Movement Standards
Every rep in a Cence competition is judged against this document — by the vision engine on every attempt, and by human review on every podium. If a rep doesn't count, the verdict names the exact standard below that it violated. No hidden rules, no judgment calls behind closed doors.
Every attempt begins from a valid start. The counter does not arm until the start position is confirmed.
Feet flat on the floor, positioned anywhere from hip-width to one-and-a-half times shoulder-width. Toes may point forward or turn out up to 30°.
Bodyweight squat, no external load. Arms are free — extended forward, crossed, or held at the chest. Arm position is not judged and does not affect the count.
The attempt arms only when you are holding a full standing lockout (see §3) for one full second. A countdown confirms the counter is live.
The bottom of the rep. This is where most reps fail — and where the standard is exact. Depth in Cence is decided by one ruling measure, checked against two others.
From standing lockout, hips and knees bend in one continuous descent. Brief deceleration is fine; reversing direction before reaching depth voids the rep.
At the bottom of the rep, the tracked hip joint center must pass below the tracked knee joint center. This is the measure that decides the rep.
Cence measures the hip joint center, not the inguinal crease used in some barbell rulebooks. Those points sit inches apart and the gap varies between bodies, so we publish the point we actually track rather than borrow language for a landmark the engine does not see. Judged side-on, hip-below-knee is deeper than thigh-parallel.
Two further measures are read at the bottom of every rep and must agree with the §2.2 ruling: the knee angle taken at the hip–knee–ankle points, and the femur angle relative to horizontal. Agreement is what makes a depth call trustworthy — hip tracking is the noisiest point on the body, and a single threshold on a noisy point is not a standard, it's a guess.
Where the corroborating measures contradict the ruling measure beyond published tolerance, the engine does not resolve the conflict in either direction. The rep is neither counted nor flagged as a miss — it is marked unresolved, and the reason is shown.
Depth is judged per rep, not on average. Every counted rep individually satisfied §2.2 with corroboration under §2.3. There is no "close enough" band.
At the top of the rep, hips and knees return to full extension and the torso reads upright at the tracked joints. This is the only position where a rep completes.
Lockout must be reached with visible control — a beat, not a bounce. Continuous cycling is allowed as long as full extension is genuinely reached each time.
A squat is a full-body movement. Position is part of the rep — but only where the camera can honestly see it.
Both heels remain in contact with the floor through the full range of every counted rep.
The torso may incline forward — that is normal squat mechanics — but may not pitch beyond the published tolerance, which converts the movement into a different lift.
Knee tracking relative to the feet — valgus collapse — is a real movement fault and is not judged under SQ v1.0. The side-on camera required by §6 cannot see it, and Cence does not score what it cannot see.
There is no minimum or maximum rep speed. Fast cycling and slow grinders both count, as long as every rep individually meets §2 and §3.
Rest is permitted in the standing lockout position (§3.1). Time continues to run; the attempt window is fixed regardless of rest taken.
A competition attempt is a single continuous recording within the published time cap for that week's event. When the window closes, the counted total is final.
The engine can only verify what it can see. Camera setup is part of the standard — and for the squat it is stricter than for the push-up.
Squat attempts are judged side-on only. The camera sits perpendicular to the direction you face, at roughly hip height, 6–10 feet away, with your full body in frame for the entire attempt.
Depth under §2.2 is a comparison of two heights. Viewed from the front or from behind, that comparison is not available at any level of tracking confidence — the body can be perfectly visible and the measurement still meaningless. The engine confirms side-on placement before the attempt arms and will not score a front or rear view.
Every rep carries a tracking-confidence reading. Reps scored below the published confidence floor are not silently counted or discarded — they are flagged for review.
The camera does not move during the attempt. Handheld or repositioned recordings are not accepted for competition.
Weight classes only mean something if the weight is real.
The standards we hold the engine to — published, like everything else.
| Expert agreement | A movement's automated scoring must agree with expert human judges at 95% or better before it decides prize outcomes without rep-level review. |
| False acceptance | No more than 2% of non-compliant reps may be accepted by automated scoring at competition grade. |
| Unresolved rate | The share of reps marked unresolved under §2.4 is published per event. A rising unresolved rate is a tracking problem we disclose, not one we quietly resolve into counts. |
| Until then | Where automated scoring hasn't yet cleared these gates, human review covers the gap — and we say so, plainly, on the event page. |
| Versioning | This standard is versioned. Changes are published before they apply to any competition window — never mid-event. Attempts are always judged by the version in force when the event opened. |