Papers

Fault-mix stabilityThe recall deficit survives large changes in fault prevalence, while Pareto membership changes with the allowed false-alert burden.

Forecastability atlasThe strong raw risk ordering follows a common energy gradient. Energy-adjusted and conditional diagnostics do not support independent predictive value from oracle instability.

Reversal envelopeThe nearest tie is an exact geometric margin on the published affine lines, not a probability distribution or a plant-uncertainty claim.

Transport failure mapRecalibration narrows the global gap without reversing it. The cell map shows why: semantic summaries help and hurt in different held-out settings.

Witness stabilityMost rejections have redundant official witnesses. Four orientations expose the exact single-fixture boundary of the recorded contract.

Topology regimesThe optimizer stays in one discrete regime over most tested weights, then switches near the exergy endpoint while held-out feasibility remains limited.

Sensor observabilityDirect measurements remove null modes one at a time. Placement matters: two three-sensor combinations still leave the balance system structurally ambiguous.