Evidence behind the numbers

Every rule that comes from research points to one of these entries. The grade says how far it was checked, not how good the study is: A opened and read, B abstract or official record read, C search summary only.

• E01Grade AHeiderscheit 2011

  Heiderscheit BC, Chumanov ES, Michalski MP, Wille CM, Ryan MB. Effects of step rate manipulation on joint mechanics during running. Med Sci Sports Exerc 2011;43(2):296-302.

  Used for: In 45 runners, 5% and 10% faster step rate lowered energy absorbed at the knee; judge cadence against your own preferred cadence.

  Open source
• E02Grade ALenhart 2014

  Lenhart RL, Thelen DG, Wille CM, Chumanov ES, Heiderscheit BC. Increasing running step rate reduces patellofemoral joint forces. Med Sci Sports Exerc 2014;46(3):557-564.

  Used for: A 10% higher step rate cut modelled patellofemoral force by 14%; lower peak stance knee flexion predicted the drop.

  Open source
• E03Grade AMillour 2019

  Millour G, Duc S, Puel F, Bertucci W. Comparison of static and dynamic methods based on knee kinematics to determine optimal saddle height in cycling. Acta Bioeng Biomech 2019;21(4).

  Used for: Knee angle at the bottom of the pedal stroke reads about 8 degrees higher when pedalling than static; 33 to 43 degrees proposed for pedalling.

  Open source
• E04Grade AStevens 2006

  Stevens NJ, Schmitt DO, Cole TM III, Chan L-K. Technical note: Out-of-plane angular correction based on a trigonometric function for use in two-dimensional kinematic studies. Am J Phys Anthropol 2006.

  Used for: Elbow and knee angles within about 20 degrees of the camera plane differ little from 3D; beyond that they can read too extended.

  Open source
• E05Grade AWade 2023

  Wade L, Needham L, Evans M, McGuigan P, Colyer S, Cosker D, Bilzon J. Examination of 2D frontal and sagittal markerless motion capture: Implications for markerless applications. PLoS ONE 2023;18(11):e0293917.

  Used for: Camera-side knee and hip angles were close to marker-based values; far-side spread was 40% to 108% larger; ankle angles were too wide to use.

• E06Grade BDingenen 2018

  Dingenen B, Barton C, Janssen T, Benoit A, Malliaras P. Test-retest reliability of two-dimensional video analysis during running. Phys Ther Sport 2018;33:40-47.

  Used for: Average at least 7 steps; contralateral pelvic drop was only moderately reliable (ICC 0.59 to 0.77, smallest detectable difference about 2.7 degrees).

• E07Grade AHopkins 2004

  Hopkins WG. How to interpret changes in an athletic performance test. Sportscience 2004;8:1-7.

  Used for: Judge a change against both the smallest worthwhile change and the typical error of the test.

  Open source
• E08Grade AEinfalt 2020

  Einfalt M, Lienhart R. Decoupling video and human motion: Towards practical event detection in athlete recordings. CVPR Workshops (CVsports) 2020. arXiv:2004.09776.

  Used for: Simple rules on 2D pose statistics detect events well with little data; a trained sequence model reached 96% F1 at plus or minus 5 ms for jump strides.

  Open source
• E09Grade ASrinivasan 2025

  Srinivasan HK, Mathunny JJ, Devaraj A, Gogoi H, Govindasamy K, Karthik V. Can BlazePose accurately assess joint angles in outdoor running environments? Biomed Hum Kinet 2025;17:13-25.

  Used for: BlazePose agreed with manual 2D measurement within about 2 degrees in 16 runners; the reference was manual, not marker-based 3D.

  Open source
• E10Grade ADPDP Act 2023 s.9

  Digital Personal Data Protection Act, 2023 (India), Sections 2(f) and 9: children, verifiable parental consent, no tracking or behavioural monitoring of children.

  Used for: Learning is opt-in, stays off for under-18s without recorded guardian consent, and shares no identifiers. Not legal advice.

• E11Grade CLeys 2013

  Leys C et al. Detecting outliers: Do not use standard deviation around the mean, use absolute deviation around the median. J Exp Soc Psychol 2013;49(4):764-766. (Authors beyond the first not verified.)

  Used for: Personal baselines use median and MAD so one bad clip cannot distort them.

• E12Grade CMunro 2012

  Munro et al. 2012, reliability of 2D video frontal-plane knee angle (as reported by a search summary; paper not opened).

  Used for: Frontal knee angle from 2D video is repeatable but its agreement with 3D is mixed: a screening aid only.

• E13Grade CMaykut 2015

  Maykut et al. 2015, 2D vs 3D pelvic drop in 24 runners (as reported by a search summary; paper not opened).

  Used for: Pelvic drop from 2D is repeatable but did not correlate with 3D: compare with the athlete’s own history.

  Open source
• E14Grade CJump apps (flight time)

  Validation studies of smartphone jump apps using the flight-time method (as reported by search summaries; papers not opened).

  Used for: Jump height from flight time (h = g t² / 8) agreed well with force plates in several studies; power estimates did not.

• E15Grade CGolf X-factor 2D vs 3D

  A 2D vs 3D comparison of golf trunk kinematics (as reported by a search summary; authors and paper not verified).

  Used for: 2D face-on shoulder-hip separation differed from 3D by about 16 degrees because of projection error: label it as a 2D projection.

• E16Grade CSprint-start review

  Systematic review of sprint-start performance factors, Int J Environ Res Public Health 2022;19(7):4074 (summary only; authors and full text not verified).

  Used for: Provisional reference for set-position knee and hip angles; verify before use.

• E17Grade CPose models in competition

  Studies of OpenPose in a long-jump final and of camera position effects (as reported by search summaries; not opened).

  Used for: Pose angles are estimates; low-confidence frames get no grade and phases can be corrected by hand.

• E18Grade CFrame rate and timing

  Smartphone sprint-timing and gait-event studies (PMC9003053 and others; summaries only).

  Used for: Timing resolution is one frame (1 ÷ fps); higher frame rates reduce timing error.

  Open source
• E19Grade CKeypoint smoothing

  Kinematics smoothing literature (cutoff choice differs by joint and athlete; summaries only).

  Used for: Smoothing windows are set per app and can be tuned; Savitzky-Golay keeps peaks better than a moving average.

• E20Grade CFederated / continual learning

  Surveys of federated learning for human sensing and activity recognition (arXiv:2501.04000 and others; summaries only).

  Used for: Personalisation, concept drift and privacy are open problems, so model updates are gated, versioned and reversible.

  Open source
• E21Grade CWashabaugh 2022

  Washabaugh EP et al. Comparing the accuracy of open-source pose estimation methods for measuring gait kinematics. Gait Posture 2022;97:188-195. Citation confirmed; results not read.

  Used for: Citation only. No number from it is used.

  Open source
• E22Grade AAltman & Davis 2012

  Altman AR, Davis IS. A kinematic method for footstrike pattern detection in barefoot and shod runners. Gait Posture 2012;35(2):298-300.

  Used for: Footstrike angle at initial contact: rearfoot above 8.0 degrees, midfoot between -1.6 and 8.0, forefoot below -1.6 (flat foot = 0).

  Open source
• E23Grade C2D frontal knee angle

  Open Access Journal of Sports Medicine, volume 7: reliability and concurrent validity of 2D versus 3D knee valgus in drop jumps. Authors and pages not read; search summary only.

  Used for: 2D frontal-plane knee angles are reliable but agree only poorly to moderately with 3D, so they are screening only and never graded.

  Open source
• E24Grade CCavedon, Bezodis et al. 2022

  Cavedon V, Bezodis NE, et al. Anthropometry, block settings and set-position joint angles in sprint starts. European Journal of Sport Science 2022. Title and full author list not verified; search summary only.

  Used for: Set-position hip and knee angles depend on where the front block sits (50-70% of leg length in 36 sprinters): closer blocks gave more flexed angles. Set angles are individual, so we report them and do not grade them.

• E25Grade CBissas et al. 2021

  Bissas A et al. Kinematic factors associated with start performance (first ground contact), 2021. University of Gloucestershire repository record 9874; abstract-level reading only.

  Used for: Trunk angle and thigh separation at take-off explained close to 90% of the variation in first-ground-contact horizontal power in elite starters; a forward-leaning trunk with large thigh separation is described as typical of strong initial acceleration.

  Open source
• E26Grade CTrunk and shank angles in sprint acceleration (ISBS 2020)

  A comparison of trunk and shank angles between elite and sub-elite sprinters during sprint acceleration. ISBS Proceedings 2020. Abstract-level reading only; authors not recorded.

  Used for: Trunk and shank angles at touchdown separate elite from sub-elite sprinters in the first steps, so both are worth measuring; the direction and size of the differences must be taken from the full text before any target is set.

  Open source