The engine
Behind every score,
a real engine.
AI gets you part of the way. Movement Theory uses something more rigorous underneath: a measurement engine purpose-built for dance, with technique standards calibrated by working dancers.
Not a chatbot looking at video.
A common pattern in 2026 is to point a general-purpose AI at a video clip and ask it to grade what it sees. That's fine for casual comparisons. It's not enough for technique.
The issue isn't intelligence. It's grounding. A language model can describe a leap convincingly. It can't reliably measure whether your ankle reached full extension, how long you were actually in the air, or whether your weight shifted before your push-off. Without that measurement layer, what you get back is confident-sounding feedback that's mostly inference from pixels.
Movement Theory does the measurement first. The narrative comes after, grounded in what was actually measured.
How it grades
Measurement, then interpretation.
Track your body in space
On-device pose tracking locks onto the anatomical landmarks that define every position in dance: the joints, the spine, the extremities. Frame by frame, those points produce a quantitative record of what your body actually did.
Interpret it against the technique
Every supported move has its own set of technique standards: a structured rubric, calibrated to what that specific move actually requires. Your record is graded against those standards. Then, and only then, the AI coach writes the notes.
The score and the notes don't come from the model's opinion of your video. They come from a measurement-first analysis the model is summarizing for you in plain language.
The human in the loop
Calibrated by working dancers.
A rubric is only as good as the technique standards behind it. Generic body-tracking apps grade movement against generic targets: straight knees, level shoulders, good posture. That's not how a judge thinks, and it's not what your teacher will tell you in the studio.
Each rubric in Movement Theory is shaped over time with working competition dancers and dance educators. Real clips. Real corrections. Real consensus on what a Platinum version of a move looks like versus a High Gold version, and why the difference matters.
This part takes the longest. It's also why a Platinum on Movement Theory means what it means.
Per-move technique
Every move judged on its own technique.
Dance technique is not one thing. The thing that makes a pirouette good is not the thing that makes a sissone good. Generic "good form" coaching collapses that distinction. The engine doesn't.
Each supported move carries its own rubric, with criteria specific to that move's technique. A turn is graded on the things turns are graded on. A leap is graded on the things leaps are graded on. A held balance is graded on the things held balances are graded on. The feedback respects what you were actually trying to do.
That catalog grows over time. Every move added goes through the same calibration loop before it ever reaches your feedback page.
What you see
What this means for your report.
When your report mentions a specific joint angle, that's a real measurement. When it describes how your line held in the air, that's a real measurement too. When the score reads 87 / 100, that number is the structured result of measured signals graded against the rubric, not a vibe.
The AI coach's role is to translate those measurements into language a dancer can use, tied to the specific trick you were drilling. The judgment underneath is the engine's.
Built so the score actually means something.
Three free analyses every month at launch. Run a trick. See where the engine puts you.