About
A speed test tells you how fast a single big download could go. It doesn't tell you whether your next Zoom call will glitch. That's a different question, and it deserves a different test.
The Mbps paradox
Speed tests like Ookla and Fast.com measure peak burst throughput over TCP. A 500 Mbps result means your link can push 500 megabits during a one-shot transfer — but real-time apps don't stream that way. Zoom, Teams, and Meet send small UDP packets at a steady cadence (typically 100–600 kbps for video) and need every packet to land within tight timing budgets.
What actually breaks calls is jitter, packet loss, and bufferbloat — none of which appear on a speed test result. You can have 500 Mbps and a broken call, or 25 Mbps and a perfect one.
What we measure
StabilityPulse runs a 30-second synthetic WebRTC stream through a public TURN relay. The browser sets up the same kind of peer connection a video-conferencing app would, then we capture the four metrics that actually predict call quality:
- Loaded latency
- Round-trip time while the link is busy. Calls feel laggy when this exceeds ~150 ms.
- Jitter
- Variance in packet inter-arrival time. The single biggest predictor of robotic audio.
- Packet loss
- Fraction of RTP packets dropped. Anything above 1% is audible.
- Bufferbloat
- Loaded RTT minus idle RTT — extra latency a saturated buffer is adding. Common on consumer routers.
The 0–100 score
Those four metrics roll up into a single stability score using a weighted model: jitter 40%, loss 40%, RTT 20%. We chose these weights because jitter and loss damage real-time audio more severely than absolute latency does — a steady 200 ms link sounds fine on a call; a 50 ms link with bursts of 500 ms does not.
Green: 80 or higher — calls should feel native.
Yellow: 50 to 79 — usable, but expect occasional artifacts.
Red: under 50 — calls will degrade noticeably.
Per-app readiness
Zoom, Teams, and Meet have published their own quality thresholds for jitter, loss, and RTT. We compare your results against each one separately so you know which calls will hold up — even when the overall score is yellow, one app might still be fine while another won't.
What we deliberately don't do
- No ICMP ping or traceroute. Browsers can't send ICMP. Server-side traceroutes add infra without changing the verdict.
- No competing speed test. The optional burst test is a contrast feature, not the product.
- No accounts, no history. Phase 0 is browser-only. Test results never leave your device except for the WebRTC stream itself.
Limitations to be honest about
- The TURN relay sits in one geographic region. If you're far from it, baseline latency will be higher than what you'd see calling someone nearby. We're addressing this in Phase 2 with multiple points of presence.
- Background tabs throttle WebRTC stats. Run the test with StabilityPulse focused.
- A bad result on Wi-Fi doesn't always mean a bad ISP. Try once on Ethernet to separate router/Wi-Fi problems from carrier problems.