PL-CausesNS3-50k

We release PLCauses-NS3, a reproducible dataset for fine-grained packet-loss causation in mobile ad hoc and IoT-like wireless networks. Traffic is generated in ns-3 (v3.45) over an 802.11g ad-hoc topology with random-walk mobility. A calendar mechanism activates distinct operating regimes in time\u2014benign, congestion, mobility, and interference\u2014to create labeled conditions without manual annotation. For each 1-second window, we aggregate per-packet and PHY\/MAC traces into features commonly used in networking and ML: PRR, ETX, SNR (dB), queue length (normalized), queue-drop rate, neighbor churn (exposed as link_changes_per_s), offered load (pps), and proxy indicators for congestion\/interference\/mobility (PQ\/PM\/P\u03b7). A lightweight quality-control (QC) flag highlights windows where evidence disagrees with the calendar label.The public release includes both raw logs (TX\/RX\/PHY-radio\/queue) and window-level tables in CSV\/Parquet, a machine-readable schema, deterministic calendars and seeds, and scripts for end-to-end regeneration from a pinned ns-3 commit. A typical build provides ~50k labeled windows (e.g., 84 runs \u00d7 600 s) and is suitable for training and benchmarking loss-cause classifiers, diagnosing wireless performance anomalies, and evaluating routing\/queueing policies under controlled mobility and interference. The dataset targets researchers and practitioners working on MANET\/IoT reliability, network telemetry, and ML-for-networks, and is designed for straightforward extension to security scenarios (e.g., selective forwarding) and additional PHY\/MAC technologies.