Identify every time reference
A delivery may include media timestamps, frame numbers, session time, and annotation offsets. List what each field means and which reference is authoritative. File creation time is not necessarily capture time. Ask for units and endpoint conventions so a reviewer can locate the same event in the video and its labels without guessing.
Test representative event alignment
Choose visible transitions that are easy to locate, then compare the media with associated action intervals or sensor records. Review the beginning and end of longer sessions as well as a short example. A fixed offset and a changing mismatch are different problems. Record what was tested instead of declaring every stream synchronized from one matching event.
Track transformations
Trimming, dropped frames, re-encoding, and joining segments can affect how a label maps to delivered footage. Retain source intervals and processing information for derived files. If labels use frame indices, confirm which version of the video they reference. A playable file and valid annotation syntax can still describe different moments when their derivation histories diverge.
Define the acceptance question
The required timing precision depends on the task. Broad activity descriptions may tolerate a different alignment policy than brief contact events. State the events that must remain correctly associated and review borderline examples. Ask Fiund about timing evidence and available metadata for the particular source; do not assume wearable footage includes a shared clock or additional sensor alignment.
Review checklist
- Document clocks and units
- Compare events across the session
- Retain transformation mappings
- Tie timing review to target events