https://tryterra.co/research/can-wearables-detect-pregnancy2026-09-08T09:44:17.802Zweekly0.9https://loved-hug-5d84552294.media.strapiapp.com/1200x630_21_4e819fd022.pngPregnancy Shows Up in Wearable Data Before a Missed Periodhttps://loved-hug-5d84552294.media.strapiapp.com/smoothed_temp_breathing_trimester_11fe7b184b.pngFigure 1: Temperature, heart rate and breathing rate trajectories during pregnancy (blue) compared to a control non-pregnancy trajectory (grey). Empirical hormone levels (in pink) plotted to show similarity to physiological biomarkers.Pregnancy detection in wearables: temperature, heart rate, and breathing rate trajectories during pregnancy versus non-pregnancy controls, tracking hormone levelshttps://loved-hug-5d84552294.media.strapiapp.com/average_acf_pre_vs_post_0_60_5d2381c92f.pngFigure 2: Autocorrelation profiles for physiological biomarkers (temperature, heart rate and breathing rate) before pregnancy in dark pink, during pregnancy in light pink, and during a pregnancy with early termination in blue.Pregnancy monthly rhythm signature: autocorrelation profiles for temperature, heart rate, and breathing rate before pregnancy, during pregnancy, and during pregnancy with symptomshttps://loved-hug-5d84552294.media.strapiapp.com/perf_vs_horizon_d5b8f160c5.pngFigure 3: Performance for pregnancy detection against window size.Early pregnancy detection performance from wearable data: model accuracy for detecting pregnancy plotted against how many days after conception the window opens https://tryterra.co/research/why-does-altitude-affect-people-differently2026-09-08T09:44:17.369Zweekly0.9https://loved-hug-5d84552294.media.strapiapp.com/1200x630_20_2ffb748790.pngYour Body's Response to Altitude Is Highly Individualhttps://loved-hug-5d84552294.media.strapiapp.com/09_day0_spo2_vs_normal_2_7e3adf8b31.pngChart 1: Distribution of Day-0 SpO₂ Δ vs home.Altitude SpO2 individual response: distribution of arrival night SpO2 change from home baseline showing a long left tail of the most-affected travellershttps://loved-hug-5d84552294.media.strapiapp.com/10_rtm_tertile_slopes_1_dafdaf241c.pngChart 2: Mean day-0 SpO₂ and min HR by tail.Altitude sensitivity extremes: mean arrival night SpO2 and minimum heart rate for the most and least affected travellers, split by trip elevationhttps://loved-hug-5d84552294.media.strapiapp.com/11_tails_compare_1_895d0eb996.pngChart 3: Day 0→5 SpO₂ by arrival tertile.Altitude recovery by arrival severity: five day SpO2 trajectory split by arrival night drop, showing the biggest-hit third also recovers the fastesthttps://loved-hug-5d84552294.media.strapiapp.com/12_raw_vs_elev_residual_hist_2_2fec20e640.pngChart 4: Raw vs elevation-residual SpO₂.Altitude versus individual response: raw SpO2 drop by elevation (about -2.6 pp per kilometre) alongside the elevation adjusted residual distributionhttps://loved-hug-5d84552294.media.strapiapp.com/13_elev_adj_tertile_slopes_3_66b877c3d5.pngChart 5: Height-adjusted residual tertiles, day 0→5. Similar to Chart 3, but with altitude accounted for.Height adjusted altitude sensitivity: five day recovery for the sensitive versus resilient third at matched 1800 to 2000m elevations, showing persistent individual differenceshttps://loved-hug-5d84552294.media.strapiapp.com/14_both_hit_by_residual_decile_1_dc9fde1f84.pngChart 6: Dual-hit rate by elevation-residual SpO₂ decile.Altitude dual hit rate by residual SpO2 decile: share of travellers who show both low SpO2 and elevated heart rate rises smoothly with sensitivity, arguing for a continuum not clusters https://tryterra.co/research/how-do-night-shifts-affect-sleep2026-09-08T09:44:16.877Zweekly0.9https://loved-hug-5d84552294.media.strapiapp.com/1200x630_18_712e5d6cda.pngIt Takes Four Days to Recover From a Night Shifthttps://loved-hug-5d84552294.media.strapiapp.com/shift_user_scatter_1_f8928c45f7.pngFigure 1: Scatter plot of sleep duration and timing shift across day-time and night-time work shift users.Night shift sleep duration analysis of 17,775 shift nights: bedtime shift plotted against hours slept, showing every extra hour of bedtime delay costs 12.7 minutes of sleephttps://loved-hug-5d84552294.media.strapiapp.com/forest_fig2_fixed_ab1191c412.pngFigure 2: Forest chart of effects of shifts compared to user baseline.Night shift REM sleep effects versus personal baseline, showing REM falls 16 minutes on shift nights while deep sleep stays essentially unchangedhttps://loved-hug-5d84552294.media.strapiapp.com/shift_blog_s3_readiness_1_55cc63e7df.pngFigure 3: Trajectory of sleep health metrics aggregate from 3 days prior to shift and up to 7 days after.How long it takes to recover from a night shift: sleep health score from three days before to seven days after a shift, showing recovery to baseline around day fourhttps://loved-hug-5d84552294.media.strapiapp.com/shift_consecutive_nights_dea4d3ae4f.pngFigure 4: Bar chart of sleep duration and REM sleep time through consecutive shift nights. Sleep quality drops on shift days, and recovers when the pattern is consistent enough.Effects of consecutive night shifts on sleep duration and REM time, showing worst disruption on the second shift and adaptation from the third shift onward https://tryterra.co/research/what-is-the-greenest-way-to-commute2026-09-08T09:44:16.454Zweekly0.9https://loved-hug-5d84552294.media.strapiapp.com/1200x630_17_25ad485e0a.pngFood is Dirtier Than the Motorhttps://loved-hug-5d84552294.media.strapiapp.com/carbon_01_active_efficiency_521618ce85.pngFigure 1: Active efficiency by mode. Metabolic cost included for cycling; DESNZ 2026 for PT and car.Active commute carbon efficiency by mode across 10 London corridors: metabolic cost for cycling and DESNZ 2026 factors for public transport and carhttps://loved-hug-5d84552294.media.strapiapp.com/carbon_03_active_metabolic_stack_86558e09cb.pngFigure 2: Active cost breakdown. Human metabolic effort vs e-bike metabolic + grid.Active commute carbon breakdown: human metabolic cost for cycling (about 15 gCO2/km) versus e-bike metabolic plus grid cost (about 3.7 plus 1.5 gCO2/km)https://loved-hug-5d84552294.media.strapiapp.com/carbon_02_operational_efficiency_1c6eb12efc.pngFigure 3: Operational efficiency. Use energy plus amortised manufacture (no metabolic, no food).Operational commute carbon per kilometre: e-bike grid at 1.5g, public transport blend at 29.9g, and car tank to wheel at 165.9g using DESNZ 2026 factors https://tryterra.co/research/can-wearables-detect-menopause2026-09-08T09:44:16.023Zweekly0.9https://loved-hug-5d84552294.media.strapiapp.com/1200x630_15_7aff84f7d1.pngMenopause Shows Up in Your HRV Before It Shows Up on the Calendarhttps://loved-hug-5d84552294.media.strapiapp.com/fig1_hrv_groups_93233aaf29.pngFigure 1: HRV distribution as a violin plot (left) before menopause in green, during perimenopause in pink and menopause in purple. HRV autocorrelation on the right. HRV is suppressed during menopause and loses its cyclical pattern when periods permanently stop.Menopause HRV signature in wearable data: night-time HRV distribution before menopause, during perimenopause, and after menopause showing an 11ms drop with menopausehttps://loved-hug-5d84552294.media.strapiapp.com/fig_sleep_groups_1_45a9cdb111.pngFigure 2: Deep sleep and wakeup events violin plot distribution before, during perimenopause and after menopause. Deep sleep duration drops and wakeup events increase with menopause.Menopause sleep changes in wearable data: deep sleep duration drops and wakeup events increase from pre-menopause through perimenopause to menopausehttps://loved-hug-5d84552294.media.strapiapp.com/fig7_cycle_trajectories_493e8099e8.pngFigure 3: Menstrual cycle sinusoidal model. Top panel shows a normal cyclic control user, middle panel is a user with perimenopause with a still cyclical pattern, and bottom panel shows high disruption during menopause.Detecting menopause from wearables: menstrual cycle sinusoidal model showing normal cyclic pattern, still cyclical perimenopause pattern, and disrupted menopause pattern https://tryterra.co/research/how-does-altitude-affect-recovery2026-09-08T09:44:15.632Zweekly0.9https://loved-hug-5d84552294.media.strapiapp.com/1200x630_14_622cb3f28b.pngYour HRV Recovers at Altitude Before Your Oxygen Doeshttps://loved-hug-5d84552294.media.strapiapp.com/01_day_curve_spo2_1_0918addc7a.pngChart 1: Within-person change in overnight SpO₂ from each traveller’s lowland home baseline, by trip elevation band (sustained cohort). Day 0 is the first altitude night with same-day high activity; ribbons are 95% CIs.Altitude SpO2 recovery curve: within-person change in overnight blood oxygen from lowland baseline across trip elevation bands from day 0 through day 5 at altitudehttps://loved-hug-5d84552294.media.strapiapp.com/01_day_curve_hrv_rmssd_a272239146.pngChart 2: Within-person change in overnight HRV RMSSD from each traveler’s lowland home baseline, by trip elevation band. Day 0 is the first altitude night with same-day high activity; ribbons are 95% CIs.Altitude HRV recovery curve: within-person change in overnight HRV RMSSD from lowland baseline by trip elevation band showing HRV rebounds while SpO2 stays lowhttps://loved-hug-5d84552294.media.strapiapp.com/04_day0_by_elevation_band_5ac254522f.pngChart 3: Mean arrival-night change from home baseline for SpO₂, resting HR, min HR, and HRV across elevation bands in the sustained cohort; error bars are 95% CIs.Altitude arrival night physiology by elevation band: change from home baseline for SpO2, resting HR, min HR, and HRV across sustained tripshttps://loved-hug-5d84552294.media.strapiapp.com/07_spo2_ridgeline_by_day_5877bf2c93.pngChart 4: Ridgeline distributions of within-person Δ SpO₂ for days 1–5 among sustained trips ≥2,000 m. Each ridge is a day’s distribution; the vertical tick is the day mean and the dashed line is home baseline.Altitude SpO2 recovery distributions by day: ridgeline plot of within-person SpO2 change from days 1 to 5 among trips at 2,000m or higher https://tryterra.co/research/how-to-handle-missing-wearable-data2026-09-08T09:44:15.151Zweekly0.9https://loved-hug-5d84552294.media.strapiapp.com/1200x630_13_7bf92a56e0.pngTerra Smart Fill: Solving Missing Health Datahttps://loved-hug-5d84552294.media.strapiapp.com/fig1_what_devices_report_1_23e20101e0.pngCoverage of health data streams reported by each major wearable device.What each wearable device reports: coverage map of which health data streams (steps, HR, HRV, sleep, temperature) each major wearable actually provideshttps://loved-hug-5d84552294.media.strapiapp.com/handbook_ranges_c8add2b000.pngTypical missing-data gaps across wearable health data streams.Missing wearable data ranges: overview of typical gaps in health data streams that Terra Smart Fill is designed to reconstructhttps://loved-hug-5d84552294.media.strapiapp.com/handbook_smartfill_6f09c70b19.pngSmart Fill reconstruction error versus per-user average baseline.Terra Smart Fill accuracy: reconstruction error for missing wearable data compared to using each user’s own average as a baselinehttps://loved-hug-5d84552294.media.strapiapp.com/handbook_devices_6bae3200cb.pngMissing-data prevalence per wearable brand and metric.Missing wearable data by device: which wearable brands leave the biggest gaps in each health data streamhttps://loved-hug-5d84552294.media.strapiapp.com/handbook_history_4e5b03f8ed.pngHistorical data depth available by wearable device.Wearable data history depth by device: how much historical data each major wearable makes available for backfill and imputation https://tryterra.co/research/Alistair-Brownlee-Norseman-power-data2026-09-08T09:44:14.731Zweekly0.9https://loved-hug-5d84552294.media.strapiapp.com/1200x630_12_fcb475ffaa.pngHow I came second in The World's Hardest Triathlonhttps://loved-hug-5d84552294.media.strapiapp.com/01_bike_power_zones_1_8cebce40cd.pngFigure 1: Time in 25 W zones. How long I spent in each 25 W bin. The band 275–349 W is most of the ride; quite a bit of time freewheeling down descents (0–24 W); almost nothing above 375 W.Norseman triathlon bike power distribution: time spent in 25W bins across Alistair Brownlee’s ride, with most time in the 275 to 349W bandhttps://loved-hug-5d84552294.media.strapiapp.com/02_climb_power_6e5bfb32dc.pngFigure 2: Climb power (first vs later) Average watts on sustained bike climbs. Left bars are the opening fjord climb (1a/1b); right bars are later climbs. Dashed line marks my first-climb level (~343 W). Later similar grades sit ~286–307 W, about 10–15% lower.Norseman bike climb power comparison: average watts on the opening fjord climb versus later climbs, showing the fast early pace dropped by roughly 100W later onhttps://loved-hug-5d84552294.media.strapiapp.com/03_run_climb_vam_87e8d77697.pngFigure 3: Run-climb VAM: meters climbed per hour on a rolling 5-minute window from ~km 25 to the summit. Dashed line ≈ climb average (~722 m/h). Shaded area = off-road summit block (~km 37+). The big dip from 34- 37 km was a flatter section.Norseman marathon Zombie Hill climb rate: metres climbed per hour on a rolling 5-minute window from km 25 to the summit, averaging about 722 m/h https://tryterra.co/research/bike-vs-tube-commuting-speed2026-09-08T09:44:14.299Zweekly0.9https://loved-hug-5d84552294.media.strapiapp.com/1200x630_10_a843225d3a.pngCycling Is Good for You, But is it Faster Than Public Transport?https://loved-hug-5d84552294.media.strapiapp.com/02_delta_histograms_37dd2ca21c.svgFigure 1: Trip-level time difference for the same morning origin–destination: bike duration minus the modeled alternative. Negative minutes mean the recorded bike ride was faster; green shading marks bike wins.Bike versus tube commute time in London: trip-level time difference between bike and alternative for the same origin destination, with negative minutes meaning bike is fasterhttps://loved-hug-5d84552294.media.strapiapp.com/08_win_rates_by_length_1_a9bfdbf6ce.svgFigure 2: Share of matched morning trips where the recorded bike ride beats TfL door-to-door or Google traffic driving. Short <6 km, medium 6–15 km, long 15–25 km, win rates stay high across lengths, with traffic wins peaking mid-distance.Bike win rate versus TfL and driving by London commute distance: short, medium, and long trips all favour the bike but the losing opponent shifts with distancehttps://loved-hug-5d84552294.media.strapiapp.com/09_median_savings_by_length_ebb8449226.svgFigure 3: Median minutes saved on the same origin–destination when cycling versus TfL or Google rush-hour driving. This is a within-trip delta, not a comparison of different absolute journey lengths across modes.Median time saved cycling versus TfL and Google rush hour driving on the same London morning commute broken down by trip lengthhttps://loved-hug-5d84552294.media.strapiapp.com/05_top_corridors_map_ef616d755e.svgFigure 4: Busiest multi-user morning bike corridors over a Greater London silhouette with the Thames. Blue routes beat Google rush-hour driving; orange means traffic still wins.Top London bike commute corridors mapped, with blue routes beating rush hour driving and orange routes where driving still winshttps://loved-hug-5d84552294.media.strapiapp.com/06_top_corridors_savings_cc5656aaa0.svgFigure 5: Each corridor has two bars: minutes saved versus public transport and versus Google rush-hour driving. Positive values mean the bike is faster; Ilford → Barking is the main case where Google driving still wins on time.Bike commute time savings on the busiest London corridors versus public transport and versus Google rush hour driving, with Ilford to Barking as the main case driving wins https://tryterra.co/research/how-spain-celebrated-winning-the-world-cup2026-09-08T09:44:13.961Zweekly0.9https://loved-hug-5d84552294.media.strapiapp.com/1200x630_6_65d5637d5a.pngWhat Winning the World Cup Does to Your Healthhttps://loved-hug-5d84552294.media.strapiapp.com/hero_hr_fc58726589.pngFigure 1: Average heart rate of Spaniards on a usual Sunday (in grey) and watching the world cup final (in red).Spain World Cup final heart rate: average heart rate of Spaniards on a normal Sunday versus final night showing elevated HR throughout the match windowhttps://loved-hug-5d84552294.media.strapiapp.com/celeb_curve_d36e0255e8.pngFigure 2: Average steps throughout a normal Sunday evening (in grey) compared to the world cup final (in red).Spain World Cup celebration steps: average steps through Sunday evening on a normal week versus final night, showing a large step spike around 1am after the matchhttps://loved-hug-5d84552294.media.strapiapp.com/city_party_6095a909c1.pngFigure 3: Spanish cities ranked by percentage of users who celebrated the world cup win after the match.Spanish cities ranked by post-match World Cup celebration participation, with Madrid leading, followed by Málaga, Barcelona, and San Sebastianhttps://loved-hug-5d84552294.media.strapiapp.com/map_concentration_68bc564580.pngFigure 4: World map of countries, color coded by percentage of viewers.World Cup final viewer distribution by country from wearable data, with Spain leading in share of users watching and Argentina second https://tryterra.co/research/does-running-in-heat-affect-your-heart-rate2026-09-08T09:44:13.592Zweekly0.9https://loved-hug-5d84552294.media.strapiapp.com/1200x630_5_0d774cff96.pngRunning in 30°C Heat Barely Raises Your Heart Ratehttps://loved-hug-5d84552294.media.strapiapp.com/Matched_pace_HR_slope_by_1_min_band_e4e4de5435.pngFigure 1: Within-person heart-rate response to WBGT when pace is matched.Running heart rate response to heat at matched pace: recreational European runners show near-zero HR increase per degree WBGT across training paceshttps://loved-hug-5d84552294.media.strapiapp.com/03_spearman_pace_hr_tradeoff_1_bd0b49e52f.pngFigure 2: Spearman pace–HR trade-off. Each runner's pace slope against their HR slope: less slowing in heat, leads to more HR rise.Heat running trade off across 1,996 runners: pace slope plotted against HR slope, showing runners who slow less in heat tend to see bigger HR riseshttps://loved-hug-5d84552294.media.strapiapp.com/Screenshot_2026_07_21_at_12_27_01_ec8b9da8a1.pngSummary of running heart rate response to WBGT across the sample.Running heart rate response to heat summary showing minimal HR increase for recreational runners across the full temperature range analysed https://tryterra.co/research/do-heatwaves-make-you-sleep-less2026-09-08T09:44:13.173Zweekly0.9https://loved-hug-5d84552294.media.strapiapp.com/1200x630_4_68b5830cc6.pngHeatwaves and Sleephttps://loved-hug-5d84552294.media.strapiapp.com/01_regional_sleep_vs_temperature_64bcbdc022.pngFigure 1: Regional time series of mean sleep (blue line), night minimum temperature (red line), and WBGT across seven European regions. Sleep troughs track temperature peaks through the heatwave window (yellow band), with the largest co-movement in France, Benelux/DACH, and the UK; Scandinavia shows a flatter response.Heatwave sleep effect across Europe: regional time series of nightly sleep, night minimum temperature, and WBGT showing sleep troughs track temperature peakshttps://loved-hug-5d84552294.media.strapiapp.com/02_dose_response_sleep_vs_night_min_temp_30212acfed.pngFigure 2: Population decile dose–response for sleep duration, deep sleep, and awake-in-bed time vs night minimum temperature. Total and deep sleep fall as nights get warmer; awake-in-bed time is essentially flat, heat shortens sleep, not by keeping people awake longer.Sleep loss dose response to night temperature: total sleep and deep sleep fall steadily across temperature deciles with no visible threshold effecthttps://loved-hug-5d84552294.media.strapiapp.com/03_within_person_hot_minus_cool_d00dfd2d03.pngFigure 3: Within-person histograms of hot-minus-cool quartile night deltas for sleep, deep sleep, and awake-in-bed time. Most users lose sleep on their hottest nights (median −11 min); deep sleep falls too; awake-in-bed change clusters around zero.Within-person heatwave sleep loss: most users lose about 11 minutes of sleep on their hottest nights compared to their coolest, with deep sleep falling more sharply https://tryterra.co/research/how-much-sleep-did-the-uk-lose-england-vs-mexico2026-09-08T09:44:12.728Zweekly0.9https://loved-hug-5d84552294.media.strapiapp.com/1200x630_3_f3e7806491.pngThe England vs Mexico Monday Morning Hangoverhttps://loved-hug-5d84552294.media.strapiapp.com/IMG_7316_1_f34a3c6176.pngFigure 1: Bedtime hour bands. Grouped bar chart of the % of UK wearable users whose sleep onset fell in each hour from midnight to 6am. Blue bars = average of prior Mondays; orange bars = Monday 6 July (sleep on Sunday night).England vs Mexico Monday morning hangover: percentage of UK wearable users falling asleep in each hour from midnight to 6am on Monday 6 July versus prior Mondayshttps://loved-hug-5d84552294.media.strapiapp.com/IMG_7317_a3ec9cef49.pngFigure 2: Sleep and recovery deltas, horizontal bar chart of within-user changes (Mon 6 Jul minus Mon 29 Jun, n = 12,777).Sleep and recovery changes after England vs Mexico match night: within-user deltas for bedtime, sleep duration, and HRV among 12,777 UK usershttps://loved-hug-5d84552294.media.strapiapp.com/IMG_7318_7fba2f8219.pngFigure 3: Short sleep and national scale: Two-panel chart. Left: grouped bars comparing % of users sleeping under 6 hours and under 5 hours on Mon 6 Jul vs prior Mondays (27% vs 19%, and 15% vs 9%).Short sleep spike after England vs Mexico: 27% of UK users slept under 6 hours versus 19% on prior Mondays, and 15% slept under 5 hours versus 9% https://tryterra.co/research/terra-research-nature-publication2026-09-08T09:44:12.341Zweekly0.9https://loved-hug-5d84552294.media.strapiapp.com/1200x630_2_56fc4009a4.pngTerra Research Published in Nature Portfoliohttps://loved-hug-5d84552294.media.strapiapp.com/day_length_mean_asleep_grid_b48054f7c6.png(a) The effect of day length on sleep time across the entire dataset. The absence of a clear trend here is expected, given the deeply hierarchical structure of the data. Estimated trends emerge once countries with significant offsets are examined individually. (b) Countries with positive b_k values, where trends begin to appear. (c) Countries whose country-specific offset runs in the same direction as the global trend, that is, a negative b_k value, where the trends are clearer still. In both (b) and (c), countries are listed in the legend in order of effect size.Sleep and daylight relationship from 185,143 nights: people sleep roughly 4.4 minutes less for every extra hour of daylight, once individual and seasonal structure is accounted for https://tryterra.co/research/sauna-and-cold-plunge-vs-sauna-alone2026-09-08T09:44:11.987Zweekly0.9https://loved-hug-5d84552294.media.strapiapp.com/1200x630_template_10_fc4295a13c.pngStacking Sauna and Cold Plunge, or? https://loved-hug-5d84552294.media.strapiapp.com/1_d90433adf6.pngFigure 1: Overnight HRV change versus a normal day. Only the sauna reached significance. Adding a cold plunge, alone or stacked, did not.Sauna versus cold plunge overnight HRV effect: only sauna alone significantly raised HRV, with adding a cold plunge alone or stacked showing no effecthttps://loved-hug-5d84552294.media.strapiapp.com/2_cc07a3afc4.pngFigure 2: Effect of each protocol on six overnight measures, with color showing significance. Sauna is significant on all six. Stacking works on heart rate but not HRV, and cold lands once in six. n = 31,424 days.Sauna and cold plunge effects on six overnight recovery measures: sauna is significant on all six while stacking works on heart rate but not HRV https://tryterra.co/research/does-heat-make-you-run-slower2026-09-08T09:44:11.654Zweekly0.9https://loved-hug-5d84552294.media.strapiapp.com/1200x630_template_9_ab26b5a558.pngDoes Heat Make You Slow Down, or Tell You to Slow Down?https://loved-hug-5d84552294.media.strapiapp.com/london_physiology_running_ridgeline_pace_temp_1_fff1cd1b0d.pngRidgeline distribution of running pace across temperature bins.Does heat make you run slower: ridgeline distributions of running pace across temperature bins showing the population pace shift as temperature riseshttps://loved-hug-5d84552294.media.strapiapp.com/london_physiology_running_ridgeline_distance_temp_b7c41d78cc.pngRidgeline distribution of run distance across temperature bins.Heat effect on run distance: ridgeline distributions of run distance across temperature bins showing shorter runs on hotter dayshttps://loved-hug-5d84552294.media.strapiapp.com/london_physiology_running_ridgeline_hr_temp_2869f1b793.pngFigures 1-3: Ridgeline distributions to visualize how the user population changes their Pace, Distance and Average Hr of runs in different temperature bins.Does heat make runners slower: ridgeline distributions of average heart rate across temperature bins from ~3,500 tracked runshttps://loved-hug-5d84552294.media.strapiapp.com/london_physiology_running_lmm_pace_temp_2_66f6cc8c9b.pngWithin-person pace-vs-temperature slope from a linear mixed model.Within-runner pace response to temperature from a linear mixed model, showing the same runner slows as start temperature rises after controlling for personal baselinehttps://loved-hug-5d84552294.media.strapiapp.com/london_physiology_running_lmm_distance_temp_1_477653b691.pngWithin-person distance-vs-temperature slope from a linear mixed model.Within-runner distance response to temperature from a linear mixed model, showing the same runner shortens runs as start temperature rises after controlling for personal baselinehttps://loved-hug-5d84552294.media.strapiapp.com/london_physiology_running_lmm_hr_temp_1_7ac7041bde.pngFigures 4-6: Linear Mixed effect Model charts to show how pace, distance, and average heart rate change with start temperature when each runner is allowed their own baseline (random intercept), so the slope is a within-person effect rather than “slow people run on hot days.” Pace creeps up and distance falls as it warms (both significant) while HR stays essentially flat, meaning runners slow down and cut distance rather than hold speed with a higher heart rate.Within-runner heart rate response to temperature from a linear mixed model, showing how the same runner’s HR shifts with start temperature after controlling for individual baseline https://tryterra.co/research/world-cup-2026-football-participation-data2026-09-08T09:44:11.295Zweekly0.9https://loved-hug-5d84552294.media.strapiapp.com/1200x630_template_8_db93af40a4.pngThe World Cup Hosts Don't Play Much Footballhttps://loved-hug-5d84552294.media.strapiapp.com/image_47_468bd81b07.pngFigure 1: Football playing users as a percentage of total users per country that manually logged a football activity session.Football participation rates by country from wearable data, showing Chile leading at 4.1% of users regularly logging football sessions, followed by Guatemalahttps://loved-hug-5d84552294.media.strapiapp.com/image_48_16453ed369.pngFigure 2: Scatter plot of football sessions average intensity and duration across the different countries. Color codes represent the continents and bubble sizes the average weekly playing frequency.How different countries play football: scatter of session intensity versus duration coloured by continent and sized by weekly playing frequencyhttps://loved-hug-5d84552294.media.strapiapp.com/05_league_day_heatmap_ce268c0c7f.pngFigure 3: Heat map of most common day of the week to play by country.Football playing day-of-week heat map by country showing weekend concentration in some regions and midweek league nights in others https://tryterra.co/research/does-eating-late-affect-sleep2026-09-08T09:44:10.940Zweekly0.9https://loved-hug-5d84552294.media.strapiapp.com/1200x630_template_7_9bedad3067.pngLate Dinners Raise Your Heart Rate When You Sleephttps://loved-hug-5d84552294.media.strapiapp.com/image_46_5bf19853f7.pngFigure 1: Mean heart rate 4h prior to bed time through to wakeup time on nights with late meals (in teal) and nights without late meals (in red). The teal peak around 2h before bedtime is likely the time of meal consumption.Late dinners and sleeping heart rate: mean HR from four hours before bed through wakeup on late-meal nights versus normal nights, showing elevated HR persists all night https://tryterra.co/research/does-weather-affect-exercise2026-09-08T09:44:10.553Zweekly0.9https://loved-hug-5d84552294.media.strapiapp.com/1200x630_template_f61c8f3bdf.pngDoes Hot Weather Change How Londoners Exercise?https://loved-hug-5d84552294.media.strapiapp.com/london_outdoor_prob_may_active_pact_temp_10_20_30_max_lines_2_0cc16d8748.pngFigure 1: Does a hot day change whether people go out? Note Walking relates to the left hand axis, running and cycling to the right. The trends are messy, but all three activities see flat activity and a small peak in the low-mid twenties, and then decline.Does hot weather change whether Londoners exercise: probability of a walking, running, or cycling session by daily max temperature, peaking in the low-mid twenties then dropping in real heathttps://loved-hug-5d84552294.media.strapiapp.com/london_outdoor_hourly_may_active_running_outdoor_hour_pact_temp_10_20_30_lines_1_7ba20f729f.pngFigure 2: When do people go running, and does timing shift with temperature? Most runs still cluster in the morning and late afternoon. On days when the daily max was ≥ 24°C, the peak hour for starting a run moved earlier (around 6:00 vs 8:00 on cooler days), and activity in the 7–8 a.m. window was lower on hot days. That fits people avoiding mid-morning heat, even when they still run.Running timing and temperature in London: hot days above 24C shift the peak start hour earlier, to around 6am versus 8am on cool dayshttps://loved-hug-5d84552294.media.strapiapp.com/london_outdoor_hourly_may_active_cycling_outdoor_hour_pact_temp_10_20_30_lines_14caa21f19.pngFigure 3: When do people go cycling, and does timing shift with temperature? Ride starts still peak around 7 a.m. on both cool and hot days. Hourly lines cross by temperature band without a simple “only cycle when it’s cool” picture. It’s consistent with Chart 1, where daily heat did not strongly predict a ride day. Again, I wonder if we are seeing a strong “commuter” effect here. A future analysis should do a detailed day of the week analysis here.Cycling timing and temperature in London: ride starts still peak around 7am on both cool and hot days without a clear temperature-driven shifthttps://loved-hug-5d84552294.media.strapiapp.com/london_outdoor_hourly_may_active_walking_outdoor_hour_pact_temp_10_20_30_lines_de354ceb21.pngFigure 4: When do people go walking, and does timing shift with temperature? Walking holds its morning and afternoon peaks across temperatures, but hot days flatten the activity throughout the day.Walking timing and temperature in London: morning and afternoon peaks hold across temperatures but hot days flatten walking activity throughout the day https://tryterra.co/research/bed-sharing2026-09-08T09:44:10.093Zweekly0.9https://loved-hug-5d84552294.media.strapiapp.com/1200x630_template_5_252f005031.pngDo People Sleep Better Alone or Together? https://loved-hug-5d84552294.media.strapiapp.com/chart1_population_summary_41b8a3f173.pngFigure 1: How do standard metrics compare on average between nights where our users shared a bed and slept alone? A hint that bed sharing is better for recovery (Resting HR decreases, but is counteracted by a decrease in RMSSD) even though Sleep efficiency and time asleep decreased?Does bed sharing improve sleep: population comparison across 25,086 nights showing lower resting HR but also lower HRV and slightly worse sleep efficiency when sharing a bedhttps://loved-hug-5d84552294.media.strapiapp.com/fig_within_person_pct_diff_ridgelines_6a0ffca642.pngFigure 2: Shows the % change distribution for each variable. The different shapes of the distributions are particularly interesting, as they demonstrate how the mean differences between cohorts are driven.Within-person bed sharing sleep effect: percent-change distributions for each sleep metric on the same user’s shared-bed versus alone nightshttps://loved-hug-5d84552294.media.strapiapp.com/fig_three_cohort_ridgeline_asleep_awake_c1ef17eaf4.pngFigure 3: shows cohort distribution using ridge lines of night-level asleep and awake distributions by group, with mean and median verticals. It demonstrates that the cohort-mean increase in time awake in bed is driven by a long tail, especially in the most-shared cohort.Bed sharing sleep quality by cohort: distributions of asleep and awake time for mostly-shared, mixed, and mostly-alone sleepers showing 70 vs 66 vs 57 minutes awake https://tryterra.co/research/parkrun-vs-vo2-max2026-09-08T09:44:09.575Zweekly0.9https://loved-hug-5d84552294.media.strapiapp.com/1200x630_template_1_3089931e01.pngIs Parkrun a Better Measure of Fitness Than VO2 Maxhttps://loved-hug-5d84552294.media.strapiapp.com/chart_1_validation_fb5bfb52fe.pngFigure 1: Training speed/HR vs Parkrun speed/HR scatter, r = 0.93. R² = 0.87. Across 78 users, the speed-HR ratio measured from everyday training runs explains 87% of the variance in the same ratio measured under standardized race conditions.Parkrun as a fitness measure: everyday training speed to HR ratio predicts parkrun speed to HR with r=0.93 across 78 users, explaining 87% of variancehttps://loved-hug-5d84552294.media.strapiapp.com/chart_2_timeseries_9abbe74ffd.pngFigure 2: Example user timeseries, training and parkrun speed/HR tracking togetherWithin-person parkrun fitness tracking: training speed to HR ratio and parkrun speed to HR ratio for one user tracking together as fitness changes over monthshttps://loved-hug-5d84552294.media.strapiapp.com/chart_3_leaderboard_f5734ec6a1.pngFigure 3: All metrics ranked by within-user median |r|. The larger r, the better. It shows the metric is “better” at predicting Parkrun pace. Note Frequency is hard to beat, mirroring earlier findings.Best training metrics for predicting parkrun performance: 13 metrics ranked by within-user median correlation, with training frequency near the tophttps://loved-hug-5d84552294.media.strapiapp.com/chart_4_within_user_1_9838fd72f0.pngFigure 4: Individual plots of volume and pace. Volume as a strong predictor, is something we keep seeing in wearable data analysis.Training volume and parkrun pace: individual runner plots showing weekly volume as a strong within-person predictor of parkrun finish pace https://tryterra.co/research/best-time-to-workout-for-sleep2026-09-08T09:44:09.200Zweekly0.9https://loved-hug-5d84552294.media.strapiapp.com/1200x630_template_184187b810.pngYour Evening Workout Is Costing You Sleephttps://loved-hug-5d84552294.media.strapiapp.com/image_40_91b1a82980.pngFigure 1: Time spent awake in bed based on the gap between activity time and bedtime. This is a small effect which is apparent by the error presented.Best time to workout for sleep: time spent awake in bed by the gap between activity and bedtime, showing later workouts mean slightly less time lying awakehttps://loved-hug-5d84552294.media.strapiapp.com/image_41_6ddbefae7c.pngFigure 2: REM event counting based on the gap between activity time and bedtime. Once again, this effect has a small magnitude. Gap distributions between plots can differ based on how much missing data was present for each sleep column.Workout timing and REM sleep: number of REM events by the gap between activity and bedtime, showing earlier workouts produce more REM eventshttps://loved-hug-5d84552294.media.strapiapp.com/image_43_6e91a69e42.pngFigure 3: Sleep heart rate based on the gap between activity time and bedtime. Interestingly, this plot resembles a mirror image of the previous ones.Evening workout sleep effect: sleeping heart rate rises when activity is done closer to bedtime and falls when the workout is earlier in the dayhttps://loved-hug-5d84552294.media.strapiapp.com/image_44_27c1166b1a.pngFigure 4: RMSSD, a measure of your HRV or heart rate variability drops when you have a late workout.Late workouts and recovery: HRV RMSSD drops significantly when workouts are done close to bedtime, indicating worse overnight recoveryhttps://loved-hug-5d84552294.media.strapiapp.com/image_45_729a5ce1b8.pngFigure 5: Sleep efficiency is considerably higher when a workout is done earlier in the day. Sleep efficiency is calculate by dividing time spent asleep by total time spent in bed.Best workout window for sleep efficiency: sleep efficiency is highest when the workout is finished 11 to 14 hours before bedtime https://tryterra.co/research/does-watching-sports-make-us-exercise2026-09-08T09:44:06.777Zweekly0.9https://loved-hug-5d84552294.media.strapiapp.com/1200x630_6c4ca98048.pngDoes Watching Sports Inspire Us to Exercise?https://loved-hug-5d84552294.media.strapiapp.com/blog_fig1_tdf_uplift_1_ff48d3904e.pngFigure 1: Tour de France cycling uplift by country. A horizontal bar chart showing the percentage change in outdoor cycling during the Tour period versus all other days. Germany (+50.5%), France (+39.4%), Italy (+33.7%), and Belgium (+17.9%) are highlighted as statistically significant; the UK and Spain show no uplift. This graphic makes the main Tour de France finding immediately visible.Tour de France cycling uplift by country bar chart, with Germany at +50.5%, France +39.4%, Italy +33.7% and Belgium +17.9%.https://loved-hug-5d84552294.media.strapiapp.com/comprehensive_rolling_average_7cb86e36ad.pngFigure 2: Grouped bar chart of the “excess” cycling uplift (cycling % change minus control % change) when activity is smoothed with 3-day, 7-day, and 14-day rolling averages. Event days are compared with non-event days. X-axis = rolling window (3-day, 7-day, 14-day). Each country has three bars (one per window). Green bars = positive excess (cycling rises more than run/walk/hike); grey bars = zero or negative excess. Y-axis = excess in percentage points.Excess cycling uplift during pro race events by country, smoothed across 3-day, 7-day, and 14-day rolling windows relative to control activities.https://loved-hug-5d84552294.media.strapiapp.com/cycling_chart_0fbdd61093.pngFigure 3: Horizontal bar chart of the share of new/returning cyclists (first-time or 30+ day gap) whose “start” falls on an event day, by country. Each bar = % of that country’s new/returning starts that occur on event days. A dashed vertical line at 9.6% marks the expected share if starts were random (event days are 9.6% of the study period). Green bars = above expected; grey bars = at or below expected. Labels include sample size (n).New and returning cyclists starting on race event days by country, benchmarked against the 9.6% random-share expectation line. https://tryterra.co/research/what-makes-us-dream2026-09-08T09:44:08.854Zweekly0.9https://loved-hug-5d84552294.media.strapiapp.com/dream1_20510ea3be.pngWhat Actually Makes Us Dream?https://loved-hug-5d84552294.media.strapiapp.com/dream1_20510ea3be.pngFigure 1: Percentage of total sleep time that deep, light or awake phases occupied.What kind of sleep leads to remembering dreams: nights with more dream recall show less deep sleep, more light sleep, and more awake timehttps://loved-hug-5d84552294.media.strapiapp.com/dream2_93d4f5800a.pngFigure 2: Sleep onset time, absolute (left) and relative to user baseline (right).Dream recall and bedtime: earlier bedtime and going to bed earlier than personal baseline is the strongest driver of higher dream recallhttps://loved-hug-5d84552294.media.strapiapp.com/dream3_97f7d2e95a.pngFigure 3: Bar plot of per user Z-score values for rest time on days preceding dreaming (red) and days not preceding the recall of dreams (blue).How your day affects dreaming: per-user Z-score comparison of rest time on days that preceded dream recall versus days that did not https://tryterra.co/research/boring-is-fast2026-05-11T15:45:43.939Zweekly0.9https://loved-hug-5d84552294.media.strapiapp.com/chart2_speed_change_distribution_2_8d20feba2d.pngData Shows Most Runners Don't Actually Get Fasterhttps://loved-hug-5d84552294.media.strapiapp.com/chart1_speed_hr_curve_c9d0517067.pngFigure 1: Population speed-HR curves, early vs late. At any given heart rate, the average runner is marginally faster in their final third of activities than their first, but the shift is small, and the curves largely overlap.chart1_speed_hr_curve.pnghttps://loved-hug-5d84552294.media.strapiapp.com/chart2_speed_change_distribution_8d3abb5a72.pngFigure 2: Distribution of speed change. The distribution of individual speed-HR improvements is wide and centered almost exactly on zero, with roughly half of runners improving and half declining. This near-perfect symmetry is evidence the method is capturing real variation, not systematic noise.chart2_speed_change_distribution.pnghttps://loved-hug-5d84552294.media.strapiapp.com/chart4_phenotype_violins_bae9387268.pngFigure 3: Phenotype violin plots. Dedicated runners, consistent, moderate-intensity, high volume, show the clearest positive shift in their speed-HR curve, while Intensity-Focused runners trend negative despite running more than the Casual group.chart4_phenotype_violins.pnghttps://loved-hug-5d84552294.media.strapiapp.com/chart3_volume_quintile_ab9872c3c7.pngFigure 4: Volume quintile bars. Runners in the highest training volume quintile (14.5 runs per month) are 17 percentage points more likely to improve than those in the lowest (2.3 runs per month). The dose-response is roughly linear: each step up in volume shifts the odds measurably in your favor.chart3_volume_quintile.pnghttps://loved-hug-5d84552294.media.strapiapp.com/chart5_consistency_scatter_1_abf8077e37.pngFigure 5: Consistency scatter. Month-to-month variability in training frequency (lower = more consistent) predicts speed-HR improvement more robustly than any other variable tested.chart5_consistency_scatter (1).png https://tryterra.co/research/london-marathon-data2026-09-08T09:44:08.489Zweekly0.9https://loved-hug-5d84552294.media.strapiapp.com/finish_time_ridgeline_173da2a1fd.pngThe Data Behind the London Marathonhttps://loved-hug-5d84552294.media.strapiapp.com/finish_time_distribution_2_e4202b7b56.pngFigure 1: Distribution of sample finish times. The whole field on one canvas: 571 runners binned into 5-minute slots, with the median finish (3:55:13) marked in coral. Look closely, and you may see some evidence of round-number magnetism. There are small spikes 3:55 and 4:55 where runners squeeze in just under their target.London Marathon finish time distribution across 571 amateur runners with median finish 3:55:13, showing 54% went sub-4 and 12% went sub-3https://loved-hug-5d84552294.media.strapiapp.com/finish_time_ridgeline_173da2a1fd.pngFigure 2: Ridgeline Plot of wave finishing time. The same data split by start wave, which exposes the structure that the single histogram hides. London's seeding works, there's nearly a two-hour gap in median finish time between the first wave and the last.London Marathon finish times by start wave showing a nearly two-hour gap in median finish between the first wave and the lasthttps://loved-hug-5d84552294.media.strapiapp.com/distance_distribution_1_25b49f5fa8.pngFigure 3: Distribution of device-recorded distance. 98% of runners over-measured by an average of 527 m. The tail to the right and the strong correlation with finish time (slower runners over-measure more) point to walking before the start, weaving in the pack, and GPS jitter as the three culprits.Marathon GPS watch over-measurement: 98% of runners recorded more than 42.195 km, with the median watch logging 527 metres extra due to weaving and stopshttps://loved-hug-5d84552294.media.strapiapp.com/calories_ridgeline_694f88c88a.pngFigure 4: Distributions of the device-reported Calorie expenditure. Same race, same data, two different stories from the watch on your wrist. Coros runs about 332 kcal hotter at the median (a 12% gap) and has a markedly heavier right tail, with some Coros watches reporting over 5,000 kcal for a marathon, a useful reminder that calorie numbers from a wearable are a model output, not a measurementMarathon calorie tracking accuracy by brand: Coros watches report about 332 kcal more than Garmin at the median, a 12% gap with a heavier upper tailhttps://loved-hug-5d84552294.media.strapiapp.com/calories_per_km_9895ddbf50.pngFigure 5: Plot of calories expended per kilometer for different activity times on different devices. Each line is the median kcal-per-kilometer reported by that brand's watches across runners in each finish-time band, with the IQR shaded and a dotted line marking the textbook physiological estimate (~1 kcal per kg per km, roughly 70 kcal/km for a 70 kg runner).Calories per kilometre reported by watch brand across marathon finish-time bands, showing Garmin tightly clustered around 2,956 kcal and Coros shifted right to 3,288 kcal https://tryterra.co/research/tube-strikes-london-activity2026-09-08T09:44:08.021Zweekly0.9https://loved-hug-5d84552294.media.strapiapp.com/commute_vector_field_map_a351c421c1.pngTube Strikes Made Londoners Activehttps://loved-hug-5d84552294.media.strapiapp.com/daily_activity_counts_34cb6a8944.pngFigure 1: Cycling activities throughout the weeks leading up to the strikes. Weekends shaded out and strikes labelled in red.Tube strike cycling activity in London tripling during the April 2026 strike compared to the prior three weeks, from wearable-tracked rideshttps://loved-hug-5d84552294.media.strapiapp.com/sample_speed_trace_clusters_pca_20c99bf253.pngFigure 2: Left: the two ride clusters in PCA space. The traces show a typical e-bike-like ride and a typical normal-bike-like ride. Right: the features that most strongly separate them.E-bike versus normal bike ride classification using k-means clustering on speed traces, showing the two distinct ride profiles used to detect e-bike commuteshttps://loved-hug-5d84552294.media.strapiapp.com/ebike_activity_percentage_bars_49743b4764.pngFigure 3: Estimated share of e-bike rides in sampled weekday activities. During the strike days, the e-bike share rises from 54.9% to 67.5%.E-bike commute share during the London Tube strike rising from 54.9% to 67.5% of sampled weekday rides during strike dayshttps://loved-hug-5d84552294.media.strapiapp.com/strike_positions_map_rounded_v2_2_cd6c126f06.pngFigure 4: Rounded weekday source and sink locations during the April 2026 strike period. Blue marks likely trip origins; red marks likely destinations.London strike cycling commute map showing likely trip origins in blue and destinations in red across 1,566 weekday rides during the April 2026 strikehttps://loved-hug-5d84552294.media.strapiapp.com/commute_vector_field_map_a351c421c1.pngFigure 5: Smoothed weekday commute field during the strike period. Blue regions act more like trip sources; red regions act more like trip sinks.London cycling commute flow map during Tube strikes showing central London as a major destination alongside many short-to-medium local trips between residential and work areas https://tryterra.co/research/endurance-training-comparison2026-09-08T09:44:07.648Zweekly0.9https://loved-hug-5d84552294.media.strapiapp.com/chart1_hr_ridge_1_1020ae4502.pngRunning Is The Hardest Endurance Sport? Not so Fasthttps://loved-hug-5d84552294.media.strapiapp.com/chart1_hr_ridge_1_1020ae4502.pngFigure 1:  HR Distribution Ridge Plot. Each distribution is normalized to its own peak to compare shapes across modalities. Dashed lines indicate the median. Running is the most concentrated and furthest to the right; ski touring and hiking are more spread out, highlighting variable session intensity.Heart rate distribution by endurance sport: ridge plot comparing HR shapes across nine activities, with running most concentrated at the high end and ski touring more spread outhttps://loved-hug-5d84552294.media.strapiapp.com/chart2_avg_vs_peak_1_835a65c163.pngFigure 2: Average HR vs Peak HR Scatter. Running occupies the top-right quadrant: both the highest average and highest peak HR, confirming genuine sustained high output. Walking clusters bottom-left — its high ratio (86%) reflects a flat, low-amplitude session rather than true intensity.Hardest endurance sport by heart rate: average versus peak HR scatter placing running in the top-right quadrant with the highest sustained cardiovascular outputhttps://loved-hug-5d84552294.media.strapiapp.com/chart3_decomposition_f21c8c23c3.pngFigure 3: Sustained Intensity Decomposition (3-panel bars). The ranking shifts between panels reveal the mechanism. XC skiing and outdoor cycling rank top-3 in peak HR but drop sharply in sustained intensity — high peaks with long low-HR recovery phases. Running leads in both average and ratio.Sustained intensity ranking by endurance sport showing running and indoor cycling deliver the most consistent high output while XC skiing has high peaks with long low-HR recovery phaseshttps://loved-hug-5d84552294.media.strapiapp.com/chart6_within_person_ri_d56b239e52.pngFigure 4: Within-Person Relative Intensity Ridge. Each user's session average HR is expressed as a percentage of their highest ever recorded heart rate. Even controlling for individual fitness, running sits highest — 93% of runner-cyclists train harder when they run.Within-person endurance sport intensity across 2,808 multi-sport users showing running still sits highest at 93% of personal max HR after controlling for fitnesshttps://loved-hug-5d84552294.media.strapiapp.com/chart7_time_weighted_bc392a2969.pngFigure 5: Time-Weighted Intensity (3-panel bars). Running is the most intense sport per minute (B), but shorter sessions mean its total cardiovascular load per session is lower than outdoor cycling, XC skiing, or hiking (C). Intensity-minutes = relative intensity × duration.Total cardiovascular load per endurance session: running is the most intense per minute but shorter sessions mean cycling, XC skiing, and hiking produce higher total loadhttps://loved-hug-5d84552294.media.strapiapp.com/chart8_weighting_bump_f9ac5ddb5e.pngFigure 6: Weighting Method Bump Chart. How you weight intensity changes the answer. Under a linear model, running ranks 7th — duration dominates. Apply the exponential weighting derived from blood lactate data (TRIMP), and running climbs to 2nd, behind only XC skiing.How the hardest endurance sport ranking changes with weighting: running ranks 7th under linear volume weighting but climbs to 2nd under exponential TRIMP weighting https://tryterra.co/research/cold-plunging-biomarker-effect2026-09-08T09:44:07.244Zweekly0.9https://loved-hug-5d84552294.media.strapiapp.com/image_30_e0fdacc17f.pngCold Plunging Might Make Your Biomarkers Worsehttps://loved-hug-5d84552294.media.strapiapp.com/image_30_e0fdacc17f.pngFigure 1: Sporadic cold exposure is associated with higher sleep minimum heart rate, consistent with an acute stress response.Cold plunging effect on sleep minimum heart rate showing sporadic exposure raises nighttime HR, consistent with an acute stress response.https://loved-hug-5d84552294.media.strapiapp.com/image_31_1a0814b440.pngFigure 2: Interaction model showing that more frequent recent cold exposure shifts the effect of cold days toward lower nighttime minimum heart rate.Cold plunging interaction model showing that more frequent recent exposure shifts nighttime minimum heart rate downward toward a recovery signal.https://loved-hug-5d84552294.media.strapiapp.com/exploratory_recovery_composite_adapted_cd5547b6ec.pngFigure 3: Exploratory recovery-oriented composite in adapted users. The composite combines recovery score, sleep score, and sign-flipped daily minimum and average heart rate, so higher values indicate a more recovery-like physiological state.Recovery composite score in adapted cold-plunge users, combining recovery, sleep, and heart rate signals across sauna, cold, and combined exposure days.https://loved-hug-5d84552294.media.strapiapp.com/panel_c_luteal_paired_sleep_avg_hr_16ff5c730c.pngFigure 4: In luteal phase, frequent cold exposure is associated with higher sleep average heart rate versus non-cold days. No equivalent effect was seen in follicular phase.Cold exposure effect during the luteal phase showing higher sleep average heart rate versus non-cold days, with no comparable follicular effect. https://tryterra.co/research/winter-activity-tracking2026-09-08T09:44:06.373Zweekly0.9https://loved-hug-5d84552294.media.strapiapp.com/chart6_modality_share_3a6af25759.pngHow The Fittest Do Winterhttps://loved-hug-5d84552294.media.strapiapp.com/chart6_modality_share_3a6af25759.pngFigure 1: Running and outdoor cycling account for the bulk of Europe's active hours year-round, but their relative share shifts with the seasons. Indoor cycling surges to roughly 15% of all hours in January before almost disappearing by June, while hiking and swimming expand to fill the summer months.Seasonal share of European exercise modalities across the year, showing indoor cycling peaking near 15% in January and fading by June.https://loved-hug-5d84552294.media.strapiapp.com/chart_volume_dual_aa031fb08c.pngFigure 2: The left panel shows per-user session counts; the right shows active hours. The seasonal swing is more dramatic in hours (35% summer-to-winter drop) than sessions (25%), because summer activities like outdoor cycling are substantially longer per session.Dual panel comparing per-user session counts and active hours by month, with a larger 35% summer-to-winter swing in hours than sessions.https://loved-hug-5d84552294.media.strapiapp.com/chart2_indoor_outdoor_1_873efcabb7.pngFigure 3: Indoor and outdoor cycling volumes show a near-perfect inverse seasonal relationship. As outdoor cycling peaks in June to August, indoor cycling drops to its minimum, and vice versa in winter.Indoor versus outdoor cycling seasonal volume chart showing a near-perfect inverse relationship across the calendar year.https://loved-hug-5d84552294.media.strapiapp.com/chart5_seasonality_cv_4afd910860.pngFigure 4: The CV measures how much a modality's volume swings across months. Walking (11%) and running (13%) are nearly year-round activities, while downhill skiing (125%) and open-water swimming (112%) are confined almost entirely to specific seasons.Seasonality coefficient of variation by sport, from year-round walking (11%) and running (13%) to highly seasonal downhill skiing (125%).https://loved-hug-5d84552294.media.strapiapp.com/chart3_running_hr_seasons_1_73b47728ee.pngFigure 5: Running average heart rate distributions for summer and winter almost perfectly overlap. Median HR sits at 144 to 145 bpm regardless of season, suggesting runners pace by perceived effort rather than being driven by environmental conditions.Running heart rate distribution overlay for summer and winter, showing median staying at 144–145 bpm regardless of season.https://loved-hug-5d84552294.media.strapiapp.com/chart4_cycling_hr_seasons_1_94488af455.pngFigure 6: Cycling shows a small but consistent winter HR elevation (median 126 bpm vs 124 bpm in summer). This likely reflects self-selection, where only the most committed riders cycle outdoors in cold months, and they tend to push harder.Cycling heart rate distribution by season showing a small winter elevation, with median 126 bpm versus 124 bpm in summer.https://loved-hug-5d84552294.media.strapiapp.com/cluster_v3_modality_b90b34deb4.pngFigure 7: The three phenotypes occupy distinct volume ranges across the year. The committed group (green) operates on an entirely different scale; their winter trough still exceeds the other groups' summer peaks.Exerciser phenotype volume ranges across the year, showing the committed group's winter trough still exceeds other groups' summer peaks.https://loved-hug-5d84552294.media.strapiapp.com/cluster_v3_modality_1_cf0452d7b2.pngFigure 8: Each phenotype has a radically different sport mix. The indoor cycling specialists are almost entirely single-modality, while the committed and casual groups spread their hours across running, cycling, and walking.Sport mix comparison across three exerciser phenotypes, with indoor cycling specialists appearing almost entirely single-modality relative to committed and casual groups.https://loved-hug-5d84552294.media.strapiapp.com/cluster_v3_pca_e1c3368ede.pngFigure 9: Principal component analysis projects each user's behavioral features into two dimensions. The indoor cycling specialists (pink) form a distinct island, while the committed athletes (blue) and summer casuals (green) separate along the volume and density axes.Principal component analysis projection of exerciser behavioral features, with indoor cycling specialists forming a distinct island from committed and casual clusters. https://tryterra.co/research/breathing-period-tracking2026-09-08T09:44:05.906Zweekly0.9https://loved-hug-5d84552294.media.strapiapp.com/image_f02c5bafca.pngYour Breathing Changes a Week Before Your Periodhttps://loved-hug-5d84552294.media.strapiapp.com/image_f02c5bafca.pngFigure 1: Nighttime breathing rate trajectory across the menstrual cycle. In red, trajectories where PMS was reported shows higher breathing rate a week before period start compared to non PMS cycles, in blue.Nighttime breathing rate trajectory across the menstrual cycle, comparing PMS-reported cycles against non-PMS cycles around period start.https://loved-hug-5d84552294.media.strapiapp.com/image_1_5837bc2ff2.pngFigure 2: Cohen d effect size of symptom tagging prevalence in PMS days compared to non PMS days. Emotional symptoms, in red show higher prevalence than physical symptoms in blue.Cohen's d effect sizes for symptom tagging on PMS versus non-PMS days, showing emotional symptoms more prevalent than physical ones.https://loved-hug-5d84552294.media.strapiapp.com/image_2_99c339628c.pngFigure 3: Correlations between PMS and other tags.Correlation matrix between PMS reporting and other user-logged tags, with the strongest link appearing between PMS and sugar cravings. https://tryterra.co/research/alcohol-effects-on-sleep2026-09-08T09:44:05.147Zweekly0.9https://loved-hug-5d84552294.media.strapiapp.com/alc_awake_usermeans_2curves_c938595429.pngAlcohol Doesn't Ruin Your Sleep The Way You Thinkhttps://loved-hug-5d84552294.media.strapiapp.com/alc_awake_dist_1_b6854cd307.pngFigure 1: Distribution of time spent awake in bed.Alcohol effect on sleep distribution chart showing time spent awake in bed, with a wider spread on drinking nights.https://loved-hug-5d84552294.media.strapiapp.com/alc_awake_usermeans_2curves_c938595429.pngFigure 2: Within user means of time spent awake in bed on alcohol and non-alcohol nights.Within-user comparison of mean time spent awake in bed on alcohol versus non-alcohol nights, highlighting a consistent per-person increase.https://loved-hug-5d84552294.media.strapiapp.com/alc_wakeup_events_dist_1beddf32c8.pngFigure 3: Distribution of number of wakeup events on alcohol vs non-alcohol nights.Wakeup event distribution comparing alcohol and non-alcohol nights, showing more frequent nighttime awakenings when drinking before bed.https://loved-hug-5d84552294.media.strapiapp.com/within_user_breaths_alcohol_vs_no_alcohol_058cc80e6a.pngFigure 4: Within user means of respiratory rate while asleep on nights with alcohol vs no alcohol.Within-user mean respiratory rate during sleep on alcohol versus non-alcohol nights, showing elevated breathing on drinking nights. https://tryterra.co/research/sauna-effect-on-heart-rate2026-09-08T09:44:05.493Zweekly0.9https://loved-hug-5d84552294.media.strapiapp.com/sauna_hr_984f14dcae.pngSaunas Lower Your Heart Rate More Than Exercisehttps://loved-hug-5d84552294.media.strapiapp.com/sauna_hr_984f14dcae.pngFigure 1: Heart rate drop from pre-sleep baseline, showed throughout sleep duration. Sauna days, in red, have lower heart rates compared to non sauna days in blue.Sauna effect on heart rate trajectory across sleep duration, with sauna nights showing lower heart rate than non-sauna nights.https://loved-hug-5d84552294.media.strapiapp.com/image_25_658d707e3f.pngFigure 2: Bar chart of the effect sizes (cohen d values) of the changes in physiology and activity on sauna days compared to non sauna days. Females (in pink) have greater increase in activity time compared to males (in blue), but lower drop in heart rate at night.Sauna effect size bar chart of Cohen's d values comparing physiology and activity changes on sauna versus non-sauna days, split by sex. https://tryterra.co/research/GLP-effects-on-training2026-09-08T09:44:04.666Zweekly0.9https://loved-hug-5d84552294.media.strapiapp.com/rolling_exercise_hrv_762341e614.pngWhat GLP-1s are Really Doing to Your Traininghttps://loved-hug-5d84552294.media.strapiapp.com/Screenshot_2026_03_16_at_17_35_09_1_d8b72b2e85.pngFigure 1: Relationship between exercise load and physiological response for GLP-1 users versus controls, drawn from over 70,000 activities.GLP-1 effects on training chart showing how the medication shifts the relationship between exercise load and cardiovascular response in active users.https://loved-hug-5d84552294.media.strapiapp.com/Screenshot_2026_03_16_at_17_42_21_51e68f9793.pngFigure 2: Heart rate and HRV recovery patterns after training sessions, contrasting GLP-1 users with matched non-users.GLP-1 training recovery chart illustrating how heart rate and HRV responses to exercise shift for users on the medication. https://tryterra.co/research/caffeine-effect-on-sleep2026-09-08T09:44:04.224Zweekly0.9https://loved-hug-5d84552294.media.strapiapp.com/caf_dist_1_b097afa9e9.pngCaffeine Kills Your Deepest Sleep Firsthttps://loved-hug-5d84552294.media.strapiapp.com/share_0658cdf8f2.pngFigure 1: Baseline share of stage duration against duration reduction in stages due to caffeine consumption.Caffeine effect on sleep stages showing baseline share of each stage duration alongside the reduction in duration attributed to caffeine consumption.https://loved-hug-5d84552294.media.strapiapp.com/caf_dist_1_b097afa9e9.pngFigure 2: Distribution of blood oxygen (SpO2) on caffeine vs non-caffeine nights. The caffeine distribution is slightly wider and left-shifted, but the difference in means is small, which is likely driven by fragmented sleep rather than a direct caffeine effect.Blood oxygen distribution comparison for caffeine versus non-caffeine nights, with the caffeine distribution slightly wider and shifted left. https://tryterra.co/research/GLP-1s-side-effects2026-09-08T09:44:03.789Zweekly0.9https://loved-hug-5d84552294.media.strapiapp.com/glp_steps_1_12e675f651.pngGLP-1s: The Real Tradeoffs Nobody Tells You Abouthttps://loved-hug-5d84552294.media.strapiapp.com/glp_steps_1_12e675f651.pngFigure 1: Daily steps relative to GLP-1 exposure - Activity trends downward before the first tag and stays below baseline for 20+ days after. Each user is compared to their own average (z-scored within person, n=538).GLP-1 side effect on daily steps, with within-person z-scored activity trending downward before the first tag and staying below baseline for 20+ days.https://loved-hug-5d84552294.media.strapiapp.com/Screenshot_2026_03_06_at_16_32_33_f74a5778bb.pngFigure 2: Minimum heart rate rises after GLP-1 exposure - the shift is most pronounced 2–7 days post-tag (z-scored within person, n=538).GLP-1 effect on minimum heart rate showing a rise after exposure, most pronounced two to seven days after the first tag.https://loved-hug-5d84552294.media.strapiapp.com/Screenshot_2026_03_06_at_16_32_50_7aebf547b7.pngFigure 3: HRV drops below baseline after GLP-1 exposure - remaining suppressed for 20+ days (z-scored within person, n=538).GLP-1 effect on HRV showing suppression below baseline that persists for more than 20 days after exposure, z-scored within person. https://tryterra.co/research/melatonin-sleep-effect2026-09-08T09:44:03.432Zweekly0.9https://loved-hug-5d84552294.media.strapiapp.com/image_14_dd1dacbd0d.pngMelatonin’s Real Effect Peaks on Day 4 - And It's NOT Sleephttps://loved-hug-5d84552294.media.strapiapp.com/image_14_dd1dacbd0d.pngFigure 1. Trends in overnight heart rate (HR) and HRV (RMSSD) z-scores across consecutive nights of melatonin use. The black line is the LOWESS‑smoothed mixed‑effects estimate; the shaded band shows the 95% confidence interval from the smoothed standard errors. Blue segments indicate intervals where the metrics are statistically different from baseline.Melatonin effect on overnight heart rate and HRV z-scores across consecutive nights of use, with peak separation around days three to five.https://loved-hug-5d84552294.media.strapiapp.com/image_16_77ea46e8d5.pngFigure 2. Trends in overnight heart rate (HR) and HRV (RMSSD) z-scores in the 14 days after stopping melatonin. The black line is the LOWESS‑smoothed mixed‑effects estimate; the shaded band shows the 95% confidence interval. Blue segments indicate intervals where metrics statistically differ from non-melatonin baselines.Melatonin withdrawal effect on overnight heart rate and HRV z-scores across the 14 days after users stopped taking the supplement.https://loved-hug-5d84552294.media.strapiapp.com/image_17_6b38153f0c.pngFigure 3. Trends in sleep duration, sleep onset, sleep latency, and time in bed z-scores across consecutive nights of melatonin use. The black line is the LOWESS‑smoothed mixed‑effects estimate; the shaded band shows the 95% confidence interval from the smoothed standard errors.Melatonin effect on sleep performance trends for duration, onset, latency, and time in bed z-scores across consecutive nights of use. https://tryterra.co/research/period-tracking-without-skin-temperature2026-09-08T09:44:03.087Zweekly0.9https://loved-hug-5d84552294.media.strapiapp.com/image_11_f2ab9a4029.pngCan We Predict Cycle Phase Without Temperature?https://loved-hug-5d84552294.media.strapiapp.com/image_10_a53c943027.pngFigure 1: Mean autocorrelation profiles for temperature, sleep metrics, and physiological metrics. Temperature and physiological metrics (HR, HRV and Breathing rate) have monthly rhythms, corresponding to the menstrual cycle. Sleep has a flat autocorrelation after the initial dip, so it does not encode monthly phase.Mean autocorrelation profiles for temperature, sleep, and physiological metrics, revealing monthly menstrual rhythms in heart rate, HRV, and breathing.https://loved-hug-5d84552294.media.strapiapp.com/image_11_f2ab9a4029.pngFigure 2: Menstruation start dates in respect to the temperature profile. The majority of menstruation starts align with the drop of temperature as the body transitions from luteal to follicular.Menstruation start dates overlaid on skin temperature profile, showing that period onset aligns with the luteal-to-follicular temperature drop.https://loved-hug-5d84552294.media.strapiapp.com/image_12_abcc9999ce.pngFigure 3: Visualization of model performance in predicting luteal (1) or follicular (-1) phases. Temperature profile for a randomly selected user is plotted in navy, the self-reported menstruation starts in red, the temperature-derived phase labels in pink and the predicted phases in cyan.Cycle phase prediction visualization for one user, comparing temperature-derived luteal and follicular labels against model predictions using non-temperature signals. https://tryterra.co/research/norway-winter-olympics-dominance2026-09-08T09:44:02.711Zweekly0.9https://loved-hug-5d84552294.media.strapiapp.com/medals_per_capita_67ca8019a6.pngThe Secret Behind Norway's Olympic Dominancehttps://loved-hug-5d84552294.media.strapiapp.com/norway_vs_others_comparison_db98fb9b4d.pngFigure 1: Activity Type Comparison - Norway vs Others (Top 5 Average) shows how Norwegians allocate time across activity types compared to the average of other top countries. It (surprisingly!) shows that Norwegians do a lot of cross-country skiing, but also rely heavily on walking and running.Norway exercise habits comparison chart showing activity time allocation across sports for Norway versus the top five other European countries.https://loved-hug-5d84552294.media.strapiapp.com/activity_modality_composition_all_users_d29d1ab021.pngFigure 2: Activity Time Composition by Modality - how total monthly activity time is split across activity types for Norway and the top 5 countries. Norwegians spend more time walking/hiking and cross-country skiing (other activities), while others rely more on indoor activities like stationary biking and elliptical machines.Activity modality composition breakdown showing how monthly training time splits by sport for Norwegian users versus other top European countries.https://loved-hug-5d84552294.media.strapiapp.com/intensity_distribution_top10_norway_vs_others_9cced2491c.pngFigure 3: Intensity Distribution - Top 10% Most Active Users (Norway vs Others) . Norway's top 10% (most active users) spend more time in the tempo and threshold zones (Z3–Z4) and less time in the maximum-intensity zone (Z5) than the rest.Training intensity distribution for the top 10% most active users, comparing Norwegian time in zones Z1–Z5 against other countries. https://tryterra.co/research/predicting-marathon-race-pace2026-09-08T09:44:02.314Zweekly0.9https://loved-hug-5d84552294.media.strapiapp.com/Whats_App_Image_2026_02_20_at_11_44_15_47e9887731.jpegSimulate Your Next Marathon Race Pacehttps://loved-hug-5d84552294.media.strapiapp.com/all_parameter_charts_37e38f5108.pngFigure 1: These figures show how individual training parameters (baseline pace, volume, and intensity zones) predict marathon time, with other variables held constant at their mean values. The bottom row demonstrates that training effects vary by baseline ability—faster runners benefit more from both volume and easy training than slower runners, highlighting the importance of personalised training approaches.Marathon race pace model panels showing how baseline pace, training volume, and intensity zones each predict finish time with other variables held constant. https://tryterra.co/research/non-linear-modeling-marathon-pace2026-09-08T09:44:01.880Zweekly0.9https://loved-hug-5d84552294.media.strapiapp.com/nonlinear_volume_no_outliers_5a2260ac62.pngHow To Train For Your Next Marathonhttps://loved-hug-5d84552294.media.strapiapp.com/nonlinear_volume_no_outliers_5a2260ac62.pngFigure 1: Piecewise Linear Fit of Marathon Time Residuals vs. Total Training Volume. Scatter plot of the relationship between total training volume and marathon timePiecewise linear fit of marathon time residuals against total training volume, showing diminishing returns beyond roughly 523 km.https://loved-hug-5d84552294.media.strapiapp.com/nonlinear_zone1_a8bb2d577c.pngFigure 2: Marathon Time Residuals vs. Percentage of Easy Training (Zone 1) This scatter plot shows the relationship between the percentage of total training spent in Zone 1 (easy/aerobic/conversational effort) on the x-axis (0% to 100%) and marathon time on the y-axis.Marathon time residuals plotted against percentage of easy Zone 1 training, revealing accelerating benefits as easy share grows.https://loved-hug-5d84552294.media.strapiapp.com/volume_intensity_percentile_heatmap_controlled_baseline_pace_b20bbc1d42.pngFigure 3: Volume × Easy Intensity Heatmap – Impact on Marathon Performance (Baseline Pace Controlled). This heatmap shows how total training volume (km) and percentage of easy/Zone 1 training combine to affect marathon performance, after controlling for each runner's natural baseline pace.Volume-by-easy-intensity heatmap showing how training kilometres and Zone 1 share combine to shape marathon performance after baseline pace control. https://tryterra.co/research/sleep-duration-variation-latitudes-countries2026-05-07T18:16:51.148Zweekly0.9https://loved-hug-5d84552294.media.strapiapp.com/users_world_map_clean_30561b7467.pngSeasonal Effects on Sleep Performance https://tryterra.co/research/training-intensity-performance-effect2026-09-08T09:44:01.483Zweekly0.9https://loved-hug-5d84552294.media.strapiapp.com/key_finding_1_intensity_zones_8949df6da9.pngThe Easier You Train The Faster You Finishhttps://loved-hug-5d84552294.media.strapiapp.com/key_finding_2_consistency_1ecf041346.pngFigure 1: Training Consistency: Mean Runs per 7 Days vs. Marathon Time Scatter plot shows a clear negative relationship: runners averaging 5+ runs/week tend to finish faster (often ≤250 min), while 0–3 runs/week spans much slower times (~350–450 min). Consistency is one of the strongest single predictors (partial r ≈ -0.40 after baseline control), suggesting steady stimulus—not sporadic hard blocks—drives adaptation.Training consistency scatter plot showing mean runs per week versus marathon finish time, with higher weekly frequency linked to faster finishes.https://loved-hug-5d84552294.media.strapiapp.com/key_finding_1_intensity_zones_8949df6da9.pngFigure 2: Intensity Distribution: % Time in Easy (Zone 1) vs. Race-Pace (Zone 3) Scatter plot of marathon time against percentage of training time spent in Zone 1 (blue, slowest/easy) and Zone 3 (red, fastest/race pace). Blue points trend downward (easier time → faster finishes), while red points trend upward (more race-pace time → slower finishes).Marathon training intensity distribution scatter comparing time in easy Zone 1 versus race-pace Zone 3 against marathon finish time.https://loved-hug-5d84552294.media.strapiapp.com/key_finding_3_longest_run_0ac6a5e75c.pngFigure 3: Longest Training Run Distance vs. Marathon Time Marathon times improve as the longest run increases, with the best averages typically in the 25–35 km range, and mean times dropping from 294 min (<20 km) to 238 min (35+ km). However, once you control for baseline pace and total volume, longest-run distance (and long-run frequency, r = -0.249 univariately) adds little independent signal, suggesting the apparent benefit is largely explained by overall training volume and consistency.Longest training run distance versus marathon finish time scatter showing best average times cluster in the 25–35 km long-run range. https://tryterra.co/research/period-tracking-womens-health2026-09-08T09:44:01.001Zweekly0.9https://loved-hug-5d84552294.media.strapiapp.com/plot1_cluster_timeseries_9541bd39ff.pngHow Temperature Predicts Women's Cycle Phaseshttps://loved-hug-5d84552294.media.strapiapp.com/plot1_cluster_timeseries_9541bd39ff.pngFigure 1. Autocorrelation profiles of night temperature recordings clustered by gender. In red, females have a cyclical autocorrelation aligning with the menstrual cycle while men have a static autocorrelation profile.Menstrual cycle temperature signature in wearable data: autocorrelation profiles showing a clear cyclical pattern in female night temperature and a flat pattern in maleshttps://loved-hug-5d84552294.media.strapiapp.com/plot2_mean_std_2865ca6007.pngFigure 2: Mean female temperature profile over 200 days, phase-aligned for visualization. Luteal is shaded green and follicular pink. The larger central peak reflects the alignment (Heaviside step), not a true increase in temperature amplitude.Female cycle temperature curve over 200 phase-aligned days showing the luteal-versus-follicular temperature shift used to predict menstrual phase from wearableshttps://loved-hug-5d84552294.media.strapiapp.com/plot3_phase_bars_4ee86911ef.pngFigure 3. Mean deviations from individual baseline (rolling z-scores) across cycle phases. The luteal phase shows higher night heart and breathing rates, lower HRV (RMSSD), and lighter, more fragmented sleep.Sleep and physiology changes by menstrual cycle phase: the luteal phase shows higher night heart and breathing rates, lower HRV, and more fragmented sleep than the follicular phase https://tryterra.co/research/WHOOP-Sleep-Tracking-Comparison2026-09-08T09:44:00.528Zweekly0.9https://loved-hug-5d84552294.media.strapiapp.com/Whoop_Sleep_Figure_1_48b3e23c18.pngSleep Metric Variation Across Wearableshttps://loved-hug-5d84552294.media.strapiapp.com/manufacturer_sleep_stages_distributions_84b9773e4b.pngFigure 1: Stacked Bar Chart of Average Sleep Stages by Manufacturer - breaks down mean hours in REM (blue), Deep (green), and Light (yellow) sleep, stacked to show total sleep time. The chart highlights Apple's notably lower deep sleep and Eight Sleep's higher REM, illustrating inter-device staging differences.Sleep stage comparison across six wearables: stacked bar chart of average REM, deep, and light sleep hours per device across 77,000+ nightshttps://loved-hug-5d84552294.media.strapiapp.com/manufacturer_population_distributions_7b954256e2.pngFigure 2: Mean Total Sleep Time with Error BarsBar chart showing average TST per device, with sample sizes noted and variability indicated. Eight Sleep leads in duration, while shorter averages on wrist-based devices like Apple and Fitbit may reflect user habits or detection algorithms. Is this because Eight Sleep users self-select as the most sleep-conscious in the sample, or because the product itself extends sleep? More of this later!Average nightly sleep duration by wearable brand with variability shown, ranging from 6.89 to 7.65 hours across WHOOP, Oura, Apple, Fitbit, Garmin, and Eight Sleephttps://loved-hug-5d84552294.media.strapiapp.com/manufacturer_tst_distribution_lines_1_7af45f68e5.pngFigure 3: Distributions of Total Sleep Time - Smoothed density curves for TST across devices, showing that most nights cluster around 6.5-8 hours. Distributions overlap significantly, but peaks vary slightly—e.g., Eight Sleep skewed higher.Total sleep time distribution curves for each wearable brand showing most nights cluster between 6.5 and 8 hours, with substantial overlap and different peak locationshttps://loved-hug-5d84552294.media.strapiapp.com/whoop_tst_comparison_chart_1_8ccf4c5eaf.pngFigure 4: WHOOP TST Differences in Multi-Device ComparisonsBar chart comparing average and median differences (minutes) for a range of different devices. Medians under 25 minutes indicate solid agreement on total time, despite larger averages pulled by outliers.WHOOP sleep tracking accuracy versus other wearables in same-night comparisons, showing average and median total sleep time differences in minuteshttps://loved-hug-5d84552294.media.strapiapp.com/whoop_sleep_stages_comparison_chart_9fce8a347c.pngFigure 5: WHOOP Sleep Stage Differences- Average absolute differences in minutes for REM, Deep, and Light stages in the same comparisons. Stage discrepancies are substantially larger than TST, emphasising ongoing challenges in consumer staging accuracy.WHOOP sleep stage discrepancies versus other wearables: average absolute differences in REM, deep, and light minutes on the same nightshttps://loved-hug-5d84552294.media.strapiapp.com/Whoop_Sleep_Figure_1_48b3e23c18.pngFigure 6: Sleep Stage Duration Breakdown by Device - wide variance in deep sleep, REM sleep, and light sleep across different wearable devices.Sleep stage duration breakdown by device showing wide variance in deep, REM, and light sleep classification across wearables in same-night comparisons https://tryterra.co/research/the-january-activity-spike2026-09-08T09:43:59.920Zweekly0.9https://loved-hug-5d84552294.media.strapiapp.com/Frame_1597880205_a52ca12e5d.pngThe January Activity Spike: A two-day Storyhttps://loved-hug-5d84552294.media.strapiapp.com/chart2_dow_duration_distribution_2_77c22a5f94.pngFigure 1: Distribution of Elevated Activity Period Durations. Bar chart illustrates the number of users maintaining activity above their personal baseline for consecutive days starting January 1st. The vast majority of elevations were short, with a sharp peak at 1 day (especially for active minutes and distance), and a rapid tail-off after 4 days.How long the New Year activity spike lasts: how many users maintained elevated activity above personal baseline for consecutive days starting January 1sthttps://loved-hug-5d84552294.media.strapiapp.com/chart_period_duration_cumulative_939e284fd1.pngFigure 2: Cumulative Percentage of Users Maintaining Elevated Activity. This line chart shows the running total of users (as a percentage of all with data) who stayed above the threshold for at least the specified number of days. Steps show the steepest early rise, while active minutes and distance catch up more gradually.Cumulative percentage of users maintaining elevated New Year activity for at least N days, showing most drop back to normal levels within one to two dayshttps://loved-hug-5d84552294.media.strapiapp.com/chart_frequency_quantity_comparison_1_c323af9f0e.pngFigure 3: This grouped bar chart shows how users who exhibited periods of increased activity on or around January 1st did so across three metrics: steps. active minutes, and distance.January activity spike details: change in activity frequency versus total step quantity between January 1st and typical baseline weekshttps://loved-hug-5d84552294.media.strapiapp.com/chart_regression_difference_1_6c6d2110ec.pngFigure 4 shows that less-active users (in the lowest baseline quartile, Q1) were far more likely to increase their activity on January 1st than on a typical Wednesday, with differences up to +25.1% in active minutes. In contrast, the most active users (Q4) showed little to no extra boost, revealing that the New Year's resolution effect primarily motivates people who weren't already very active to get moving.January New Year activity spike statistical significance: regression difference of activity levels on and after January 1st versus expected baseline https://tryterra.co/research/research-article-global-sleep-patterns2026-09-08T09:43:59.324Zweekly0.9https://loved-hug-5d84552294.media.strapiapp.com/GSP_pic1_1da25c4c79.pngSleep Patterns Across the World https://loved-hug-5d84552294.media.strapiapp.com/GSP_pic3_abd20e6802.pngAverage nightly sleep duration by country across wearable users in October 2025.Global sleep patterns comparison across countries showing average nightly sleep duration ranked from 8,298 wearable-tracked nights in October 2025https://loved-hug-5d84552294.media.strapiapp.com/GSP_pic2_384b156b9e.pngAverage bedtime and wake time by country from October 2025 wearable data.Global sleep timing comparison showing average bedtime and wake time across countries in North America, Europe, and beyond from wearable datahttps://loved-hug-5d84552294.media.strapiapp.com/GSP_pic4_c1c8bc345a.pngSleep efficiency and stage composition by country, October 2025.Global sleep quality comparison across countries showing sleep efficiency and stage composition drawn from 8,298 wearable-tracked nights https://tryterra.co/research/new-year-resolutions2026-09-08T09:43:58.545Zweekly0.9https://loved-hug-5d84552294.media.strapiapp.com/post_new_year_duration_user_max_948b54de60.pngNew Year’s Resolutions Cliffhttps://loved-hug-5d84552294.media.strapiapp.com/b1pic1_468e8f9a3b.pngFigure 1: Monthly Total Sleep Time and Sleep Latency: The left panel displays average total sleep time per month with interquartile ranges, showing a small but consistent January peak. The right panel highlights sleep latency's skewed distribution, where January's higher mean contrasts with a more typical median, indicating many struggled to fall asleep despite extra time in bed. So even when we try to get more sleep, nature is against us!New Year sleep resolution effect: monthly average sleep time and sleep latency across 60,000+ wearable users showing a small but consistent January peakhttps://loved-hug-5d84552294.media.strapiapp.com/b1pic2_37c52e4769.pngFigure 2: Individual-Level Significance of January 2 Periods: This two-panel breakdown compares each user's January 2 improvement period to their typical periods elsewhere in Q1, using 75th- (left) and 90th- (right) percentile thresholds. It reveals that 60% of users experienced a meaningfully standout January 2 effort—either longer-lasting, larger in magnitude, or both—providing strong individual-level evidence of resolution behaviour.Individual-level significance of the January 2 sleep bump compared to each user’s typical Q1 sleep, measured against 75th-percentile baselineshttps://loved-hug-5d84552294.media.strapiapp.com/b1pic3_3d7dc6c750.pngFigure 3: Sleep Trajectory for January 2 Starters: The blue line tracks average daily sleep for January 2 starters relative to their personal baselines (dashed grey), with shaded areas indicating gains or losses. It illustrates an enthusiastic early boost that steadily declines, returning to baseline within days and dipping slightly below it by late January.Sleep trajectory of January 2 resolution starters showing minutes gained above personal baseline in the days following the resolution starthttps://loved-hug-5d84552294.media.strapiapp.com/b1pic4_63b977e969.pngFigure 4: Duration of January 2 Sleep Increases: This bar chart counts how many users maintained a 5%+ increase above their personal baseline starting January 2 (capped at January 10). It dramatically shows the heavy skew toward brevity, with most periods ending after just 1–2 days.How long New Year sleep resolutions last: how many users sustained a 5%+ sleep increase from January 2 onward, dropping off sharply within dayshttps://loved-hug-5d84552294.media.strapiapp.com/b1pic5_0c00153b77.pngFigure 5: Magnitude vs. Duration of January 2 Increases: Each point represents one user's January 2 period, plotting extra minutes slept (magnitude) against days maintained (duration), with a regression trend line. The dense bottom-left cluster highlights the most achieving small, short gains, while the sparser top-right points show rare, sustained, larger improvements.Magnitude versus duration of the January 2 sleep bump: each user plotted by extra minutes slept against how many days they maintained the improvementhttps://loved-hug-5d84552294.media.strapiapp.com/b1pic6_a9272cd9c1.pngFigure 6: Weekend vs. Weekday Differences Over Time: The top panel details Friday/Saturday advantages weekly; the bottom tracks trends (imputed values marked). Weekends consistently delivered extra sleep, with January—especially early Saturdays—showing the strongest premium, supporting post-holiday catch-up efforts.Weekend versus weekday sleep gap over the year showing Fridays and Saturdays consistently deliver extra sleep compared to weeknights https://tryterra.co/research/how-hrv-actually-works2026-09-08T09:43:58.040Zweekly0.9https://loved-hug-5d84552294.media.strapiapp.com/hrv_all_distributions_kde_93day_f4864421a4.pngHow HRV Actually Works https://loved-hug-5d84552294.media.strapiapp.com/hrv_all_distributions_kde_93day_f4864421a4.pngFigure 1: The different distributions of all the HRV data from each variable across the period. We can see that HRV usually hangs around 50ms, before gradually falling off. The differences in deviations are looked into in further detail below.HRV distribution by wearable across a 93-day period showing values centre around 50ms on Apple Watch, Fitbit, Garmin, and Oura, with differing spread and tailshttps://loved-hug-5d84552294.media.strapiapp.com/daily_hrv_trends_combined_4hour_filter_93day_225da54508.pngFigure 2: The populations HRV trend over the three months. You can see the weekly and general (seasonal trend). Ignore Fitbit’s “shaky” start, this is because they had few users for the first few weeks.Daily HRV trends over 93 days across wearables showing Oura’s strong weekly cyclicality, Garmin’s low-variability steadiness, and Apple’s smoother but more scattered curve https://tryterra.co/research/sleep-tracking-accuracy-adherence-analysis2026-09-08T09:43:57.178Zweekly0.9https://loved-hug-5d84552294.media.strapiapp.com/image_1_4ae0732a51.pngHow Do People Really Use Wearables? https://loved-hug-5d84552294.media.strapiapp.com/Gemini_Generated_Image_4v8l634v8l634v8l_7ce4d2d3ae.pngFigure 1: Fitbit users are also highly compliant. Apple shows more spread, with many high-compliance users but also a large low-compliance group. Garmin trails with the lowest overall compliance.Wearable adherence comparison across Fitbit, Apple Watch, and Garmin showing Fitbit users are the most compliant and Garmin trails with the lowest overall wear ratehttps://loved-hug-5d84552294.media.strapiapp.com/output_11_1_bd0443e287.pngFigure 2: Daily rhythms of wearable sleep tracking across the week.Weekly wearable usage patterns showing Thursday as the highest-adherence night and Saturday as the lowest across Fitbit, Apple Watch, and Garmin users https://tryterra.co/research/sleep-tracking-accuracy2026-09-08T09:43:56.370Zweekly0.9https://loved-hug-5d84552294.media.strapiapp.com/output_6_3d52da6a1c.pngBenchmarking Wearable Sleep Data Reliabilityhttps://loved-hug-5d84552294.media.strapiapp.com/output_6_3d52da6a1c.pngFigure 1: Average Total Sleep Time with Variability The mean total sleep time (TST) recorded by each platform, with error bars representing variability (standard deviation). Garmin users averaged the longest sleep at just over 7 hours, while Oura recorded the shortest at 6.3 hours. Apple and Fitbit fell in between. Garmin also had the most consistent readings (tight error bars), whereas Apple and Oura showed greater variability. These differences reflect both device algorithms and the inclusion of naps or multiple short events in some platforms.Wearable sleep tracking accuracy: Garmin records the longest average nightly sleep at 7.17 hours while Oura records the shortest at 6.34 hours across 5,000+ nightshttps://loved-hug-5d84552294.media.strapiapp.com/output_7_9d1322e29a.pngFigure 2: Sleep Stage Composition by Platform (≥4h episodes) The proportion of REM, Light, and Deep sleep detected by each wearable. The totals should sum to ~100% of sleep, but Apple consistently under-reports by around 10%, leaving some sleep time unclassified. Across the other platforms, REM sleep is steady (16–22%), Light sleep hovers around 60–66%, and Deep sleep sits at ~18%. Apple stands out with markedly lower Deep sleep (10.5%), highlighting a fundamental difference in how its algorithm stages sleep.Sleep stage composition by wearable: proportion of REM, light, and deep sleep detected by Apple Watch, Fitbit, Garmin, and Oura, with Apple consistently under-summing to less than 100%https://loved-hug-5d84552294.media.strapiapp.com/output_8_5e53c44f19.pngFigure 3: Data Exclusions (<4h) and Outlier Rates by PlatformThis compares the percentage of records excluded as short sleep (<4 hours) against the percentage of outliers detected after filtering. Garmin excluded the largest share of records (16.3%), but produced very few outliers afterwards (~1.4%). Oura and Fitbit both identified naps explicitly and had moderate exclusion rates (12.7% and 9.1% respectively), while Apple recorded virtually no naps (<1% exclusions) but suffered the highest outlier rates (up to 4.3%). Garmin and Apple do not detect naps, which partly explains why Garmin required more exclusions (short fragments that had to be removed manually), and why Apple’s dataset retained implausible values instead.Sleep tracking data exclusions and outlier rates by wearable showing which platforms most often drop short-sleep records or flag anomalies after filtering https://tryterra.co/research/is-consistency-actually-more-important-than-intensity2026-09-08T09:43:55.539Zweekly0.9https://loved-hug-5d84552294.media.strapiapp.com/chart2_weekly_pattern_f7df2e7b18.pngIs Consistency Actually More Important Than Intensity?https://loved-hug-5d84552294.media.strapiapp.com/chart3_activity_scores_a07d02a43c.pngFigure 1: Activity Scores by Profile. Wearable Activity Scores increase with step volume across all energy levels. High-step users achieve scores of 87-92, compared to 74-79 for medium-step and 62-70 for low-step users, regardless of active energy intensity.Wearable activity scores by step and energy profile: high-step users hit 87–92 while low-step users sit at 62–70 regardless of energy expenditurehttps://loved-hug-5d84552294.media.strapiapp.com/chart1_steps_consistency_8b295dc9a1.pngFigure 2: Steps & Consistency Comparison. High-step users average 70% more daily steps (11,537 vs 6,761) and show significantly lower variability in their movement patterns (CV: 36.4% vs 54.9%), demonstrating the power of consistent daily activity.Daily step consistency comparison: high-step users average 11,537 daily steps with 36% variability, versus 6,761 steps and 55% variability for the resthttps://loved-hug-5d84552294.media.strapiapp.com/chart2_weekly_pattern_f7df2e7b18.pngFigure 3: Weekly Step Pattern. High-step users maintain over 10,600 steps daily throughout the week, with weekend days (especially Sunday at 12,810 steps) showing increased activity while weekday consistency remains strong.Weekly step pattern of high-step users staying above 10,600 steps every day of the week, peaking at 12,810 on Sundays https://tryterra.co/research/which-activity-metrics-actually-matter-28-000-days-analysed2026-09-08T09:43:54.511Zweekly0.9https://loved-hug-5d84552294.media.strapiapp.com/weekly_trends_line_plot_5783d3fa2a.pngWhich activity metrics actually matter? 28,000+ days analysedhttps://loved-hug-5d84552294.media.strapiapp.com/variable_distributions_7721e72531.pngFigure 1: Distributions of Sleep Activity metricsSmart ring activity metrics compared: distribution shapes of steps, active seconds, calories, MET minutes, and activity score across 28,000+ days of user datahttps://loved-hug-5d84552294.media.strapiapp.com/weekly_trends_line_plot_5783d3fa2a.pngFigure 2: Weekly trends in general activity metricsWeekly rhythms in wearable activity metrics showing coordinated weekday-to-weekend patterns across steps, active seconds, calories, and MET minutes https://tryterra.co/research/burn-more-calories-sleep-like-a-log2026-09-08T09:43:53.648Zweekly0.9https://loved-hug-5d84552294.media.strapiapp.com/blog_5_graph_1_9911c19473.pngBurn More Calories, Sleep Like a Log? https://loved-hug-5d84552294.media.strapiapp.com/blog_5_graph_1_9911c19473.pngFigure 1: Matrix of sleep duration vs activity levels.Does exercise improve sleep: matrix of sleep duration against daily activity levels across 1,000+ smart ring users showing no strong population-level relationshiphttps://loved-hug-5d84552294.media.strapiapp.com/blog_5_graph_2_37ce067d4c.pngFigure 2: Person-level correlations between average calories burned and sleep variables. Heart rate metrics show weak negative correlations (r = -0.164 to -0.172), while HRV shows a weak positive correlation (r = 0.166) across 341 participants (all p < 0.05).Person-level correlations between calories burned and sleep metrics, showing weak negative links to resting heart rate and a small positive link to HRV rather than to sleep duration https://tryterra.co/research/gamm-analysis-can-complexity-tell-us-anything-we-don-t-already-know2026-09-08T09:43:52.612Zweekly0.9https://loved-hug-5d84552294.media.strapiapp.com/gamm_1_b21a6d8ee4.pngGAMM analysis – can complexity tell us anything we don’t already know?https://loved-hug-5d84552294.media.strapiapp.com/gamm_1_b21a6d8ee4.pngFigure 1: The wild spread of correlations; only 15% of users (dark shading) reach within-person significance.Person-level sleep and HRV correlations from GAMM analysis showing only 15% of individuals reach a meaningful within-person link between sleep duration and next-day HRVhttps://loved-hug-5d84552294.media.strapiapp.com/gamm_2_9cb403a558.pngFigure 2: Top Predictive Factors: Distribution shape factors (kurtosis, skew) are most important for prediction., Strong relationship predictors include: Sleep CV, delta kurtosis, HRV CV, sleep range, and mean sleepTop predictors of the personal sleep–HRV link: distribution shape factors such as HRV variability, sleep variability, and kurtosis matter far more than average sleep duration https://tryterra.co/research/think-a-good-hrv-score-follows-a-good-night-sleep-think-again2026-09-08T09:43:52.139Zweekly0.9https://loved-hug-5d84552294.media.strapiapp.com/hrv_graph_1_6c5125694b.pngThink a good HRV score follows a good night’s sleep? Think again!https://loved-hug-5d84552294.media.strapiapp.com/hrv_graph_1_6c5125694b.pngFigure 1: A big blob of a scatter plot.Sleep duration versus next-day HRV Z-score scatter across nearly 90,000 nights forming a dense cloud with no visible relationshiphttps://loved-hug-5d84552294.media.strapiapp.com/hrv_graph_2_0922083a3c.pngFigure 2: Do trends fare any better?Sleep and HRV trend correlation showing even smoothed multi-day trends produce essentially no relationship between sleep duration and next-day HRV https://tryterra.co/research/descriptive-hrv-using-z-scores2026-09-08T09:43:50.628Zweekly0.9https://loved-hug-5d84552294.media.strapiapp.com/image_59dc136ad4.pngDescriptive HRV using Z scoreshttps://loved-hug-5d84552294.media.strapiapp.com/image_59dc136ad4.pngFigure 1: Z-Score Violation Severity Distribution and by weekdayHRV Z-score distribution across ~100,000 records showing 48.5% positive and 51.4% negative daily changes, with severity broken down by weekdayhttps://loved-hug-5d84552294.media.strapiapp.com/image_2_611efc6105.pngFigure 2: Z-Score violation breakdown by frequencyWeekly HRV pattern showing large positive HRV swings cluster early in the week while large negative swings peak on weekends https://tryterra.co/research/why-mondays-might-be-your-healthiest-day2026-09-08T09:43:49.882Zweekly0.9https://loved-hug-5d84552294.media.strapiapp.com/chart_2_e39825b9bd.pngWhy Mondays Might Be Your Healthiest Day: What 1,000+ Sleepers Taught Us About Recoveryhttps://loved-hug-5d84552294.media.strapiapp.com/chart_1_0412fdb833.pngFigure 1: Sleep and HRV Weekly TrendsWeekly sleep and HRV patterns across 1,000+ wearable users showing HRV declines through the week and hits its lowest values by the weekendhttps://loved-hug-5d84552294.media.strapiapp.com/chart_2_e39825b9bd.pngFigure 2: Number of Users with Highest Sleep, HRV, or Delta HRV by DayBest and worst days for sleep and HRV recovery: how many users hit their weekly high and low for sleep, HRV, and HRV delta on each dayhttps://loved-hug-5d84552294.media.strapiapp.com/chart_3_261e08a5b0.pngFigure 3: 28-Day Trend of Sleep and HRV Metrics28-day rolling trend of sleep and HRV showing repeating weekly cycles of recovery early in the week and accumulated load by weekend