Define relevant differences
People vary in working style, dominant hand, pace, posture, and familiarity with a task. Identify which differences are relevant to the intended model and which details are unnecessary to collect. Do not assume a broad participant count establishes useful diversity. A collection can involve several people repeating the same tightly scripted movement in one unchanged setting.
Separate people from sessions
Request stable, appropriately handled participant identifiers that distinguish repeated sessions from independent participants. Explain whether the model should generalize to new people or only new recordings. This affects how the collection is divided for evaluation. A new filename or recording day does not automatically make an example independent of a participant already seen in training.
Observe how capture changes behavior
A participant may slow down, reposition objects, or hold a tool unnaturally to satisfy filming instructions. Determine whether you want controlled demonstrations, ordinary performance, or both. Record the distinction in session metadata. Review early footage for unintended staging so a large collection does not quietly become a demonstration of the capture protocol itself.
Review coverage with appropriate detail
Use a matrix of task, participant, environment, and session rather than a single total. Ask which combinations are sparse and whether specific people performed only particular activities. Keep personal data requests limited to what the project actually needs and address handling requirements in the source review. Send behavioral requirements rather than unsupported claims about population representativeness.
Review checklist
- List relevant behavioral variation
- Distinguish participants from sessions
- Document staged and natural performance
- Review task-by-participant coverage