How accurate is Atlas? The honest answer.
The real methodology behind every red, yellow, and green: what's measured well, what's deliberately not reported, and where every band edge came from. Nothing here is gated.
For hip and knee bending — flexion and extension — Atlas lands in roughly the same error range as the tool a clinician uses in a clinic (a goniometer).
Some things a camera can't measure reliably — like rotation, or ankle angles. Atlas doesn't report those as graded findings, instead of estimating them.
Atlas measures. It never diagnoses. Any screen that becomes part of your care is read by Brian Jones, DPT, who makes every clinical decision.
Everything below is the same story at full technical depth, public so you can check it.
Compared to what?
The realistic alternative isn't a $150,000 motion-capture lab — it's a clinician judging your movement by eye over video. Against that, repeatable timestamped measurement is a real improvement, and it doesn't require overstating what a single camera can do.
The claims we can back up
Roughly the same error band as clinical goniometry itself — about 2–5° — with zero rater drift, because there is no rater.
Cadence, stance-time symmetry, and pelvic drop — frame-timing measurements, exactly what a camera is good at. Walking speed needs a calibration we don't perform, so we leave it out.
Same protocol, same math, every time — built for tracking your change over time: a baseline, then re-checks against it.
Forward-head posture, shoulder-height asymmetry, pelvic obliquity — a method with published reliability, reported as screening indices worth a closer look, never a diagnosis.
A single camera has real limits. We built around them instead of hiding them.
Rather than show a number we don't trust, Atlas suppresses these findings:
- ✕Rotational measurements — the literature on single-camera rotational accuracy isn't strong enough, so we don't display them.
- ✕Ankle joint angles — a known weak point for markerless motion capture generally, so no graded finding.
- ✕A diagnosis — Atlas flags patterns worth a closer look. Naming the condition is Brian's job.
- ✕Precision past what the method supports — a method with degrees of real-world error can't defend a decimal place, so we round and band.
We label our own confidence. Nobody asked us to.
Every posture finding carries one of three honesty labels — we'd rather tell you exactly how confident we are than let you assume more.
Band edges (green ≥75°, yellow 60–75°, red <60°) were set from three deliberately-graded reference poses — a real derivation, but statistically thin, and we say so. We also publish the failure mode we found: up to 24° side-to-side disagreement on the most exaggerated pose.
Anchored to a published radiographic study, read through a camera proxy instead of an X-ray — a real gap, and we say so rather than borrow the study's precision.
Band edges (2°/5° tolerance) come from Brian's clinical judgment as a starting point — stated outright, moving to calibrated status as capture data accumulates.
Two findings on some posture screens — kyphosis (a rounded upper back) and rounded shoulders — aren't graded by Atlas today. The camera can't measure them reliably yet, so instead of guessing, the screen tells you honestly that it isn't a reliable read.
A screening tool a clinician reviews. Not a diagnostic device.
Every finding traces back to the video and measurement behind it, and Brian Jones, DPT reviews anything Atlas flags before it becomes part of your care. Atlas never makes an autonomous diagnostic call.
And we won't call Atlas "FDA registered" the way some competitors do — that's a fee and a form, not an approval.
Try the free movement screen — same honesty, real time.
Every band in the demo is built on exactly what's described on this page. Try it free — no email first.