NABS-SIM results

Generated 18 September 2026, 14:41 · 462 simulated scan cycles

Headline

By cohort

Clean control has no lesion zones, so AUC is undefined by construction — it measures false-positive behavior only.

Mission profile

The two wrong-sign calls are both heel L under R3 during the febrile window, where a large lesion contaminates its own annulus. The P3 precondition exists to catch that and its 1.6× threshold is too permissive — a known open item.

Figures

Pennes forward model

Left: synthetic skin-surface temperature with the eight monitored prominences. Right: the perfusion field driving it. Perfusion is the controlled variable; no field is hand-drawn.

Two-level scan schedule

Warnsdorff tile order at left, Warnock leaf quads shaded by allocated dwell at right.

Why a knight's tour, not a raster

Acquisition index against position under both orderings, and the spurious zone delta-T that results. Shaded band is the reference sigma.

Sensor chain under enclosure conditions

Discriminability as steam and residual surface water are introduced.

Operating envelope

378 cycles across the steam-extinction by residual-water grid.

Reference ablation - the headline result

Same phantoms, same sensor, same conditions. The only change is whether an admission baseline is on file.

Per-cohort discriminability

AUC measured by rank from the zone-level statistic, not assumed. Amber cohorts are adversarial by construction.

40-cycle mission telemetry

Path transmittance, error terms, effective sigma and resulting AUC across a full bath profile.

Mitigation stack

Applied cumulatively at the worst-case operating point. Note the diminishing return on the sensor improvement.

Files