Describe the temporal task
State whether the model needs to identify a moment, follow a sequence, or reason over a whole session. Define the shortest useful example and the longest acceptable file. This prevents a collection of short highlights from being mistaken for complete activities.
Separate capture from delivery
The source recording and the exported file may differ. Ask how footage was captured and what transformations occur before delivery. List required resolution, frame rate, aspect ratio, encoding, and audio tracks, along with any tolerance for mixed source quality.
List exclusions precisely
“Clean video” is ambiguous. Enumerate requirements for captions, visual overlays, watermarks, background music, edits, synthetic scenes, and third-party clips. Decide how exceptions should be documented and whether they are accepted, excluded, or reviewed individually.
Review before expanding
Use representative full-length examples and their metadata to check fit. Agree on how missing tracks, corrupt files, repeated footage, and interrupted sequences are handled. A small accepted sample should establish a shared interpretation of the brief, not substitute for review of the complete delivery.
Review checklist
- Model task and content categories
- Desired duration and total volume
- Video and audio specifications
- Overlays, edits, and content exclusions
- Commercial use, sample criteria, and timing