Answering: Complete the movement
Analyze a clip to select one primary movement.
The same retained interval will drive the moments, evidence, chart, and next action.
Browser-local attempt coach
One attempt.
One next move.
Choose what you are trying to improve. The app follows one retained movement from setup to peak, connects every claim to a frame and measurement, and gives one evidence-backed focus for the next attempt.
01
Start with the decision
What should the next attempt improve?
The analyzer will find the strongest retained rise, identify the most limiting visible evidence, and choose one next-attempt focus.
1 · Findone primary movement
→
2 · Explainmoments + evidence
→
3 · Actone next-attempt focus
02
Checking camera support…
Capture one complete attempt
No camera permission requested. Choose a source to begin.
A defensible clip shows:
whole body + both hands
the bar throughout
a fixed side or slight-front view
textured background around the athlete
03
Build one evidence chain
observed 2D model estimate
Recording 0:00
Pose idle
Analyze once; answer the selected goal
Pose, confidence, camera flow, bar roll, scale, primary movement, key moments, scorecard, and next action all derive from the same retained samples.
Pose model loads on first use. Background stabilization runs automatically with clip analysis.
Live evidence diagnostics
Pose confidence—Waiting for a pose
Elbows— / —left / right image-plane angle
Pose depth spread—root-relative model units
Bar correction—applied to the same result axes
Review movements0primary movement is selected after analysis
04
Analyze a clip
Answer for this attempt
No answer yet.Choose a goal and analyze one complete attempt. Low-confidence motion will be rejected instead of turned into advice.
Select the unit of analysis
Review the suggested movement
Analysis will suggest the strongest confidence-gated rise. Confirm it—or choose another—before using the coaching action.
No review movements yet.
Evidence chain
One movement · four linked moments
Select a moment to seek the same local clip and inspect the pose overlay.
Setup, fastest rise, peak, and finish appear after analysis.
Connected scorecard
Each observation answers a question
No universal technique score; the selected goal determines relevance.
No observations yet.
Progress without uploading
Compare with your previous saved summary
Only the goal and compact derived evidence are stored in this browser—never media, frames, or landmarks.
No local comparison yet.
Metric conversion and camera evidence
Evidence timeline for the same movement
The selected interval and four review moments are synchronized with the local video.
positionvelocityacceleration
Methods, alternative movements, gravity check, and limits
What changed the result
- Waiting for analysis.
Experimental bar-normal acceleration check reference g: 9.80665 m/s²
Optional evidence lab only. Mark the projected horizontal bar, then mark genuinely unsupported flight with at least six stabilized samples spanning 0.25 s. The fitted axis is normal to the projected bar under a near-parallel-plane approximation—not independently calibrated gravity. Contact, active tucking, camera tilt, sparse samples, poor stabilization, or out-of-plane motion invalidates it. It never supplies or modifies the report metric scale.
No free-flight interval marked. Human-height scale remains independent.
Limits and uncertainty
- Single-view output is an estimate, not ground truth.
- No force, power, torque, injury-risk or clinical inference is made.
Export the complete evidence chainGoal, primary movement, moments, scorecard evidence, one action, calibration, series, alternatives, and limitations—never raw media.
Experimental evidence lab · depth and root-relative pose
These views can help inspect occlusion or depth ordering. They do not change the primary answer or produce metric 3D.
Optional dense depth
Estimate one paused frame with local Depth Anything V2. This relative/ordinal map can help interpret depth ordering; it does not produce metres and does not gate pose results.
Not loaded. Relative/ordinal output only.
Root-relative pose view
Rotate the model-estimated skeleton. Its origin is near the hips; it is not aligned to the room, camera or bar and is not metric scene reconstruction.
Interpretation boundary
More defensible: timing, stabilized 2D displacement, gross joint configuration, left/right asymmetry and trends from a clear clip.
Less defensible: absolute values under heavy perspective, occlusion, blur, poor background texture, unverified free flight, or a body moving strongly toward the camera.