What does this map leave out?

Our record holds 1,082 ties for the CNRS: laboratories it runs, projects it shares, papers its researchers sign with companies. Open the map and you find 21 of them, drawn faintly, and none drawn plainly. The other 1,061 are not on the map at all. They are not gone. Every one of them is in the database, and the data page measures them. The map simply did not draw them, for two reasons we can state exactly — and that is the point of this story: when a line is missing, the first question is whether the map left it out, and the second is whether we ever found it.

Rule one: a tie is drawn only when at least one end is a core organisation — the 917 small and medium robotics organisations this study is about. That is241,575 ties, 96.5% of everything we hold. Rule two: a kept tie with a large body at one end — a ministry, a national research organisation, a multinational, 4,873 of them — is drawn faintly, so that it frames the picture instead of swallowing it. That is 2,402 ties. The CNRS is a large body; hence its 21 faint lines, and none plain. Why cut at all? 250,437 lines would be a blot.

No. Removing a rule cannot recover evidence we never collected. 40.9% of our event records come from one register, the EU’s list of the projects it funds, and one project with thirty participants gives thirty records. Our collection is deep where a register publishes freely and thin where a company would have to be read one page at a time. We measured that against lists other people keep: we find the organisations that call themselves robotics companies, and we miss the incumbents — of the 300 largest holders of US robotics patents, another source of ours knows 102. So when a line is missing, ask two questions: did we find evidence for it, and did the map include it?