Seeing Through Encrypted QUIC: A Dual-Substrate Dataset for Early Congestion Prediction and Cross-Network Generalization

Dual-substrate open dataset for early congestion prediction in encrypted QUIC: a 3,600-flow synthetic tc-netem testbed (220,099 windows; 6 regions x 4 CC x 2 queue disciplines) and a 17,193-flow real-network corpus (~1.23M windows over 14 multi-continent nodes), with qlog-derived congestion ground truth, the collection and feature/label pipeline, and all benchmark/evaluation scripts and per-experiment result files. Accompanies the paper of the same title. Code is MIT-licensed (see bundled LICENSE); data is CC-BY-4.0.