Analyzing Video Metrics like Richard Feynman
summary
Concrete claims and tools replace platitudes: reviews of QoS metrics like rebuffering startup time watch time, use of Anscombe's quartet to show identical summaries can hide different truths, and practical techniques — density distributions, quantiles with error bars, choosing a process sigma, and adjusting tests for auto correlated data. Surprising results include Akamai’s 2s/6% rule not matching Snap and Snap’s finding that 44% of viewers see a P90 stall over a week.
AI-written summary · about
chapters (4)
- 0:00Introduction
- 4:20Understanding the Nature of Metrics
- 7:45What do other people think
- 11:00Richard Feynman story
related talks
17mMetrics Driven Obsession with Viewer Experience
15mTowards Measuring Perceptual Video Quality & Why
14mUsing Performance Signals to Improve Video Recommendations
33mThe Challenges of Testing Video Playback at Scale
26mMagic Mirror: Understanding the Diversified Network Conditions of Twitch Community
48mHow I learned to stop worrying and love UX metrics
22mA taxonomy of video "quality," or: was Strobe right or wrong about quality?
18mDebunking the Myth of Video Delivery Monitoring on End-User Devices