LinkedIn has a page called "Get a copy of your data". It offers two archives: a fast one and a larger one that can take a day. I asked for both.
The fast one came first. A ZIP with twenty-six files. Most are spreadsheets with plain names: Connections, Invitations, Positions, messages. Connections has a name, a profile link, a company, a job title and the date we connected. Mine has 339 rows. Two of them have no name and no link. Those are people who deleted their accounts. They are still in the count.
What the archive doesn't contain is the thing I wanted. It does not say who knows whom. The site shows mutual connections; the export leaves them out. My own data turned out to be a set of spokes without a wheel.
So the ties have to be guessed. Two people at the same company probably know each other, or at least share a kitchen. Two people with similar job titles are probably in the same trade. That is all the tool assumes, and the page labels every such tie as inferred.
Then there are dates. Dates are where the archive starts talking about me.
For about two and a half years I added fewer than one person a week. Then almost two. Since August, about fourteen. One week in September holds 65 new connections, nearly five times the rate around it. The tool finds weeks like that with an exact binomial test against the rate of their period, and shows them as a group with dates. It doesn't name them. Naming is left to me.
It also looks at messages without reading them. It keeps who wrote, when and in which direction, and drops the text before anything is saved. From that it can tell who starts conversations and how quickly people answer. It also looks for contacts I talk to in the same weeks, the way two colleagues turn up together around one project. It tested the pairs and corrected for the number of tests. It found nothing. My conversations, it turns out, are independent events.
A word on privacy, because it is the part I would ask about first. The archive is mostly other people: their names, their employers, the day they agreed to know me. So the tool sends it nowhere. It reads the ZIP in place, builds the map on your computer and writes one HTML file that opens with a double-click. The page carries a security policy that forbids it to load or send anything. Only your user can read the files. The screenshots in the repository come from a synthetic export of 320 invented people at invented companies. Nobody in them agreed to be published, because nobody in them exists.
For the curious: DuckDB in memory, NetworkX for communities (Louvain), PageRank and betweenness, CUSUM changepoints for the growth periods, and Benjamini-Hochberg correction for the bursts. Under a steady rate, false bursts show up in fewer than 5% of simulated exports.
There are five views. Constellations groups contacts into communities. Companies gives each employer a hub. Interaction puts you in the centre and the people you write to closest. Timeline shows when each connection happened, one lane per group. Cohorts shows the bursts.
The tool is called tigra. It is open source under Apache-2.0 and is not affiliated with LinkedIn. It takes three steps: request the archive, download the ZIP, run one command.
I built it to see my network. Mostly I saw my calendar.