Green marks the best values within tolerance; all-equal rows or columns have no highlight. There is no combined ranking.
A plain row completed the category. An amber row finished but the server could not hold the 60 fps budget (frame p99 past 33 ms or more than a sixth of frames dropped): its bandwidth and CPU describe a saturated server and can be lower than a healthy one's, so they are shown but never marked best. Delivery and connection problems stay in the notes.
Each category is reported separately. A red row did not complete the category (crash, resource limit, a truncated test or an unconfirmed delivery check): no averages, no best mark; completed categories remain usable after a later failure. Run errors stay in the notes. Idle and Static are baselines.
Peak RSS is the process-lifetime maximum, not category-local memory. It remains available in the per-test view.
Same workload and machines within each session. Traffic includes protocol overhead; CPU covers the whole server process. Fusion uses Photon Cloud. See the README for setup changes and measurement limits. Detailed diagnostics are in the raw results.