Why these four thresholds.

Four gates run at ingest. A clip that fails any of them is rejected or flagged; nothing that fails is quietly mixed into a delivery. The thresholds are not picked for how they sound. Three follow from the optics and timing of head-mounted capture. One does not. Three follow from the optics and timing of head-mounted capture, and one does not.

1 · Hand visibility, 50%

Below half the frames, a hand-pose track contains more interpolated frames than observed ones. MANO fitting on an occluded hand is extrapolation from the last good observation, and per-joint contact state inherited from an extrapolated pose is a model output, not a measurement.

Fifty percent is where the balance tips. It is a floor, not a target: clips that clear ingest typically run 70 to 90 percent, and the sample records we publish sit in that band. What the floor protects is the meaning of the L5 contact layer — if contact labels can be produced on clips where the hands are mostly unseen, the label stops being evidence of anything.

2 · Inertial drift, 1 ms per 60 s

At 30 fps a frame occupies 33.3 ms. One millisecond is three percent of a frame. The threshold is expressed per minute because drift accumulates: a three-minute clip at the limit has drifted 3 ms, still under a tenth of a frame; a nine-minute clip has drifted 9 ms, about a quarter.

The reason the budget is a fraction of a frame rather than a fraction of a second is that contact onset is a single-frame event. If the inertial stream cannot be assigned to one specific frame, it can only be assigned to a neighborhood, and every downstream use — motion segmentation, phase boundaries, force correlation — inherits that ambiguity. Hardware timestamping is what makes that budget affordable. Software synchronization after the fact does not get there.

3 · Reprojection error, 3 px at 1080p

This is the only gate with a direct metric consequence, so it is worth showing the arithmetic — and stating the assumption before the numbers, because at this field of view the assumption does real work.

Reprojection error is computed on the rectified image, after lens distortion has been solved and removed. A wide raw field of view is not linear in angle per pixel; near the frame edge a naive division would be wrong by a large factor. On the rectified image, near the optical axis, the approximation below holds. Across the fleet the figure runs 0.063°–0.083° per pixel (120° to 160° horizontal at 1920 px, equidistant fisheye), so three pixels is 0.19°–0.25° of angular error. A wider field of view buys more metric error for the same pixel budget. At a working distance that becomes:

The error this gate does not show. Reprojection is a 2D residual. With a 62 mm stereo baseline the same three pixels of disparity error give a depth uncertainty of roughly 5–6 mm at 300 mm, 13–18 mm at 500 mm and 34–45 mm at 800 mm — depth error grows with the square of distance while lateral error grows linearly. These are orders of magnitude, not measurements.

Around two millimeters of lateral uncertainty at bench distance is tolerable for most manipulation and intolerable for fine assembly, where clearances run under a millimeter. So the gate is not the whole answer: fine-assembly clips carry a tighter internal target, and the published 3 px is the corpus-wide floor beneath which metric geometry stops being usable for anything.

Frame rate is nominal. The measured rate on delivered assets is 29.90 Hz, so reconstructing time as frame index divided by 30 drifts by about 200 ms per minute — two hundred times the gate. Use the per-frame timestamps.

4 · Composite score, 70 — the convention

The first three gates are necessary and not jointly sufficient. A clip can clear all three and still be unusable: the object leaves frame during the critical phase, the retarget is feasible but degenerate, the reconstruction is locally correct and globally drifting. The composite weights five sub-scores into one number: reconstruction, hand pose, tracking, retarget feasibility (does an IK solution exist within joint limits) and simulation replay (does the solved trajectory hold up under dynamics). The last two are separate because a trajectory can be kinematically solvable and dynamically impossible.

Seventy is not derived from anything. It is a working threshold, set by review rather than by measurement, at roughly the point where reviewers stopped agreeing on whether a flagged clip was usable. We would move it if that point moved, and we would say so.

We publish the distinction because a vendor that presents all four numbers as equally principled is overstating three of them or one of them. Ours is one.

Each gate’s pass rate should be measured on the units that reached it, not on the raw pool. Defects are correlated — a badly lit session fails perception, reconstruction and interaction together — so multiplying marginal rates overstates the loss.

For MOVAS-Ego 100K the published end-to-end figure is 49%: of 100,000 candidate hours, 49,000 reach the trainable set. The per-gate breakdown ships on the data card of every delivery, by scene family.

Why floors rather than targets

The published work on scaling egocentric pre-training reports a log-linear relationship between hours and validation loss, with a comparatively small volume of robot data needed afterward to align the action space. A log-linear curve has an unforgiving property: to keep improving, the corpus has to keep multiplying, and a gate that removes ten percent of the hardest hours costs more than it returns.

That is the argument for floors. A gate should exclude data that is wrong, not data that is difficult. Every threshold above is set at the point where the measurement stops being trustworthy — not at the point where the clip stops being easy.

What the gates do not do

None of this measures whether a task is worth learning. Every gate above is a measurement-quality check on the capture. A clip can pass all four and still be a poor training example, because the manipulation is trivial, the operator’s intent is ambiguous, or the corpus already holds a thousand near-identical instances. That judgment is made separately and by people, and it is the part of data curation that has not been automated in any pipeline we know of, including ours.

There is a failure mode in the other direction too, worth naming because the incentives run that way. Tightening these gates would raise every published number and produce a corpus of clean, well-lit, unoccluded, slow manipulations — which is to say, a corpus with the hard cases removed. Occlusion, fast motion and clutter are not defects in egocentric data. They are the conditions a deployed policy will meet.