Industries / Media, arts and recreation / Fitness and gyms
Tracking progress
Tracking progress is measuring what training does to the body: fitness tests, heart rate, steps, calories, lifting strength and movement, by hand or by device.
7
Milestones achieved, from 1968 to 2020.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1968
Twelve minutes on a track
Kenneth H. Cooper of the US Air Force reported in JAMA in January 1968 that a 12-minute running test correlated at 0.897 with maximal oxygen intake measured on a treadmill in 115 Air Force officers and airmen. Maximal oxygen intake (VO2max) is the most oxygen the body can use in a minute, the standard laboratory measure of aerobic fitness. The test let a coach estimate it with a track and a stopwatch. It is still in use: Samsung's running coach for its Galaxy Watch8 begins with a 12-minute run and sorts the runner into one of 10 levels.
Credit
Kenneth H. Cooper, US Air Force
1983
Heart rate without wires
Light heart rate monitors that send the heartbeat by radio from skin electrodes to a receiver have been available since 1983, according to a 1998 review by two researchers at Polar Electro, the Finnish company that made them and that dates its own start to 1977. A 2003 review by J. Achten and A. E. Jeukendrup, outside the company, found that over the previous 20 years such monitors had become a widely used training aid, mainly to set the intensity of a session. It also found that estimates of VO2max from heart rate can be up to 20% off, and that calorie estimates from heart rate are fair for a group but not accurate for one person.
Lessona reading that is accurate on average can still be far off for the person wearing it.
Credit
Polar Electro Oy, Kempele, Finland
2010
Bar speed shows the load
J. J. González-Badillo and L. Sánchez-Medina of Pablo de Olavide University in Seville tested 120 strength-trained men on the bench press and found that the bar's average speed while pushing tracked the load as a share of each man's one-repetition maximum almost exactly (R² = 0.98). The link held after six weeks of training, though maximum strength rose 9.3%. A lifter could now tell how hard a load was from the first repetition, without a maximum test, which is the basis of velocity-based training. A 2022 systematic review from outside the group, of 25 studies with 842 people, 16 exercises and 12 speed-measuring devices, found that the accuracy of such estimates depends on the exercise, the number of loads, the speed measure and the device.
Credit
J. J. González-Badillo and L. Sánchez-Medina
2017
Wrist calorie counts miss
Anna Shcherbina, Euan Ashley and colleagues at Stanford University tested seven wrist devices, including the Apple Watch, Fitbit Surge and Samsung Gear S2, on 60 volunteers while measuring heart rate by electrocardiogram and energy use by indirect calorimetry (measuring the oxygen a person breathes in and the carbon dioxide breathed out). Six of the seven measured heart rate during cycling with a median error under 5%. None estimated energy use with an error under 20%, and errors were larger for men, for people with a higher body mass index and for darker skin.
Lessona device can count heartbeats well and still guess calories badly.
Credit
Anna Shcherbina, Euan Ashley and colleagues, Stanford University
2018
A heatmap shows bases
In November 2017 Strava, an app where people record and share runs and rides, published a global heatmap of 1 billion activities, more than 3 trillion GPS points, recorded through September 2017. In January 2018 Nathan Ruser, an analyst with the Institute for United Conflict Analysts, pointed out that it showed US military bases and soldiers' regular jogging routes in places such as Helmand province in Afghanistan. Strava suggested that military users opt out of the map.
Lessonlocation data pooled from many people and stripped of names can still show places someone needs to keep hidden.
Credit
Strava; Nathan Ruser
2019
The 10,000-step goal tested
I. M. Lee and colleagues at Brigham and Women's Hospital and Harvard Medical School followed 16,741 women, average age 72, who wore step counters for a week, for an average of 4.3 years. Death rates fell as daily steps rose, up to about 7,500 steps a day, and then leveled off. Women averaging about 4,400 steps had lower death rates than those averaging about 2,700. The authors write that the 10,000-step goal likely comes from the name of a pedometer sold in Japan in 1965, the manpo-kei, or "10,000 steps meter".
Lessona round number taken from a product name became a health target before anyone tested it.
Credit
I. M. Lee, E. J. Shiroma and colleagues, Brigham and Women's Hospital and Harvard Medical School
2020
Phone cameras find joints
Valentin Bazarevsky and colleagues at Google described BlazePose, a model that finds 33 points on a person's body in video at more than 30 frames a second on a Pixel 2 phone, aimed at uses such as fitness tracking. It is used through Google's MediaPipe software. In 2026 a team at the University of Aveiro in Portugal used MediaPipe to count repetitions of 12 exercises, from squats and deadlifts to burpees, in 1,513 videos, and matched hand counts 97.2% of the time.
Credit
Valentin Bazarevsky, Matthias Grundmann and colleagues, Google
Watch for
A chat coach in Fitbit
Google opened a public preview of a personal health coach "built with Gemini" for Fitbit Premium users in the United States, first on Android. It sets goals, suggests weekly and multi-week plans built around the user's equipment and recent activity, and changes the schedule after setbacks such as an injury. Google says it checked the coach with more than 1 million human ratings and more than 100,000 hours of human evaluation by generalists and specialists in sports, sleep, medicine and behavior. It reports no accuracy figures and no outside evaluation.
Achieved whena group outside Google publishes a test of the coach's plans and advice against expert coaches or published guidelines, or a controlled study of users' fitness.
Claimed by
Google, October 2025
Open
01
Calories for a weight session
An open calculator that turns a gym log (exercise, sets, repetitions and load) plus height, age and body composition into an energy estimate with an error range, built from published laboratory measurements, and a public comparison of its error with that of watches.
The briefCredit · unclaimed02
A rep-count scorecard for watches
A filmed, repeatable test of the automatic repetition counting in current watches. Lifters wear watches through a fixed list of sets at light to moderate loads; each watch's count is compared with the count from video, by watch, exercise and software version, and the test is rerun after each major software update.
The briefCredit · unclaimed
Open 01 · the brief
Calories for a weight session
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An open calculator that turns a gym log (exercise, sets, repetitions and load) plus height, age and body composition into an energy estimate with an error range, built from published laboratory measurements, and a public comparison of its error with that of watches.
Why it is still open
In a 2026 test of four watches (Apple Watch Series 8, Galaxy Watch 5, Fitbit Charge 5 and Garmin Vivosmart 5) on 70 men doing bench press, row, squat and deadlift, heart rate was close to the electrocardiogram but calorie estimates hardly agreed with indirect calorimetry at all (intraclass correlation 0.021 to 0.033, where 1 is perfect). A 2019 equation predicts the calories of a machine-weights session from the log and body composition (reported fit 0.773, standard error 28.5 kcal), but it was built on 52 people in one lab, and none of the 13 papers that cite it in Europe PMC tests it on other data. A 2024 review of 166 studies found that reported energy costs of resistance exercise varied widely.
Built with what exists
The study list of the 2024 scoping review and the 2024 Adult Compendium of Physical Activities as data sources; an AI model such as Claude to pull values out of papers, checked by hand; Python for the calculator and the error analysis.
First step
Take the 2024 review's list of studies, enter the energy values from the first 20 into a spreadsheet, and run the 2019 equation on them.
Done looks like
A public table of at least 100 published measurements with their methods; on studies held out from building it, the calculator's median error per session reported beside the watches' published errors; code and table public.
Open 02 · the brief
A rep-count scorecard for watches
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A filmed, repeatable test of the automatic repetition counting in current watches. Lifters wear watches through a fixed list of sets at light to moderate loads; each watch's count is compared with the count from video, by watch, exercise and software version, and the test is rerun after each major software update.
Why it is still open
The peer-reviewed checks found are narrow. A 2021 study at the Swiss Federal Institute of Sport Magglingen of one Apple Watch app found counts accurate for squat and deadlift but poor for bench press. A 2023 conference abstract on four Garmin watches during light dumbbell circuits found none valid for front squats, lunges, push-ups or shoulder presses (errors up to 67.5%) and concluded that lifters should count by hand. No cross-brand comparison against video was found; buyers' guides give figures without a method.
Built with what exists
Two phones on tripods to film each set; watches whose makers list automatic rep counting; open pose software such as MediaPipe to count repetitions from the video as a cross-check on a person counting; a spreadsheet.
First step
Borrow two watches, wear one on each wrist through 30 sets of your own training, swap wrists the next session, and score them against your own video.
Done looks like
At least 20 lifters, 10 exercises and 3 sets each, with every watch's count set against the video count; mean absolute error per watch and exercise; counts, code and consented videos public; a rerun within a month of each major software update.