PeerSig
Pilot phase: the relay runs on the same machine as this server and is operated by the same team, so for now it does not hide participants' network addresses from the study team. The other protections (blind tokens, counts only, randomized response, thresholds) apply in full.
PeerSig finds which applications send which TLS fingerprints (JA4). Volunteers run a small app that watches the start of the encrypted connections their computer makes and, once a week, contributes which applications sent which fingerprints, as anonymous ballots. The result is a public ground-truth table for traffic classification research.
Take part: get a code
From the UCLouvain network
On campus or through the UCLouvain VPN: the server checks that your connection comes from the UCLouvain network and hands you a code, with nothing to fill in. It does not keep your network address.
By e-mail
The code is sent to this address. The server does not keep the address: it keeps a keyed hash of it, so that one address gets at most 3 codes (one per computer). Nothing links that hash to your code.
Then
- Install PeerSig from the releases page (macOS or Linux).
- Open it and enter your code. The addresses are filled in already; they should be:
- Server: https://peersig.org
- Relay: https://relay.peersig.org
- Each week, PeerSig shows you what it will send and waits a day before sending it.
Questions about the study: contact@peersig.org.
How data is protected
- Only application names and versions, the fingerprints of their handshakes and the OS family leave a participant's computer. Never addresses, visited sites, times, file paths or names.
- Each observation is a separate anonymous ballot carrying a blind-signed token: the server knows it comes from a participant, not which one, nor which ballots come from the same person.
- Ballots go through an Oblivious HTTP relay, which sees the sender's address but not the content; the server sees the content but not the address. This hides who sends what as long as the relay and the server are run by different parties.
- The server stores one counter per item and week: no addresses, no arrival times, no individual ballots.
- Items already known are reported with randomized response (p = 0.90, ε = 2.20 per item and week).
- An item is published once at least 3 participants reported it (5 for domains), with Laplace noise (ε = 1).
Current week: 0. Configuration. The full design is in docs/privacy.md.