Field collection fails in the same handful of ways on every project. None of them are exotic. All of them are invisible at the moment of capture, which is exactly what makes them expensive.
The nine checks
- Sync drift — streams that have slipped against each other beyond the agreed tolerance.
- Dropped frames — gaps in a stream, truncated recordings, files that end mid-episode.
- Exposure and focus — blown highlights, underexposed interiors, motion blur.
- Occlusion — hands or the manipulated object out of frame during the part of the task that matters.
- Calibration drift — stereo or depth extrinsics that moved since the rig was last calibrated.
- Protocol compliance — episodes that skipped a step or deviated from the task definition.
- Duration outliers — runs far shorter or longer than the distribution.
- Coverage gaps — under-represented operators, sites or lighting conditions in a batch.
- Consent and redaction — a missing consent record, or a redaction rule not applied.
Why automate them
A trained reviewer can spot every one of these. The point of automating is not that machines are better at it, but that reviewer attention is the scarcest thing in the pipeline. Spending it on "is this file broken" instead of "is this a good demonstration" is the single most expensive mistake we made early on.
Write more here.