A phone call is two streams of tiny audio packets — one coming in (the caller, what the AI hears) and one going out (the AI's voice, what the caller hears). These numbers measure how cleanly those packets flow.
call_receivedLoss In
% of the caller's audio packets that went missing. High = the AI hears the caller cutting out. The most reliable number here. Under 1% is great; over 8% sounds badly broken.
call_madeLoss Out
% of our outgoing packets the caller's phone reported losing — i.e. the caller hears the AI cutting out. Only available when the phone line reports it back; a 0 can mean "no data", not always "perfect".
speedThroughput
Roughly how much audio data is flowing in, in kbps. A normal voice call sits around 64–80 kbps. It's an estimate from packet counts, not an exact byte measurement.
graphic_eqJitter
How uneven the packet timing is, in milliseconds. Packets arriving at irregular gaps make audio sound robotic or stuttery even when nothing is lost. Under 30 ms is good; over 50 ms is audible.
timerLatency
Round-trip delay in milliseconds — the "walkie-talkie lag" feeling. Under 150 ms is good; over 300 ms causes people to talk over each other. Needs the phone line to report back, so it may show 0.
grainPkt Rate
Packets received per second right now. A healthy voice call holds steady around 50/s. A dropping rate means audio is faltering.
verifiedThe badge (Excellent → Poor)
A one-glance rating based on the worse of Loss In / Loss Out. If it says FAIR or POOR, check which Loss side is high to see whose audio is suffering, then Jitter/Latency for why.