Geometric versus Packet-Confirmed Biconnectivity in UAV Recovery: Research Dataset and Reanalysis

This dataset accompanies the manuscript Geometric versus Packet-Confirmed Biconnectivity in UAV Recovery: A Matched Edge-Semantics Evaluation. The study investigates the difference between geometric range-based and bidirectionally packet-reliability-qualified edge admission in post-failure unmanned aerial vehicle (UAV) network recovery. Both approaches use the same bounded recovery planner and matched experimental conditions. The primary E3 experiment comprises 600 paired recovery scenarios across four UAV failure strata (F = 1, 3, 6, and 9). An outcome-independent selection of 100 paired scenarios was evaluated using independent bidirectional packet measurements. The archive contains scenario definitions, radio calibration data, planner outputs, endpoint records, 61,454 UAV-pair measurement results with corresponding CSV summaries, and SHA-256 integrity evidence. Packet-confirmed vertex biconnectivity was achieved in 64 of 100 scenarios using reliability-qualified planning, compared with 27 of 100 using geometric range-based planning. The paired difference was 37 percentage points, with a 95% scenario-stratified bootstrap confidence interval of 29–45 percentage points. A separate exploratory E4 extension includes 1,200 paired planning scenarios across network scale, traffic load, and reliability-threshold variations. These results characterize planner-level sensitivity; they do not establish independent packet-confirmed connectivity across the additional operating conditions. The deposit also includes exploratory geometric comparisons and two Python analysis scripts that independently reconstruct the reported connectivity outcomes and reproduce the original 10,000-replicate bootstrap from the archived measurements. It does not include the complete original ns-3 simulation and recovery-solver source code required for rerunning all experiments from scratch. Research data are licensed under CC BY 4.0; the two included analysis scripts are licensed under MIT. File-level license assignments and data integrity information are provided with the dataset. Funding support was provided through the TÜBİTAK 2219 International Postdoctoral Research Fellowship Program for Turkish Citizens (application no. 1059B192200615).