duration without altering overnight metrics). Includes a continuous adjustment for how efficient your time in bed was.
• Architecture (20%) — how much of your sleep was deep (slow-wave) sleep and REM. Both matter. Deep sleep is when most physical recovery happens; REM is when your brain processes memory and emotion. Targets are age-continuous to account for the natural biological decline in deep sleep across the lifespan (Ohayon 2004).
• Restoration (25%) — how rested your estimated recovery-related physiology looks. We use the natural log of RMSSD (lnRMSSD) and overnight resting heart rate (RHR) to compute Z-scores. The log transformation is the scientific gold standard (Plews 2013, Buchheit 2014) for monitoring recovery, as it normalizes the skewed distribution of raw HRV data.
• Fragmentation (15%) — evaluates sleep continuity using Wake After Sleep Onset (WASO) between sleep onset and final awakening, alongside discrete awakenings of at least 1.5 minutes. Because normal adult sleep includes brief wake periods, a grace allowance of 20 minutes of WASO and 2 awakenings is permitted before continuous exponential penalties apply.

Weight emphasis profiles

You can customize the relative emphasis of each sleep component in Settings to suit your recovery focus (Balanced, Duration Focused, Recovery Focused, Architecture Focused, or Continuity Focused). Changing your weight profile requires running the Recalculate scores action in Settings to apply the new weighting across your history.

Sleep Regularity multiplier

Your sleep score incorporates Sleep Regularity as a penalty-only multiplier between 0.92 and 1.00 derived from your circadian consistency score. Consistent sleep timing preserves 100% of your score (multiplier 1.00), while irregular schedules incur a mild proportional deduction. During baseline calibration or when circadian data is missing, the multiplier remains neutral (1.00).

If a source provides a sleep session with no stage records, Readylytics uses the raw session span as total sleep duration. Because Architecture and Fragmentation are unavailable, the Sleep Score renormalizes the remaining sub-scores (Duration and Restoration) according to your active weight profile. This differs from suspicious but non-empty stage data: the source supplied stages, but their distribution failed plausibility checks.

Reading the score

• 85–100 Excellent. You slept enough, your sleep stages looked balanced, and your autonomic recovery markers were strong.
• 70–84 Good. Most components are healthy; one is slightly below your norm.
• 50–69 Fair. Likely a duration shortfall, fragmented sleep, or an off night for HRV/RHR.
• Below 50 Poor. Multiple components are below typical. One bad night is rarely meaningful; a streak deserves attention.

Deep and REM sleep targets — age matters

As you age, the amount of deep sleep naturally declines. We adjust your targets based on your age so you’re never unfairly compared to a younger person.

These ranges come from polysomnography studies in healthy populations. They represent the healthy mid-range; your personal healthy normal may sit anywhere within your age band. Note that the age-related decline in deep (slow-wave) sleep is much steeper than the decline in REM, which falls only modestly across adulthood (Ohayon 2004). We do not penalise you if your wearable reports unusual numbers — wearable stage detection is imperfect.

Biphasic sleep and naps

We fully support segmented or biphasic sleep. If you sleep in multiple blocks (like a main overnight sleep and an afternoon nap), your “core” sleep drives your recovery metrics (Restoration/HRV/RHR), while all supplemental sleep adds to your total Duration. This means naps will improve your Sleep Score by adding duration, but they won’t rewrite your overnight Readiness signal or baseline HRV.

HRV sensitivity by profile

Heart rate variability (HRV) is noisy on any single night. To avoid false positives, we only flag HRV as notably “high” or “low” once it crosses a Z-score threshold. While your personal baseline is still being learned (the Early Baseline phase), we estimate your day-to-day variability from a population value tiered by profile, so the threshold shifts with it.

• Athlete — HRV is normally stable, so a genuine departure is more likely to mean something (a high reading can even signal parasympathetic hyperactivity during functional overreaching; Le Meur 2013). We flag sooner: Z beyond ±1.2 is notable.
• Active — moderate sensitivity: Z beyond ±1.5 is notable.
• Sedentary — more natural night-to-night noise, so we require a larger departure before flagging to avoid false alarms: Z beyond ±2.0 is notable.

This tuning means a single noisy night isn’t treated as a signal as readily for a Sedentary person as the same Z-score would be for an Athlete, because higher baseline noise produces more large deviations by chance. Once your personal baseline matures (60+ nights), the Z-score is computed against your own standard deviation, which already accounts for your individual variability.

Implemented in: SleepScoringStrategy.kt, SleepArchitectureTargets.kt, ScoringConstants.kt

Restoration/HRV Z-scores implemented in: LoadScoringStrategy.kt, BaselineComputer.kt, HrvBaselineProvider.kt, RhrBaselineProvider.kt

Circadian Consistency

This score asks: do you go to bed and wake up at roughly the same times each day? Schedule regularity is independently linked to better metabolic, cognitive, and cardiovascular outcomes — sometimes more strongly than sleep duration itself (Windred et al. 2023, UK Biobank).

We compare each night’s bedtime and wake time to your typical (median) bedtime and wake time over the last 14 days. The bigger the deviation, the lower the daily score. The number you see is a 7-day rolling average so a single late night doesn’t tank it.

Profile-specific thresholds

How strict we are about “consistent” depends on your profile:

Within each threshold band:

• Within the threshold → full score.
• Threshold to threshold+60 min → score decays linearly.
• Beyond threshold+60 min → score is 0 for that day.

You can override the resolved threshold with your own value in Settings if the profile default doesn’t fit your schedule.

A caveat for biphasic sleepers. This metric is calibrated for people with one main sleep period per day. If you sleep in two segments by choice (e.g., 2:00–4:00 AM and then again at 6:00–7:00 AM), the score may misclassify your schedule. We exclude any single sleep period under 3 hours from the median calculation so naps don’t pull your “typical” times around — but this rule has imperfect coverage of every sleep pattern.

Implemented in: CircadianConsistencyRepository.kt, CircadianThresholdDefaults.kt

Readiness

A daily 0–100 composite number summarising three signals:

Readiness = 0.4 × Restoration (sRest) + 0.3 × Sleep Score + 0.3 × Load Score

Each component is described in its own section above. Restoration carries the largest single weight (0.4), so your overnight recovery markers are the biggest lever on Readiness — but a significant load spike or poor sleep will also pull the score down.

The Load Score (one of the three components) itself is based on your training load ratio:

We compute two rolling averages of your training load:

• Acute load (ATL) — roughly the last week
• Chronic load (CTL) — roughly the last 6 weeks

The ratio (ATL ÷ CTL) tells us whether you’ve recently spiked above your recent norm. Around 1.0 means you’re training in line with your fitness; substantially above 1.0 means a relative spike.

How we score the ratio

• sr ≤ 1.3 → 100 (in your normal range — “sweet spot”)
• sr > 1.3 → 100 × exp(−2.5 × (sr − 1.3)²) — a smooth Gaussian decay that starts gently and accelerates as the spike grows (approximating Gabbett 2016’s elevated-risk zone)

Tooltips

• Peak (85–100) — your recent load is consistent with your fitness.
• Maintain (60–84) — manageable load increase.
• Caution (30–59) — meaningful load spike; consider an easier day.
• High Fatigue (<30) — large spike vs. your norm.

Emergency signals

When HRV is much lower than usual and resting heart rate is elevated for more than one day, readiness may be capped as a cautious possible-illness signal (which caps it at 50). When HRV is much higher than usual and resting heart rate is much lower than usual, the app treats this as an encouraging recovery signal rather than overreaching. Strong recovery signals do not cap Readiness. To ensure accuracy and filter out acute noise (e.g., alcohol or minor stress), the algorithm requires the thresholds to be breached on two consecutive nights (Mishra 2020, Le Meur 2013). This is informational only, not medical advice. Workout-impact and rest-day flags shown elsewhere are informational only and do not cap your Readiness number.

What we don’t do. We don’t penalise you for resting. A week of light activity will not drop Readiness; the score is designed for load spikes, not undertraining.

Implemented in: LoadScoringStrategy.kt, RasScoringStrategy.kt, ComputeSleepMetricsUseCase.kt

Training Readiness

Training Readiness is an optional, default-hidden Dashboard score. It keeps legacy Readiness unchanged while adding Residual Fatigue to the load branch. Residual Fatigue is a raw, unbounded, non-normalized, provisional, workout-only derived input: it feeds Training Readiness when available but does not modify legacy Readiness or Load Score.

AcuteLoadRecovery = clamp(100 * exp(-ResidualFatigue / S), 0, 100)

TrainingLoadReadiness = clamp(w * LoadScore + (1 - w) * AcuteLoadRecovery, 0, 100)

TrainingReadiness = existingReadinessCalculator(Restoration, Sleep, TrainingLoadReadiness, recoveryFlags)

• S default = 100; range 75..175; UI step 5. Increasing S raises Acute Load Recovery for the same Residual Fatigue, so fatigue has less downward influence.
• w default = 0.90; range 0.80..1.00; UI step 0.01 / 1%. Increasing w gives Load Score more influence and Acute Load Recovery less influence.

At w = 1, Training Readiness equals the stored Readiness exactly, and Training Load Readiness preserves the source-matched Load Score.

When Residual Fatigue is unavailable, Acute Load Recovery is unavailable, Training Load Readiness equals Load Score, and Training Readiness equals Readiness.

Training Readiness uses the active Strain / Training Load source for its Load Score branch. Residual Fatigue remains workout-only and never includes everyday heart-rate load.

Edits to S and w remain pending until you choose Recalculate Training Readiness. The durable projection-only recalculation reads stored daily summaries; only success advances the applied values, so current scores stay on the last-applied settings during a run or failure.

The AI Advisor still uses Readiness, not Training Readiness. Training Readiness is excluded from Advisor/AI, Workouts card layouts, trends, insights, details/history, widgets, and standalone exports in this release.

Implemented in: ComputeTrainingReadinessUseCase.kt, TrainingReadinessProjectionRecomputeUseCase.kt, FinalSummaryAssembler.kt

Load Sources

Two independent settings control which heart-rate data feeds your strain/training-load metrics versus your Readylytics Activity Score (RAS):

• Strain / Training Load source (default: Workout only) — controls TRIMP, acute/chronic load (ATL/CTL), Strain Ratio, Load Score, Readiness, and the Load Score branch of Training Readiness. Readiness always uses this source; the RAS source never affects Readiness or Training Readiness.
• RAS source (default: Everyday heart-rate load) — controls your daily and 7-day total RAS only, independent of the Strain / Training Load source above.

Workout only counts heart-rate load from your logged exercise sessions only — the original behaviour.

Everyday heart-rate load also counts elevated heart rate outside workouts (e.g. from stress, illness, or heat) on top of your workout TRIMP. Your workout TRIMP is folded into this total exactly once — it is never double-counted. Sleep is always excluded from the everyday calculation.

For the everyday calculation, every waking, non-sleep, non-workout minute with at least one heart-rate sample is classified into a heart-rate zone using your configured zones/TRIMP settings — the same model used for workouts. Zone 0 minutes (below your Zone 1 threshold) are excluded from TRIMP but still counted toward coverage. Zone 1 and above minutes contribute TRIMP using the standard per-minute formula.

• coverageMinutes — waking, non-sleep, non-workout minutes with ≥1 heart-rate sample (Zone 0 included).
• validBucketCount — the subset of those minutes in Zone 1+ that actually contributed TRIMP.
• Confidence is derived from coverageMinutes: 0 → None, 1–179 → Low, 180–479 → Medium, 480+ → High. A day needs at least 180 coverage minutes to be a valid everyday-load estimate; below that, Readiness shows a low-confidence indicator whenever the Strain / Training Load source is set to Everyday heart-rate load. For the AI Advisor with that source selected, Low coverage caps a high confidence at Medium; None lowers the base confidence by one level, never below Low. The AI Advisor’s base confidence starts at Low during Calibration/Early Baseline, Medium once your baselines are Maturing (or Low if HRV or sleep-stage data is missing today), and High once Mature (or Medium if HRV or sleep-stage data is missing today). Everyday-load coverage can then only lower this, never raise it, per the coverage rule above.

Both source variants are calculated and stored for every day, so switching either setting is instant — no recalculation or history rewrite is needed.

New installs default to Strain / Training Load = Workout only and RAS = Everyday heart-rate load. Existing users upgrading keep their prior behaviour automatically: the first time the app runs after upgrade, the RAS source is set to Workout only as a one-time default if you already have workout history — you can change it in Settings at any time.

Implemented in: EverydayHeartRateLoadCalculator.kt, LoadSourceSelector.kt, LoadSourceMode.kt

BMI and Body Fat

BMI (Body Mass Index) classifies your weight relative to height using WHO-aligned bands:

• Underweight — BMI below 18.5
• Healthy weight — 18.5 to 24.9
• Overweight — 25 to 29.9
• Obesity — 30 and above

The Weight card’s BMI reference gauge uses visual anchors of 15, 21.7, and 35 so its midpoint is shown at 21.7. These anchors position the gauge only; they do not change the BMI status bands above.

Body Fat Percentage uses continuous, gender-specific bands when your profile records a biological sex:

• Male — Below essential <2%, Essential 2–5.9%, Athletic 6–13.9%, Fitness 14–17.9%, Acceptable 18–24.9%, Obese 25%+
• Female — Below essential <10%, Essential 10–13.9%, Athletic 14–20.9%, Fitness 21–24.9%, Acceptable 25–31.9%, Obese 32%+

If gender is set to Other, Prefer not to say, or is unset, we show a fixed 10–30% reference band centered on 20% instead of a gendered scale — values inside the band are Optimal, at or below 10% is Neutral, and above 30% is Poor.

The Body Fat card also shows a reference midpoint — a target value for your physiology profile, used only to position the marker on the gauge, not to change your status:

• Male — Athlete 9.5%, Active 15.5%, Sedentary 19.5%
• Female — Athlete 17%, Active 22.5%, Sedentary 26.5%
• Other / Prefer not to say / unset — fixed at 20%, independent of profile

Reading the status. Optimal (green) and Neutral (informational) describe healthy or expected ranges; Warning flags dangerously low essential-fat levels, or a BMI that is either underweight or overweight — the same status covers both directions, so check the category label (not just the color) to see which one applies; Poor flags the obesity range for BMI or body fat.

The status colors on Weight and Body Fat trend charts use these same canonical bands. Visual reference anchors and profile markers never redefine a reading’s status.

Implemented in: BodyCompositionAssessment.kt, BmiService.kt, HealthMetricsService.kt, HealthMetricsCalculator.kt

Blood Pressure

The dashboard classifies each blood-pressure reading with an inclusive, component-wise ladder:

• Optimal — systolic ≤120 and diastolic ≤80 mmHg
• Neutral — otherwise, systolic ≤129 and diastolic ≤89 mmHg
• Warning — otherwise, systolic ≤139 and diastolic ≤99 mmHg
• Poor — all other readings

For example, 121/80, 120/81, and 129/89 are Neutral, while 130/90 is Warning. This is a dashboard status, not a diagnosis.

Blood-pressure trend charts obtain their component reference bands from this same ladder.

Implemented in: HealthMetricsService.kt

Overnight Oxygen Saturation

The dashboard card uses your overnight average oxygen saturation. Overnight oxygen saturation: below 90% Poor; 90–94% Warning; 95–97% Neutral; 98% and above Optimal.

This means an overnight average of 96% is Neutral. The status describes the displayed overnight average rather than a scoring-engine calculation.

Body Temperature

The dashboard card and Vitals trend chart show your overnight average body temperature (nightly average of Health Connect readings within your sleep-session window). This is an optional metric — it requires the separate READ_BODY_TEMPERATURE Health Connect permission, and the card stays hidden until you grant it.

Once you have 14 nights of readings, we show a rolling 14-day baseline — a plain trailing average of your own recent nights. Before that, the card shows “Calibrating” instead of a deviation badge. We compare today’s reading against that baseline and flag a day as elevated when it deviates by at least your configured elevated-deviation threshold (Settings, default 1.0°C, adjustable from 0.25°C to 1.5°C) in either direction — a wider swing than typical night-to-night noise.

This baseline is intentionally a simple average, not the log-normal statistical model used for HRV/RHR — it exists purely to flag a possibly-illness-related change for you to notice. Body temperature does not affect Sleep Score, Load Score, or Readiness in any way. It is a display-only insight, entirely outside the scoring engine.

Workout GPS and Route Details

When outdoor workouts include GPS location tracks, Readylytics visualizes route contours, pace, and elevation profiles directly inside workout details.

• Privacy-preserving offline Canvas rendering — routes are drawn entirely on-device using native Android Canvas vector graphics. The app never embeds or initializes third-party mapping SDKs (such as Google Maps or Mapbox), downloads no raster or vector map tiles, and executes zero network calls. Location coordinates remain strictly private on the device.
• Douglas-Peucker line simplification — GPS tracks containing hundreds or thousands of raw coordinates are simplified on-device via the Douglas-Peucker algorithm using an adaptive tolerance. This preserves sharp turns, curves, and route shape while keeping rendering lightweight and responsive.
• Pace and elevation performance charts — route waypoints provide distance, altitude, and timestamp metrics used to generate elevation profiles and pace or speed progression charts across the activity.
• Local storage and cascade lifecycle — route coordinates are stored in the local encrypted Room database (workout_route_points table). When a workout session is deleted or purged during historical retention cleanup, all associated route points are immediately cascade-deleted.
• Optional permission & graceful fallback — reading route data requires the optional Health Connect exercise route permission. If route permissions are not granted or route data is unavailable for a session, the app continues to display heart-rate metrics, zone distribution, TRIMP, and recovery analysis normally without route contours.

Implemented in: RouteSimplifier.kt, RouteDistanceCalculator.kt, RouteContourCard.kt, WorkoutPerformanceCharts.kt, WorkoutDetailScreen.kt

Cardio Fitness (VO2 Max)

Cardio Fitness (VO2 Max) is an indicator of your cardiovascular endurance. The app determines your VO2 Max based on your selected source preference (Auto, Wearable only, Estimate only).

If an estimate is allowed, Readylytics can compute it two ways, selectable in Settings:

• Heart-rate-ratio method — the Uth et al. (2004) resting heart rate ratio formula (15.3 * (HRmax / HRrest)).
• Materko-adapted method (experimental) — combines your resting heart rate and HRV baselines. It is an adaptation of a published regression (Materko 2018) from which the CDR term is omitted and pNN50 is approximated from RMSSD; it is predominantly driven by resting heart rate. This adaptation is not equivalent to the published model, and it is not validated across the broad population — the original model was developed in young, healthy, physically active men. Its MeanRR input is derived from your resting-HR baseline rather than a measured beat-to-beat recording, which can introduce a systematic offset; estimates falling outside the supported range are withheld rather than clamped.

Both estimation methods use stable daily baselines and are unavailable until scoring calibration completes.

Your VO2 Max is then benchmarked into five categories (Superior, Excellent, Good, Fair, Poor) based on normative percentiles from the Cooper Institute tailored to your age and sex.

Training Stress Balance (TSB)

Training Stress Balance (TSB) represents your readiness to perform based on your recent training load. It is calculated as TSB = CTL - ATL, where CTL is your Chronic Training Load (42-day average) and ATL is your Acute Training Load (7-day average).

The TSB value places you into one of five zones:

• > +25 (Very Fresh): You are fully recovered, but detraining may occur if prolonged.
• +5 to +25 (Fresh / Peaked): Optimal state for a race or peak performance.
• -10 to +5 (Optimal / Productive): The sweet spot for maintaining and building fitness.
• -30 to -10 (Fatigued / Overload): Heavy training load; you are absorbing training but building fatigue.
• < -30 (High Risk / Overreached): Warning zone; high fatigue and increased injury risk.

You can view your TSB on the Workouts tab (via a toggle) and optionally enable it as a Dashboard card.

Workout Recommendation

An optional, default-visible Dashboard card that turns this morning’s HRV, last night’s sleep, and your recent training fatigue into one qualitative daily signal: Rest, Easy workout, or Push harder. It reuses the same profile-tuned HRV deviation bounds and illness-detection logic your Readiness score already uses (see Emergency signals, above) — no new coefficients, no separate model.

What decides the recommendation

• HRV vs. your own usual range — the same per-profile Z-score bounds described under HRV sensitivity by profile, above (±1.2 Athlete / ±1.5 Active / ±2.0 Sedentary), evaluated against this morning’s reading rather than last night’s frozen baseline.
• Possible illness signals — the same emergency-flag logic described under Readiness caps guidance to Rest.
• Last night’s Sleep Score — a score below 60 caps guidance at Easy, even when HRV looks fine.
• Recent training fatigue — a Residual Fatigue value above 70 × your configured gain (the same Warning threshold your Residual Fatigue card already uses) also caps guidance at Easy.
• Calibration — like your other scores, guidance stays “Calibrating” until at least seven valid nights of data exist.

Missing-data precedence. If we can’t find last night’s sleep at all, or this morning’s HRV reading is missing, we say so directly (“Sleep data missing” / “HRV data missing”) rather than guessing — a missing HRV reading takes priority over any illness signal, since illness detection itself depends on HRV.

Past examples, when guidance is Easy or Push harder. We show up to three of your own past workouts from the preceding 30 days that matched a similar effort level — one per distinct workout type, each longer than 15 minutes, most recent first. Tapping one opens that exact recorded workout. Rest and “no data” days never show examples; if fewer than three workouts qualify, we show fewer.

Existing users. If you already have history, your past days are backfilled with this guidance automatically the next time your local recompute runs — no separate action needed.

Visibility. The card is shown by default and can be hidden from your Dashboard layout at any time, like any other card.

What this is not. There is no generated TRIMP target, workout duration, or heart-rate target attached to this guidance — it is deliberately qualitative, not a number. It is not a “safe,” “optimal,” or validated strength or training prescription, and it is not medical or training advice. It reflects the same recovery-adjacent signals your other scores already use, meant to inform your own judgment, not replace it.

Privacy. Everything above is computed and stored entirely on your device from data you already have locally — no network request is made to produce it. If you use encrypted local backups, your stored recommendations (and the examples they reference) are included like any other computed data. No cloud feature or telemetry was introduced to support this.

Local backup recovery retains verified archives matching the saved password after an interrupted password change. Restore recovery finishes before sync resumes and retains its recovery record until maintenance is released.

Heart-rate history older than 90 days is kept as one-minute summaries rather than individual samples. Each summarised minute now also records which device or app it came from, and that coverage information is included in encrypted local backups so a restore reproduces the same history. Refreshing a source replaces its evidence in these source-backed summaries while preserving other sources; device selection removes discarded sources before rebuilding. Expired summaries, coverage and source evidence are removed together by retention cleanup. Every minute is read from exactly one of the two stores — the raw samples or its summary, never both added together — so a minute can never be counted twice. When a sleep session or workout is later corrected, the summaries it covers are re-derived from that stored per-device record rather than from the previous result, so repeating the correction always lands on the same answer; the only residual difference is that a session boundary falling part-way through a minute can place at most one summarised sample on the wrong side of it. Summaries created before this existed have no per-device record to re-derive from, so they stay marked as approximate and are left exactly as they are — never silently overwritten or presented as corrected. There is currently no way to repair them: “Resync Health Connect data” re-reads your history but does not replace those older summaries, and no other in-app action does either. All of it stays on your device.

Implemented in: ComputeWorkoutRecommendationUseCase.kt, MorningRecommendationAssembler.kt, SelectWorkoutRecommendationExamples.kt

What the app needs from you

We read from Android Health Connect:

• Sleep sessions (with stages if your device records them)
• Heart rate during sleep (for restoration metrics)
• Heart rate variability (RMSSD) during sleep
• Heart rate during exercise sessions (for training load)
• Exercise routes & GPS data (optional, for offline route visualization, pace, and elevation profiles)
• Distance & elevation gained (optional, so a workout shows the same figures as the app that recorded it — re-measuring the GPS track instead reads about 1–3% short)

The app reads only — it never writes. You can revoke access at any time in Health Connect settings.

If a particular metric is missing on a given day, we’ll either:

• show the score with a “data partial” badge and explain which component was estimated, or
• skip the score for that day entirely if too much is missing (especially total sleep time).

Local encryption keys are stored through Android Keystore. On devices that support StrongBox, Readylytics attempts to use StrongBox-backed key protection and falls back to standard Keystore when StrongBox is unavailable. Backup passwords and database keys remain local to the device.

How long until your scores stabilise

Biological baselines take time to learn. We are explicit about the phases:

• Calibration (0–6 valid nights, confidence: Not Ready). We collect baseline data and show raw sleep/recovery measurements where available, but Sleep Score, Load Score, and Readiness stay hidden until there are at least 7 valid nights. This avoids unstable early scores while HRV and RHR baselines are still forming.
• Early Baseline (7–20 nights, confidence: Low). We start showing all three scores, but Restoration uses a population-typical estimate of how much your HRV varies night to night (tiered by your profile). Expect more variability than the mature score.
• Maturing (21–59 nights, confidence: Medium). Your personal HRV mean is settled, and we begin blending your personal night-to-night variability with the population estimate. A confidence indicator becomes visible.
• Mature (60+ nights, confidence: High). All scores use your own personal baselines for both the average and the variability. This is when small day-to-day differences become trustworthy.

Progress through these phases depends on the number of nights with usable HRV/RHR data, not calendar age. If you wear your tracker only 3–5 nights a week, the timeline lengthens proportionally.

Implemented in: Phase.kt, PhaseCalculator.kt

A short glossary

• HRV (Heart Rate Variability) — the millisecond-level variation in time between heartbeats. Higher generally indicates better autonomic recovery, up to a point.
• RMSSD — the specific HRV measure most apps use. We work with the natural log of RMSSD (lnRMSSD) internally because it linearizes the naturally skewed distribution of heart rate variability, making statistical comparison (Z-scores) valid (Plews 2013, Buchheit 2014). The HRV Baseline (ms) shown on your dashboard is the geometric mean of your recent nightly RMSSD — computed as exp(mean(lnRMSSD)) — not a simple average, since a plain average would be skewed upward by occasional high outlier nights.
• Deep sleep / Slow-Wave Sleep / N3 — the deepest stage of NREM sleep; growth-hormone release is concentrated here.
• REM — the dreaming stage, important for memory and emotional processing.

• RHR (Resting Heart Rate) — your true resting heart rate, calculated as a user-defined low percentile (default 5%) of your heart rate samples across the detected sleep period. This sits below your average overnight heart rate, which is pushed up by REM and brief awakenings — using the low percentile gives a more stable night-to-night baseline. Note: This nightly-frozen nocturnal floor is the foundational baseline used directly in all downstream recovery, Heart Rate Reserve (HRR), and TRIMP calculations to ensure training load metrics are highly stable and unaffected by wake-time noise or systemic average inflation.

  Implemented in: SleepPercentileRhrCalculator.kt, BaselineComputer.kt, RhrBaselineProvider.kt

• TRIMP (Training Impulse) — a single number summarising the intensity-weighted duration of an exercise session. Advanced models (such as LT-TRIMP) rely on the specific Heart Rate Zones configured in your app settings. These zones are always active; it is your responsibility to ensure they accurately reflect your current fitness level. On workout screens, overall load is derived primarily from total TRIMP. Intensity is derived from TRIMP/min (load density) and may promote a Moderate workout to Hard when TRIMP >= 90 and TRIMP/min >= 1.75, or a Hard workout to Very Hard when TRIMP >= 140 and TRIMP/min >= 2.25. These categorical labels do not replace numeric TRIMP in ATL, CTL, Strain Ratio, Load Score, RAS, or Readiness calculations.

  Implemented in: RasCalculator.kt, ComputeWorkoutTrimpUseCase.kt, RasScoringStrategy.kt, HeartRateFormulas.kt

• iTRIMP (individualized training impulse). When selected as your training-load model, iTRIMP is a Manzi-inspired fixed-exponent variant: each minute of exercise accrues TRIMP as duration × hrR × exp(b × hrR), where hrR is your fractional heart-rate reserve and b is a single fixed exponent — default 2.1 (adjustable 1.0–4.5 in Advanced Settings). The fixed exponent makes high-intensity minutes scale more steeply than the default Banister model. No sex-specific weighting factor is applied in this variant.

• ATL / CTL — short-term and long-term rolling averages of TRIMP. Borrowed from Banister’s training-load model and used by most cycling and running apps.
• Z-score — a standardized number telling you how many standard deviations above or below your average a metric is. Z=0 is your average; Z=+2 is very high; Z=−2 is very low.

Score adjustments you may not see

A few smaller modifiers shape the numbers behind the scenes. We list them here for transparency:

• HRV-score saturation. Above a Z-score of 1.5, additional HRV improvement contributes less to your Restoration score (a 0.25 slope beyond that point) — so an extraordinarily high reading doesn’t dominate the score the way a moderate one does.
• Late-nadir penalty. If your lowest overnight heart rate occurs in the final third of your sleep period (after 67% of total sleep time has elapsed), we apply a small 0.95 multiplier to the restoration component. A very late RHR nadir often reflects a shortened or fragmented night rather than genuine recovery.
• Banister training-load multiplier. Your Banister training-load model converts heart-rate-reserve intensity into TRIMP using a multiplier of 1.0 for every physiology profile, so TRIMP is a consistent, profile-independent measure of training load — the same effort produces the same load regardless of profile (under the default Banister model). The multiplier remains adjustable in Advanced Settings if you want to scale your personal TRIMP magnitude.
• TRIMP dead zone. Heart rate must exceed your resting-HR baseline by at least 5bpm before any training load (TRIMP) accrues for that minute — below that, the minute contributes zero load rather than a small fractional amount. This avoids counting sensor noise around your baseline as training stress.
• Residual Fatigue. The app calculates an internal, exponential-decay residual fatigue metric from your logged workouts using continuous timestamps and customizable half-life (default 24 h, range 6–96 h) and gain (default 1.0, range 0.1–5.0) parameters. Each day’s value is an end-of-day snapshot evaluated at the following midnight; the Dashboard card shows that snapshot for past days, but for today it instead shows fatigue decayed through the current moment, on the same basis as the Workouts tab decay chart’s “now” point, and refreshes about once a minute while you are looking at it. If a workout logged today has not yet had its training load calculated, the card shows no value rather than a total that would leave that workout out. Residual Fatigue is a raw, unbounded, non-normalized, provisional, workout-only derived input: it feeds Training Readiness when available but does not modify legacy Readiness or Load Score. Users can optionally visualize it by enabling the default-hidden Residual Fatigue card on the Dashboard or the decay curve chart (with selectable 1D, 3D, and 7D time ranges) on the Workouts tab via their respective layout customization sheets. The curve is drawn only up to the present moment, with a dot marking where “now” sits; the remainder of the day is left blank rather than projected. The card’s status colours and gauge scale move with your configured gain, so changing gain re-scales the reading rather than pushing every day to one extreme.
• Readylytics Activity Score (RAS). RAS is a PAI-style motivational activity metric with a daily cap and rolling 7-day accumulation. It is separate from the physiological Load Score: RAS never feeds Readiness, and Readiness/load continue to use TRIMP → ATL → CTL → Strain Ratio → Load Score.
• Suspicious sleep-stage reweight. If your wearable’s sleep-stage data for a night looks implausible (e.g., no deep or REM sleep detected at all), we reweight the Sleep Score: Architecture and Fragmentation drop out and the score renormalizes Duration and Restoration according to your active weight profile. This avoids penalising you for a wearable data glitch rather than your actual sleep.
• Missing-day handling in load averages. Acute and chronic training-load averages (ATL/CTL) are exponential moving averages where a day with no logged exercise counts as zero TRIMP, not “no data”. When you only have one day of history, that single day’s value is used directly as the starting average.
• Estimated max heart rate. If you haven’t entered your own max heart rate, we estimate it from your age using the Tanaka formula (208 − 0.7 × age), rounded down to a whole beat per minute (age 35 gives 183.5, so 183 bpm), which is more accurate across adult age ranges than the older “220 − age” rule of thumb.

Determinism & timezone

Your scores are computed against a stored scoring timezone, so the same underlying health data and settings always produce the same scores, and scores remain consistent if you travel or change your device’s timezone. Recomputing your history (e.g., after a resync) reproduces identical numbers while the underlying heart-rate samples are still stored raw; once that history has moved to the compact per-minute summary (see Honest limitations), recomputation is a measured, close approximation rather than a bit-identical replay.

Honest limitations

1. Wearable stage detection is imperfect. Even premium devices misclassify deep and REM sleep on individual nights. We use age-adjusted targets and treat architecture scores as approximate readiness indicators. Architecture differences on individual nights should not be over-interpreted.

2. Population norms are not destiny. The age-banded deep sleep ranges come from polysomnography studies of healthy adults. Your healthy normal may sit anywhere in your age band. Our scoring uses age-banded targets so the ceiling shifts with you.

3. Profiles are engineering heuristics, not physics. The cutoffs (Athlete ±20 min, Active ±30 min, Sedentary ±45 min circadian threshold) are chosen for practical usability, not derived from prospective studies. We monitor whether these cutoffs are working well and will adjust if needed.

4. The ACWR (Readiness load ratio) is descriptive, not predictive. The methodological literature (Lolli et al. 2019; Impellizzeri et al. 2020, 2021) has demonstrated that the acute-to-chronic ratio is often a mathematical artifact and is not a validated causal injury predictor. We approximate Gabbett’s (2016) elevated-risk zone above the 1.3 sweet-spot ceiling with a Gaussian decay penalty — the curve shape is our own modelling choice — and present it strictly as a load-change indicator to help you visualise spikes in training intensity, not as a diagnostic injury-risk score.

5. One night is noise; trends are signal. Treat any single day’s score as a data point, not a verdict. Look at the 7-day trend.

6. This app does not diagnose anything. If you suspect sleep apnea, a heart condition, an infection, an injury, or any other health concern, see a clinician. Physiological metrics such as HRV, sleep staging, and resting heart rate are non-specific and can be influenced by numerous behavioral, environmental, pharmacological, and measurement-related factors.

7. Heart-rate history older than 90 days is stored as a compact per-minute summary (min/max/average plus a five-point percentile sketch) rather than every raw sample. With the same inputs, settings, and storage tier, a historical replay is repeatable in the stored scoring timezone. Heart-rate history that has moved from raw samples to minute summaries is a measured approximation, so raw-to-minute-summary recomputation can differ slightly; replays are repeatable within the same storage tier. Each minute is read from either the raw samples or its summary, never both, so nothing is double-counted; and when a session is corrected the summaries are re-derived from the stored per-device record. Summaries created before that record existed cannot be re-derived at all — they stay marked approximate and are left untouched. No in-app action repairs them today: “Resync Health Connect data” does not replace them, so that part of your history stays approximate.

8. The Workout Recommendation is not a safe, optimal, or validated training prescription. It is a qualitative Rest/Easy/Push-harder signal derived from the same recovery-adjacent HRV, sleep, and fatigue thresholds used elsewhere in this app — not a strength or endurance program, not a numeric load target, and not medical or training advice. Treat it the same way you would treat any other single-day score: informative, not authoritative.

Selected primary sources informing the scoring: Buysse 1989 (PSQI); Buysse 2014 (RU-SATED); Ohayon et al. 2004, 2017; Hirshkowitz et al. 2015 (NSF); Boulos et al. 2019 (Lancet Respir Med); Lauer et al. 1991; SIESTA database; Plews et al. 2012, 2013, 2014 (HRV monitoring); Buchheit 2014 (Front Physiol); Le Meur et al. 2013 (parasympathetic hyperactivity); Mishra et al. 2020 (Nat Biomed Eng); Quer et al. 2021 (Nat Med); Phillips et al. 2017 (Sleep Regularity Index); Lunsford-Avery et al. 2018; Windred et al. 2023/2024; Khalsa et al. 2003 (phase-response curve); Banister 1991; Foster 1998; Gabbett 2016; Lolli et al. 2019; Impellizzeri et al. 2020/2021.

Scientific and medical disclaimer

This app describes a fitness and training-readiness monitoring framework derived from consumer wearable signals and sports-science literature. The framework is not validated for medical diagnosis, disease screening, treatment guidance, or injury prediction. Profiles and their associated thresholds optimize estimated recovery signals and are engineering heuristics chosen for practical usability, not clinical validation.

If you have concerns about your health, sleep, or recovery, consult a qualified healthcare provider.

Development Transparency

This application was developed with the assistance of AI. While the implementation and scoring logic have been thoroughly reviewed, both AI and human developers can make errors. We encourage community review and welcome issue reports to help improve the application.