Angle
Top-arm extension
High
Both tracked endpoints are usually visible; scale does not affect the angle.
Accuracy · the evidence contract
VaultLab produces pose-derived angles and calibrated distance estimates from one side-on video. It is built for repeatable coaching evidence—not laboratory certainty from a phone.
ANGLES · POSE-DERIVEDDISTANCES · CALIBRATEDCONFIDENCE · PUBLISHED
TRUNK AXIS169°
POSE MODELVISIBLE
01 · Real-world scale
Two known lengths
anchor the frame.
Dallin’s standing height and the crossbar create vertical scale. The recorded run provides context. Perspective still matters, so confidence follows the geometry.
ATHLETE6′0″
RECORDED RUN68.5 FT
PHONESIDE-ON · LEVEL WITH BOX
02 · Metric confidence
Not every number
deserves equal weight.
Each report should tell the coach whether to trust the exact value, the direction of change, or only the phase-level pattern.
Angle
Max trunk inversion
High
A second pose model agreed within about 10–11° on average across eight checked jumps.
Angle
Takeoff angle
Good
Geometric, but the plant and pole occlude joints at the exact takeoff frame.
Angle
Pole angle
Good
Tracked separately from pose; busy backgrounds can weaken the pole line.
Rate
Approach speed
Good
Frame timing is exact; real-world displacement depends on camera perspective.
Distance
Grip + finish height
Good
Vertical measurements sit near the known crossbar reference.
Distance
Takeoff distance
Directional
Horizontal depth carries more perspective error than vertical height.
Distance
Under / free position
Directional
Trust direction and session change more than the exact inch count.
03 · Independent model check
Same jumps.
Different pose model.
Eight jumps were also run through RTMPose. The two systems agreed on the coaching classification—finished or stalled—on every checked jump. Their inversion angles differed by roughly 10–11° on average.
Agreement between two models is not ground truth. It reduces the risk of a single-model hallucination; it does not replace physical validation.
0°90°180°
VaultLab poseCLASSIFIED
Independent poseCLASSIFICATION MATCHED
AVERAGE ABSOLUTE AGREEMENT≈10–11°
04 · Physical validation
The error table
is still being earned.
A ground-truth study will compare taped takeoff position, measured grip, recorded bar height and video-only output jump by jump. Until that study is complete, VaultLab publishes qualitative confidence—not invented percentages or sub-degree claims.
STUDY IN PROGRESS
Join the validation study What stays human
Measurement informs the decision.
The coach still makes it.
NO INJURY PREDICTIONNO AUTOMATIC POLE SELECTIONNO SAFETY OVERRIDE
See every measured signal Get started
See what one honest camera can show.
One complete measured jump is free. No card required.