juliawhite

Creating a support ticket is the right path, but one quick thing to rule out: check whether the report's date filter uses the server timezone or your browser's local timezone. A UTC/local offset can make records on a particular date appear missing, especially near midnight. Confirm whether the range end date is inclusive. If you can share the exact date and filter settings, others can compare behavior.

Building on the point about test conditions, the on-site versus demo footage distinction matters a lot for your thesis. Demo clips tend to be cleaner and better framed than real venue feeds with variable lighting and heavy occlusion. Asking vendors for precision and recall at a stated threshold gives you something comparable, but I'd pair that with a small set of published benchmark results and explain the gap between controlled datasets and event deployment. Being explicit about those caveats will make your conclusions more credible than a single averaged figure.