overlap

For room codes, a service holds one connection blob from each side for ten minutes so the two browsers can find each other. It never sees a name or a hash, and the copy and paste route skips it entirely.

This is not a private set intersection protocol, and calling it one would be a lie. Both browsers agree a random salt and then swap salted hashes, so the person you are comparing with holds the same salt you do. They can hash a list of people they are curious about and see whether any of them come back. This is a tool for two people who have chosen to compare notes. It is not a defence against the person on the other end.

The honest version of the promise is short: your list never leaves your browser. The other browser receives salted hashes, the salt and a count, never your file or its names.

They also learn two things that are not on the list. Roughly how many connections you have, because the number of hashes is the number of names. And your IP address, because the two browsers connect straight to each other and that is what a direct connection is. Nothing about this tool avoids either.

The room code service holds one connection blob from each side for ten minutes and a daily-changing hash of your address for up to an hour. It never sees a hash from your list, a name, or the file. The blob does carry your address inside it, because that is how two browsers find each other.

This tool writes nothing to your machine: no cookie, no local storage, nothing. The forget command has nothing to wipe here.

Can't see

• Second-degree paths. It compares two lists of people you are each already connected to, so somebody you both reach through a third person is invisible to it.
• Warmth. A colleague you speak to weekly and a stranger who sent a request in 2019 look exactly the same in an export, and this tool does not read the connected-on date to guess between them.
• Anyone who has changed their profile URL since one of the two exports was taken. The slug is the identifier, so an old file and a new one hold two different people as far as the hashing is concerned, and nothing can detect that.
• Rows with no profile link. LinkedIn leaves the URL out when a connection has restricted it, and those rows are counted and reported rather than guessed at.
• Old style /pub/ links from exports taken years ago. They are a different identifier space from an /in/ slug and comparing the two would invent matches.