Four input sequences run through the executable AUTHREX-style decision pipeline. Every stage verdict is computed and every stage gates the terminal decision - SATA provenance, the ADARA research detector, IFF roster, independent-node MAIVA consensus (fail-closed), the HMAA authority tier, a bus-latency TIMING gate, FLAME deliberation, and ERAM risk. Hover any stage for its decision trace.
Seeded Poisson-arrival mixed-protocol traffic with simulation-assigned truth labels (65% bursty, 35% low-and-slow). Reports both system detection and ADARA-only detection so the anomaly detector's skill is not conflated with the IFF roster rule. Batch size is capped at 25,000 so every processed message is captured in the traffic export with no silent loss; batches run in async chunks to keep the UI responsive.
Load an external JSON array of feature rows and evaluate the pipeline against it, instead of the internal generator. This addresses the circular-validation concern: the detector is scored on data it did not generate. Each row: {src, proto, op, klass:"benign"|"malicious", features:{burstRate,fcEntropy,roleMismatch,provenanceGap,offHours}}. Use “Export traffic” from any run to produce a compatible file.