Website fingerprinting (WF) enables an adversary to predict the website a user is visiting, despite the use of encryption or Tor. Previous work almost exclusively uses synthetic datasets to evaluate the success of WF attacks. We present GTT23, the first dataset of genuine Tor traces, intended especially for WF analysis. We obtain it through a measurement of the Tor network, and, with 1.4 ×10^7 traces, it is larger than any existing WF dataset by an order of magnitude. We survey 28 WF datasets published since 2008 and compare them to GTT23, discovering common deficiencies of synthetic datasets for drawing conclusions about the WF effectiveness. We have made GTT23 available to other researchers.
@inproceedings{gtt23-pam2026,
title = {A Measurement of Genuine Tor Traces for Realistic Website Fingerprinting},
author = {Jansen, Rob and Wails, Ryan and Johnson, Aaron},
booktitle = {Passive and Active Measurement Conference},
year = {2026},
}