# Muscle-up practice analysis — 16 August 2026

## Executive result

Seven attempts were reviewed across three clips. 3 reached stable top support. The first attempt in each clip succeeded (3/3); all later attempts did not (0/4). The limiting question has therefore shifted from absolute transition capability to repeatability of projected bar clearance and coordinated turnover under continued work.

## Evidence ledger

| Set | Clip attempt | Outcome | Window | Pre-turnover chest / bar-normal proxy | Max bar-normal rate proxy | Min elbow | Frontal 2D asymmetry |
|---:|---|---|---|---:|---:|---:|---:|
| 1 | IMG_6710 / 1 | full turnover to stable top support | 1.45–7.51 s | +0.06 m | 1.00 m/s | 46° | view-limited |
| 2 | IMG_6710 / 2 | high pull without turnover; returned to hang | 11.41–15.39 s | -0.14 m | 1.34 m/s | 63° | view-limited |
| 3 | IMG_6711 / 1 | full turnover to stable top support | 2.21–7.91 s | +0.02 m | 0.96 m/s | 47° | view-limited |
| 4 | IMG_6711 / 2 | high pull without stable turnover; returned to hang | 18.30–22.79 s | -0.15 m | 1.18 m/s | 70° | view-limited |
| 5 | IMG_6711 / 3 | partial asymmetric turnover; no support | 31.41–34.96 s | -0.10 m | 0.91 m/s | 62° | view-limited |
| 6 | IMG_6712 / 1 | full turnover to stable top support | 1.30–6.78 s | +0.07 m | 1.57 m/s | 37° | 67° |
| 7 | IMG_6712 / 2 | partial asymmetric turnover; returned to hang | 13.81–18.31 s | -0.12 m | 1.29 m/s | 34° | 83° |

## Findings

1. **Repeatability is the current bottleneck.** All three first attempts reached support; all four later attempts did not. Within the same camera/view, later attempts lost 0.11–0.19 m of pre-turnover projected chest-to-bar-normal clearance versus the first attempt; the combined sensitivity envelopes did not overlap. Corrected maximum bar-normal rate did not decrease consistently, so speed loss is not presented as the shared mechanism. Use rested singles and stop accumulating fatigued misses.
2. **Turnover symmetry remains variable.** Later attempts in the side-oblique and frontal clips show one-side-first partial turnover; the frontal miss also includes a pronounced leg split. In the frontal clip only, the 2D elbow-asymmetry proxy increased from 67° to 83° and tracked ankle spread from 0.53 m to 1.21 m; the absolute values are projection/model-sensitive, but the direction agrees with the visible miss.
3. **The tuck creates height but does not guarantee support.** Compact knee tuck is visible in successes and failures. The discriminating phase is getting the chest over the bar while upward momentum remains.
4. **Setup is confounded with fatigue.** Successful first efforts follow a fresh jump/grip; several misses begin from a continuing hang. Standardize grip, swing and rest before interpreting failure as pure fatigue.

## Practice prescription

- 3–5 clean singles, 2–3 minutes rest; end the set when lower-chest pull height or bilateral turnover degrades.
- Low-bar/band-assisted turnover: lower sternum to bar, both wrists above the same line, both elbows back together.
- High-pull singles with a controlled tuck; after chest reaches bar height, close the leg split and send the torso over.
- Next recording: fixed frontal camera, hand/grip marks, one prescribed swing, 2:30 timed rest, 60–120 fps if available.

## Method

- MediaPipe Pose Landmarker Heavy at native decoded frame rate; YOLO11m-pose cross-check sampled every three frames (~10 Hz).
- Fixed-background SIFT/RANSAC camera compensation and clip-specific paired-edge active-grip-tube tracking. Clips 1 and 2 use visually reviewed positive-slope anchors, with a regression test that rejects the earlier wrong near-horizontal line. Active-bar residual p95 was 5–14 px.
- Across reviewed windows, median MediaPipe confidence was 0.80–0.97 and median YOLO/MediaPipe landmark disagreement 16–23 px.
- Stature-scaled signed image-normal projection proxies with ±7% apparent-scale sensitivity, tracked-bar uncertainty and 9/15/21-frame smoothing sensitivity.
- Attempt windows and stable-support outcomes were visually reviewed; the model did not decide success.
- Successful-attempt clearance proxy is capped at the last clearly pre-turnover frame so supported top-position geometry is not compared with failed hanging-side peaks.

## Boundaries

Single-camera footage cannot directly measure calibrated depth, true vertical clearance or velocity, bar force, joint torque, muscle activation or injury risk. Absolute anatomical angles and rates are projection-sensitive. Timing and within-clip comparisons are more defensible than cross-view absolute values. Force/power proxy fields in raw data are not used for the coaching conclusion.


## Source upload boundary

Google Photos still labelled the shared items as **Partial upload** at the recorded fetch time. This release analyzes the complete 1080×1920 copies then available. The manifest preserves this state and exact SHA-256 hashes; a later original-file rerun must not be presented as byte-identical continuity.