Supplementary data and analysis code for:Cross-Layer SCTP–NAS Recovery of 5G Standalone PDU Session Establishment under N2 Transport Partitions

This dataset and analysis package accompanies the manuscript “Cross-Layer SCTP–NAS Recovery of 5G Standalone PDU Session Establishment under N2 Transport Partitions.” It provides sanitized run-level results and supporting analysis material from controlled experiments using Open5GS 2.8.0 and a pinned, patched UERANSIM build in a single-host Ubuntu 22.04.5 virtual machine. The formal dataset comprises 30 independently reset, prospectively interleaved experiments involving selective bidirectional N2 transport interruptions during manually triggered PDU Session Establishment: five attempts at a nominal 0.5-second condition, ten at 4 seconds, and fifteen at 8 seconds. All 30 attempts had verified fault exposure and raw-artifact checksums; no attempts were excluded or replaced. Calibration observations are documented separately and are not included in the formal denominator. The archive contains: A complete 30-run TSV table with run identifiers, nominal conditions, trace-defined recovery classes, restoration-to-original-DATA cumulative-acknowledgment intervals, NAS T3580 events, changes in the current-run SMF session trajectory, and bounded cleanup outcomes. The frozen experimental sequence and build/configuration hash manifest. Python analysis scripts, classifier regression tests, and recorded regression results. Selected calibration reports, the formal results report, the analysis amendment, and an evidence manifest linking derived records to retained raw archives. The experiments distinguish N2 path restoration from recovery of outstanding SCTP signaling and the ongoing NAS procedure. Median restoration-to-original-DATA cumulative-acknowledgment intervals were 0.613, 3.172, and 15.540 seconds for the three nominal conditions, respectively. All 30 attempts established bidirectional user service. All fifteen attempts in the 8-second condition experienced T3580 expiry before original DATA recovery and failed the separate 30-second cleanup audit. An SMF classification defect was corrected before execution resumed at slot 9. Slots 1–8 were retained and reanalyzed using the corrected classifier; results for the post-fix slots 9–30 are reported separately. This release contains derived data and analysis source. Raw packet captures, system logs, subscriber identifiers, credentials, and private effective configurations are excluded. Raw evidence remains in a separate private archive; requests for appropriately sanitized material may be directed to the corresponding author. The observations apply to the documented emulated testbed and experimental conditions. They do not establish a universal outage-duration threshold or validate recovery behavior in physical radio networks or other core implementations.