Curve-level dominance of the structure-aware monitor

Establish whether the OnTrack quadratic structure term yields generally superior detection precision or curve-level dominance over reference-free and no-structure variants across operating thresholds and task distributions.

Background

The component study reports that, on homogeneous SWE-agent traces, the quadratic structure term improves the deployed false-alarm operating point but does not improve recall or ranking in a clear, general way. The no-GW and reference-free variants can achieve comparable or higher recall at substantially higher false-positive rates.

The authors therefore distinguish an operating-point advantage from a demonstrated precision–recall or ROC-curve advantage. Whether the structure-aware transport formulation is uniformly or generally superior across thresholds and more diverse tasks remains unresolved.

References

The component study clarifies what the transport machinery changes on real traces: not recall --- L1 heuristics and even a prerequisite-rule baseline match or exceed it there --- but the operating point: at shipped thresholds the quadratic structure term selects a point with less than half the benign false-alarm rate of every reference-free variant (13/40 vs. 29--32/40 benign traces flagged; removing the frontier mask instead shifts the count by one trace); establishing curve-level dominance remains open, and we do not claim generally superior detection precision.

— OnTrack: Real-Time Monitoring and Intervention in LLM Agent Trajectories via Streaming Structure-Aware Optimal Transport  (2610.12375 - Barazandeh et al., 8 Oct 2026) in Conclusion