Research
The hard parts, written down.
Measuring the internet from the edge is not a solved problem. These are the questions the design is currently wrong about, kept here so they can be argued with.
Open questions
Four problems without clean answers yet.
-
01
Sampling bias
Contributors are not a random sample of connections. People who install a measurement client skew toward better lines and more technical households. A map built from them overstates how good the internet is, and no amount of verification fixes that. Bounties help, and they do not solve it.
-
02
Adversarial readings
Paying for measurements creates an incentive to fabricate them. Corroboration between independent devices is the main defence, which works where density is high and fails exactly where coverage is thin and the data is most valuable.
-
03
What a probe can see
An edge probe measures a path to a reference endpoint. It cannot tell you which hop was at fault, and attributing a bad reading to an operator rather than to peering, to the home network or to the device itself is mostly inference.
-
04
The observer effect
A throughput test consumes the capacity it is trying to measure, and on a congested line it changes the thing it reports. Bounded transfers reduce it. They do not remove it.
Prior art
This is not the first attempt.
Edge measurement has a long history: academic measurement platforms, regulator-run hardware probes, browser-based speed tests, and operator telemetry. Each gets something right. The platforms have rigour and almost no density. The regulator probes have density in a handful of countries. The speed tests have enormous reach and a self-selected, single-moment sample. Operator telemetry has everything except independence.
The gap being aimed at here is continuous, independent, verified measurement at consumer density, with the method published. Whether contributor rewards can produce that without producing fraud at the same scale is the open question the whole project rests on.
Disagree with any of this
The method is meant to be argued with before it is built on.