Data Structure & Analysis

rtcStats makes use of WebRTC statistics via the getStats() API call. That gives the foundations of much of the analysis performed.

rtcStats organizes its analysis as a pyramid of five layers. Each layer is more interpreted, more compressed, and closer to an answer than the one below it. The pyramid orders layers by insight density, not by where the computation runs.

The five layers

• Foundations - the raw WebRTC metrics as reported by getStats(), minimally processed for readability.
• Calculations - derived values computed from the Foundations: rates, averages, aggregations, and quality estimates like MOS.
• Features - structured per-session facts (client, connection, track) distilled from the metrics and written to SQL, enabling cross-session analysis.
• Observations - expert rules that read a session's metrics and flag what is worth knowing, each with a title, severity, and domain tags.
• Deductions - the top of the pyramid: weighs the Observations and scores the session to point at root causes. This is where the quality scores and the AI summary live.

Where each layer runs. Foundations, Calculations, and Features are the open-source, local extraction layer: they run inside your own rtcstats-server. Observations and Deductions are produced by the rtcStats enrichment engine, available both on rtcstats.com and as an on-premise enterprise package.