GPS Accuracy & Watch Data: Understanding Your Running Metrics

The Global Positioning System works through trilateration: your watch receives timing signals from multiple satellites orbiting at approximately 20,200 km altitude, and by calculating the tiny differences in signal arrival time, it determines its position in three-dimensional space. Each satellite broadcasts its precise orbital position and the exact time the signal was transmitted, using atomic clocks accurate to nanoseconds. Your watch compares the received time against its own internal clock to compute the distance to each satellite. With signals from three satellites, the receiver can determine latitude, longitude, and altitude. A fourth satellite is required to correct for clock error in the watch's far less precise quartz oscillator — without this correction, even a one-microsecond timing error translates to 300 meters of position error.

Modern GPS watches do not rely on the American GPS constellation alone. They receive signals from multiple Global Navigation Satellite Systems (GNSS): the US GPS (31 satellites), Russia's GLONASS (24 satellites), the European Union's Galileo (30 satellites), and China's BeiDou (35+ satellites). Using multiple constellations increases the number of satellites visible at any given time, which improves position accuracy and reduces the time to first fix. Most current running watches can track two or three constellations simultaneously. Satellite-Based Augmentation Systems (SBAS) such as WAAS (North America) and EGNOS (Europe) provide additional correction signals from geostationary satellites, refining accuracy from the typical 3-5 meter civilian GPS error down to 1-2 meters in ideal conditions.

The signals themselves travel on specific radio frequencies, and this is where the distinction between single-band and multi-band receivers becomes important. Traditional consumer GPS uses the L1 band (1575.42 MHz). Newer multi-band receivers add the L5 band (1176.45 MHz), which was designed with a wider bandwidth and stronger signal structure specifically for improved accuracy. The L5 signal is more resistant to multipath interference — the phenomenon where signals bounce off buildings, trees, and terrain before reaching the receiver, creating phantom position readings. When a watch processes both L1 and L5 signals simultaneously, it can cross-reference them to identify and reject multipath-corrupted readings, dramatically improving accuracy in challenging environments.

Despite these advances, fundamental physical limitations constrain what any wrist-worn GPS receiver can achieve. The antenna is tiny — typically 15-20mm — compared to survey-grade equipment with 100mm+ antennas. The antenna sits on one side of your wrist, meaning your body blocks satellite signals from certain directions, creating asymmetric reception. The receiver's power budget is severely constrained by battery life requirements, limiting the computational complexity of position solutions. And atmospheric conditions — ionospheric electron density, tropospheric water vapor — introduce variable signal delays that even multi-band processing cannot fully eliminate. Understanding these inherent constraints sets realistic expectations: your watch is an extraordinary piece of engineering, but it is solving an incredibly difficult physics problem with a device that weighs 50 grams and runs on a battery smaller than a coin.

Wundersitz and colleagues (2020) conducted one of the most comprehensive assessments of consumer GPS watch accuracy, testing eight devices across three distinct environments: open fields with unobstructed sky view, tree-lined paths with partial canopy coverage, and urban canyons with tall buildings on both sides. The results quantified what every runner intuitively suspects — environment matters enormously. In open conditions, the best devices achieved distance errors below 1%, with position accuracy of 1-2 meters. Under moderate tree cover, errors increased to 2-4% for distance and 3-8 meters for position. In urban canyons, the worst-case errors reached 8.9% for distance, meaning a watch could display 5.45 km for a true 5.00 km course. The study also found significant variation between devices in the same environment, confirming that hardware and software differences between brands produce meaningfully different results.

Multipath interference is the dominant error source in non-open environments. When GPS signals reflect off buildings, water, or dense foliage before reaching the watch antenna, the receiver calculates an incorrect distance to the satellite because the reflected signal traveled a longer path. In a narrow urban street flanked by glass-fronted buildings, a single satellite's signal might arrive via three or more paths, each suggesting a different position. The receiver must select or average among these conflicting inputs, and even sophisticated algorithms cannot always identify the direct-path signal. The Fellrnr comparative GPS dataset, comprising over 12,000 miles of runs across multiple watches and environments, provides extensive empirical evidence of this effect: systematic position offsets occur consistently in the same locations, demonstrating that multipath errors are environmental features, not random noise.

Cold start versus warm start significantly affects the first minutes of a run. A cold start occurs when the watch has no recent satellite almanac data — perhaps after a firmware update, a long period of inactivity, or travel to a new region. The receiver must search for and acquire satellites from scratch, which can take 30-120 seconds during which position accuracy is severely degraded. A warm start, where the watch has recent almanac data and approximate position from its last fix, typically achieves a lock in 5-15 seconds with better initial accuracy. Most modern watches use assisted GPS (A-GPS), downloading predicted satellite positions via Bluetooth from your phone, which dramatically speeds acquisition. The practical advice is simple: wait for a solid satellite lock before starting your run, and allow 30-60 seconds of standing still after the watch reports a fix to let the position solution stabilize.

GPS Accuracy by Environment

Body position introduces a subtle but consistent bias. Because the watch sits on one wrist, your body blocks satellite signals from roughly half the sky at any given moment. When you run north, you block signals from southern satellites; when you turn, the blocked satellites change. This creates a systematic position wobble synchronized with arm swing and turns. Some watches compensate by using accelerometer data to model arm motion, but the correction is imperfect. Runners who wear their watch on the left wrist will see slightly different GPS tracks than those wearing it on the right, particularly on courses with many turns. For racing, this effect is negligible compared to environmental factors, but it contributes to the overall error budget and explains why two identical watches on different wrists can disagree on distance by 0.5-1%.

The GPS chipset inside your watch determines the fundamental ceiling of its positioning accuracy, and the industry has undergone three distinct generations in the past decade. From approximately 2018 to 2022, most premium running watches — including Garmin Forerunner and Fenix series, COROS PACE and VERTIX, and Suunto models — used the Sony CXD5603GF chipset. This single-band L1 receiver represented a significant improvement over earlier chips in power efficiency and time-to-first-fix, enabling the all-day battery life that modern GPS watches offer. However, it was fundamentally limited to L1 signals, meaning it had no defense against multipath interference beyond algorithmic filtering. In open environments, the Sony chipset performed well. In cities and under tree cover, its limitations were evident in the characteristic wandering GPS tracks that runners learned to accept.

The transition to the Airoha AG3335M chipset in 2022-2024 marked the arrival of multi-band GNSS in consumer running watches. Garmin's Forerunner 265/965, Fenix 7X, and Enduro 2, along with COROS VERTIX 2S and APEX 2, adopted this chip. The Airoha chip receives both L1 and L5 signals simultaneously, enabling the multipath rejection described earlier. Real-world testing by independent reviewers, including the DC Rainmaker and Fellrnr comparative datasets, showed approximately 50% improvement in urban accuracy compared to equivalent Sony-chipset models. A run through downtown that might show 6% distance error with L1-only dropped to 3% with dual-band. Trail runs under moderate canopy improved from 4% to 2% error. The improvement was most dramatic precisely where runners needed it most: in the challenging environments where single-band GPS struggled.

The latest generation, beginning in 2024, features the Synaptics SYN4778 (previously Broadcom BCM4778) and similar next-generation chipsets. These chips further improve multi-band performance while reducing power consumption, addressing the primary trade-off of multi-band GPS: battery life. Early multi-band implementations consumed 30-50% more power than single-band mode, forcing runners to choose between accuracy and battery duration. The new chipsets narrow this gap considerably, making multi-band GPS practical for ultra-distance events where 20+ hour battery life is necessary. Additionally, these chipsets support more satellite signals per constellation and implement more sophisticated signal processing algorithms, including machine learning-based multipath detection that adapts to the specific environment rather than applying generic filtering.

For runners choosing a new watch, the chipset generation is one of the most impactful specifications — often more so than headline features like display type or training metrics. If you frequently run in cities, on wooded trails, or in mountainous terrain, a multi-band capable watch will produce meaningfully better distance and pace data than even a flagship model from two years ago using a single-band chipset. The practical difference is not academic: on a 10 km urban run, the difference between 6% and 3% error is 600 versus 300 meters of phantom distance — enough to significantly distort your pace calculations and training analysis. When comparing watches, look for explicit mention of multi-band, dual-frequency, or L1+L5 support, and check whether multi-band mode is the default or must be manually enabled at the cost of battery life.

Optical heart rate sensing works through photoplethysmography (PPG): LEDs on the back of the watch illuminate the skin, and photodiodes measure how much light is absorbed or reflected. With each heartbeat, blood volume in the capillaries beneath the skin pulses, changing the light absorption pattern. The sensor detects these micro-variations and extracts heart rate by identifying the dominant frequency in the signal. Modern watches use green LEDs (approximately 525 nm wavelength) during exercise because green light is well-absorbed by hemoglobin and provides good signal contrast during motion. For resting measurements and continuous all-day monitoring, infrared LEDs (approximately 940 nm) are used because they penetrate deeper into tissue and consume less power, though they are more susceptible to motion artifact.

Gillinov and colleagues (2017) conducted one of the most rigorous clinical assessments of optical HR accuracy, testing four consumer devices against a reference electrocardiogram (ECG) in a hospital setting across multiple exercise types. The study revealed that accuracy is highly activity-dependent: during treadmill running, the best devices achieved 2-5% error rates, while cycling and rowing produced errors of 10-20%. The critical variable was motion artifact — rhythmic wrist movements during cycling and rowing created periodic light intensity changes at frequencies that the PPG algorithm misinterpreted as heartbeats. Running, despite the arm swing, produces less problematic motion patterns because the wrist acceleration during running is less synchronized with the heart rate frequency range than the repetitive cycling cadence. Pasadyn and colleagues (2019) found even stronger results specifically for Apple Watch during treadmill running, achieving a correlation coefficient of r = 0.96 with a chest strap — excellent agreement for a wrist-worn device.

Several physical factors affect optical HR accuracy beyond the activity type. Watch fit is paramount: a loose watch allows ambient light to leak under the sensor and permits the watch to shift position, both of which degrade signal quality. The watch should sit snugly approximately one finger-width above the wrist bone, tight enough to stay in place but not so tight as to restrict blood flow. Skin tone affects signal strength because melanin absorbs green light, reducing the depth of the pulsatile signal in darker skin. While manufacturers have improved algorithms and increased LED power to mitigate this, studies including Bent et al. (2020) have documented reduced accuracy in individuals with darker skin tones, particularly during high-intensity exercise. Cold temperatures cause peripheral vasoconstriction, reducing blood flow to the wrist and weakening the PPG signal — winter runners commonly report HR dropouts or anomalous readings during cold starts.

Optical HR Accuracy by Activity Type

The practical question for runners is when to trust wrist-based HR and when to use a chest strap. For steady-state running — easy runs, long runs, tempo efforts at consistent pace — modern wrist-based sensors are accurate enough for training zone management. The 3-7% error in outdoor running translates to approximately 3-10 bpm for most heart rate ranges, which is acceptable for zone-based training where zones span 10-15 bpm. For interval training with rapid HR changes, wrist sensors introduce a lag of 5-15 seconds in detecting HR spikes and drops, which can misrepresent the actual intensity of short intervals. For lactate threshold testing, drift testing, or any protocol where precise HR data is critical, a chest strap (ECG-based) remains the gold standard. Chest straps like the Polar H10 achieve correlation coefficients above 0.99 with clinical ECG and have effectively zero motion artifact during running, making them the appropriate tool when accuracy cannot be compromised.

Your watch calculates distance by recording a series of GPS positions over time and computing the total length of the polyline connecting them. At the typical 1 Hz sampling rate (one position per second), a runner moving at 5:00/km pace generates a position fix every 3.3 meters. The watch draws straight lines between consecutive fixes and sums the segments. This approach introduces two competing error sources. First, the straight-line segments cut corners on curves — the true path around a bend is an arc, but the recorded path is a chord. This corner-cutting effect systematically undercounts distance on curvy courses. Second, GPS position jitter — random fluctuations of 1-3 meters in each fix — adds phantom distance on straight segments, because the recorded path zigzags slightly around the true line. On a perfectly straight road, jitter alone can add 1-2% to the recorded distance.

Track running illustrates these errors clearly. A standard 400-meter track has tight curves where corner-cutting reduces recorded distance, but the GPS jitter on the straights adds distance back. The net effect depends on which lane you run in, how your watch handles the curves, and the satellite geometry at the time. Most runners find their watch overcounts track distance by 2-4%: a GPS-recorded 5 km on a track often reads 5.10-5.20 km. This is primarily because the jitter effect on the relatively straight sections dominates the corner-cutting on the curves. Some watches now offer track detection modes that snap the GPS path to the lane geometry, significantly improving accuracy for track workouts, but these features require the watch to correctly identify that you are on a track.

Treadmill distance uses an entirely different measurement approach. With no GPS movement, watches estimate distance from wrist-based accelerometer data — counting steps and estimating stride length from arm swing characteristics. Initial calibration is based on the watch's default stride length model for your height and pace, which may not match your actual biomechanics. Most watches allow manual calibration: run a known distance on a measured course or calibrated treadmill, then adjust the watch's stride length factor. Even after calibration, treadmill distance accuracy typically falls in the 2-5% range because stride length varies with pace, fatigue, and incline. A watch calibrated at your easy pace may overcount distance at your tempo pace if your stride lengthens. Some runners find that running with a foot pod (like Stryd) provides more consistent treadmill distance because the foot-mounted accelerometer captures ground contact dynamics more accurately than wrist motion.

Sampling rate plays a more significant role than most runners realize. The standard 1 Hz rate is a compromise between accuracy and battery life. At 1 Hz, a runner making a sharp 90-degree turn around a corner effectively skips the turn — the positions before and after the corner are connected by a straight line that cuts across the actual path. Higher sampling rates (some watches offer smart recording modes up to 4 Hz in turns) capture more points around curves, reducing corner-cutting error. However, higher sampling rates also amplify the jitter effect on straight sections because each additional noisy position adds potential phantom distance. The optimal strategy varies by environment: on a winding trail course, higher sampling rates improve accuracy; on a straight road, 1 Hz may actually produce a more accurate distance than multi-Hz recording because the jitter averages out over fewer samples.

Instantaneous pace — the number displayed on your watch screen as you run — is perhaps the most frustrating metric for runners because it is derived from the noisiest data your watch collects. Pace is calculated from GPS speed, which is the distance between consecutive position fixes divided by the time interval. At 1 Hz sampling, a 2-meter position error in a single fix translates to a 2 m/s speed error, which at running speeds represents a pace fluctuation of approximately 30-60 seconds per kilometer. Without any smoothing, the pace display would be unusable — jumping wildly between 3:30/km and 7:00/km on what feels like a perfectly steady 5:00/km run. Every watch manufacturer applies smoothing algorithms to tame this noise, but they make very different choices about how aggressively to smooth, and these choices have real consequences for training.

The core trade-off in pace smoothing is responsiveness versus stability. A short averaging window (3-5 seconds of GPS data) produces a pace display that responds quickly to actual speed changes — useful for interval training where you need to hit a specific pace within the first 100 meters of a repeat. However, short averaging windows also let more GPS noise through, making the display jittery during steady running. A long averaging window (15-30 seconds) produces a stable, calm pace display during steady running but lags badly during pace changes — you might be 200 meters into a recovery jog before the watch acknowledges you have slowed down. Garmin, COROS, Suunto, and Apple all make different default choices on this spectrum, which is why the same run can feel smooth on one watch and erratic on another, even when the underlying GPS data is identical.

Environmental factors amplify pace display instability in predictable ways. Running through a tunnel or under a dense bridge overpass causes complete GPS signal loss, during which the watch either freezes the last known pace or switches to accelerometer-based estimation — both of which produce visible discontinuities when satellite tracking resumes. Running alongside a tall building often creates multipath-induced speed errors that manifest as sudden pace spikes or drops lasting 5-10 seconds. Tree canopy causes less dramatic but more persistent pace noise because the degraded position accuracy continuously feeds noisier speed calculations into the smoothing filter. The practical impact is real: a runner doing a threshold workout in a forested park may see their pace display fluctuate by 15-20 seconds per kilometer even while maintaining perfectly steady effort, making pace-based training unreliable in that environment.

Some modern watches address the smoothing problem by combining GPS speed with accelerometer-derived speed in a sensor fusion approach. The accelerometer provides a smooth, responsive speed estimate based on stride frequency and length, while GPS provides the absolute reference that prevents accelerometer drift. By weighting the two sources according to their reliability in the current conditions — relying more heavily on the accelerometer when GPS quality is poor, and more on GPS when signals are strong — the fused pace display can be both responsive and stable. Stryd's foot pod takes this further by providing a ground-truth speed reference from foot-mounted accelerometry that is independent of GPS entirely. Runners who find GPS pace display unreliable in their typical training environments may find that a foot pod provides the most consistent real-time pace feedback, particularly for tempo and interval workouts where pace precision matters.

Running watches determine elevation through two fundamentally different methods, each with distinct error characteristics. GPS-derived altitude uses satellite geometry to calculate the vertical component of your position, but because satellite signals travel through the atmosphere at angles optimized for horizontal positioning, the vertical accuracy is inherently 2-3 times worse than horizontal. A watch with 3-meter horizontal position error typically has 10-20 meter vertical error. This means that GPS-only elevation data is unsuitable for meaningful ascent/descent calculations — on a flat run, random GPS altitude fluctuations would accumulate into hundreds of meters of phantom elevation gain. This is why the elevation profiles from older or cheaper watches without barometric sensors often look like jagged noise rather than smooth terrain profiles.

Barometric altimeters, present in most mid-range and premium running watches, use atmospheric pressure to determine relative altitude changes. The standard atmosphere model defines a pressure decrease of approximately 12 Pa per meter of altitude gain at sea level. Barometric sensors in modern watches can resolve pressure changes of 1-2 Pa, corresponding to altitude resolution of approximately 0.1-0.2 meters — far superior to GPS. For relative altitude change over short periods, barometric accuracy is typically 1-3 meters, making it the clear choice for tracking hills during a run. Garmin, COROS, Suunto, and Apple Watch all use barometric pressure as the primary elevation source when available, with GPS altitude serving as a periodic calibration reference to correct for long-term barometric drift.

The barometric altimeter's Achilles heel is weather-induced pressure change. A passing weather front can alter atmospheric pressure by 200-500 Pa over several hours, equivalent to a 20-40 meter apparent altitude change. During a 3-hour long run, a falling barometer (approaching storm) will add phantom elevation gain because the watch interprets decreasing pressure as ascending altitude. A rising barometer during a run subtracts real elevation gain. Watches partially compensate by using GPS altitude to periodically recalibrate the barometric reference, but this correction is intentionally slow — updating too aggressively would import GPS's 10-20 meter vertical noise into the barometric data. Temperature also affects barometric readings: moving from a warm indoor environment to cold outdoor air changes the pressure at the sensor before atmospheric conditions have time to equilibrate, which is why the first few minutes of elevation data after stepping outside can be unreliable.

The reason two watches on the same run report different total ascent — a near-universal frustration among runners — stems from differences in how each manufacturer processes the raw barometric data. Total ascent is calculated by summing only the positive altitude changes and ignoring the negative ones, but this summation is extremely sensitive to noise filtering. A watch that applies aggressive smoothing will ignore small altitude fluctuations and report lower total ascent. A watch that applies minimal smoothing will capture real micro-terrain (speed bumps, curb steps, mild undulations) but also accumulate barometric noise as phantom ascent. Strava and Garmin Connect apply their own post-processing to the recorded data, which may further modify the total ascent number. The discrepancy between your watch's displayed ascent, Garmin Connect's number, and Strava's number for the same run is not a bug — it is the inevitable consequence of three different algorithms making three different noise-versus-detail trade-off decisions on inherently noisy data.

The VO2 Max number on your watch is not a measurement — it is the output of a mathematical model that infers your maximum oxygen consumption from the relationship between your running speed and heart rate. The dominant algorithm in the industry is developed by Firstbeat Analytics (now part of Garmin), and it powers the VO2 Max estimates on Garmin, Suunto, and numerous other brands. The core principle is elegant: at a given submaximal running speed, a fitter runner will have a lower heart rate because their cardiovascular system delivers oxygen more efficiently. By observing how your heart rate responds across different running speeds and comparing the pattern to a large reference database, the algorithm estimates where your aerobic capacity sits on the fitness spectrum.

Firstbeat's published validation data reports a standard error of estimate of approximately 3.5 ml/kg/min when compared to laboratory treadmill VO2 Max testing. This is a meaningful margin: if your watch displays 50 ml/kg/min, your true laboratory value is statistically likely to fall between 46.5 and 53.5. For a male runner, that range spans from the 50th to the 75th percentile for age 30-39 on the ACSM normative tables — a significant fitness difference. The algorithm performs best during steady-state runs of 10+ minutes at moderate intensity, where the speed-HR relationship is most stable and informative. Short runs, interval workouts, and very easy jogs provide less reliable input data, which is why your estimated VO2 Max may not update after certain workout types.

Multiple confounders can skew your watch's VO2 Max estimate independently of any change in actual fitness. Cardiac drift — the progressive rise in heart rate during prolonged exercise due to dehydration and thermal stress — makes a steady-pace run in heat look like declining fitness to the algorithm, because heart rate is higher than expected for the speed. Caffeine elevates resting and exercising heart rate by 3-5 bpm in some individuals, which the algorithm interprets as lower efficiency. Altitude reduces the partial pressure of oxygen, raising submaximal heart rate and depressing the VO2 Max estimate by approximately 3-5% per 1,000 meters of elevation above sea level. Beta-blockers and other heart rate-lowering medications artificially suppress HR, causing the algorithm to dramatically overestimate VO2 Max. Running on soft surfaces (sand, grass, trails) requires more energy per kilometer than road running, increasing heart rate at a given GPS speed and pushing the estimate downward.

Apple Watch uses a different estimation approach that incorporates data from walking and daily activities, not just dedicated running workouts. This broader data collection means the estimate updates more frequently and may be more stable over time, but it also responds to non-exercise factors like illness, stress, and sleep quality. The ACSM normative tables provide useful context for interpreting your number: for males aged 20-29, the 50th percentile is approximately 44 ml/kg/min; for females in the same age range, approximately 37 ml/kg/min. Competitive recreational runners typically fall in the 50-60 range, while elite marathoners reach 70-85. The most productive way to use your watch's VO2 Max estimate is as a long-term trend indicator rather than an absolute number — a consistent upward trend over months reflects genuine aerobic improvement regardless of whether the absolute value matches what a laboratory would measure.

The central insight for interpreting watch data is distinguishing between metrics that are reliable enough for single-run decisions and those that are only meaningful as long-term trends. Heart rate during steady-state running is in the first category: wrist-based sensors are accurate enough during continuous running that you can trust the displayed HR for zone-based training decisions. Distance on open roads falls into the same reliable category, with errors small enough to be inconsequential for training purposes. Instantaneous pace, elevation gain, and VO2 Max estimates belong firmly in the second category — single-run values carry too much noise for precise interpretation, but multi-week trends reveal genuine physiological changes. Understanding this distinction prevents both under-trusting your watch (ignoring useful real-time HR data) and over-trusting it (agonizing over a 0.02 km discrepancy on your morning loop).

Cross-referencing data sources improves confidence in any individual metric. If your watch shows an unusually fast average pace for a run that felt easy, check the GPS track for obvious shortcuts or wobbles that inflated the speed calculation. If your VO2 Max estimate suddenly drops by 2 points, look at recent runs for confounders — were they in unusual heat, at altitude, or after poor sleep? Platforms like Strava, Garmin Connect, and TrainingPeaks post-process your raw data with their own algorithms, and comparing results between platforms can highlight when a metric is likely distorted by processing choices rather than reflecting reality. Downloading your FIT files and examining the raw sensor data — using a tool like Hashiri.AI's FIT Viewer — provides the most granular view of what your watch actually recorded, before any platform's smoothing or correction algorithms were applied.

For training decisions, the hierarchy of trust should be: perceived effort first, heart rate second, pace third, and derived metrics (VO2 Max, training load, recovery time) as supporting context. Perceived effort integrates information from every system in your body — cardiovascular, musculoskeletal, neurological, metabolic — in a way that no external sensor can replicate. Heart rate provides an objective cardiovascular load measurement that largely agrees with effort perception during steady running. Pace reflects the output of that effort but is contaminated by GPS noise, terrain, wind, and environmental factors. Derived metrics use pace and HR as inputs, which means they inherit and compound the errors from both. A runner whose watch says 5:00/km pace and 150 bpm heart rate can trust the HR more than the pace, and can trust both more than the recovery time estimate that was calculated from them.

The most valuable long-term tracking practices involve controlling variables to maximize the signal-to-noise ratio in your data. Run the same route regularly as a benchmark — a fixed route eliminates GPS course-measurement error, so changes in time or heart rate reflect genuine fitness changes. Perform HR-controlled tests (like the drift test or MAF test) monthly under consistent conditions to get clean physiological data uncorrupted by pacing or environmental variation. Log the conditions of each run — temperature, humidity, sleep quality, fueling — so that outlier data points can be explained rather than misinterpreted. Over months and years, this disciplined approach to data collection produces a rich, reliable picture of your fitness trajectory that is far more valuable than obsessing over whether today's run was 10.02 or 9.98 kilometers.