ClarusC64/clinical-intervention-competition-sepsis-v1

Clinical Intervention Competition Sepsis Detection Overview

This dataset tests whether a model can determine which intervention pathway best stabilizes a sepsis-like clinical system.

In real clinical settings multiple interventions may be available at the same time. Each intervention affects system dynamics differently. Some actions move the system toward recovery while others fail to meaningfully counteract the instability trajectory.

The task is to determine which intervention most effectively stabilizes the system.

Prediction target

labelbestintervention

A = intervention A provides the strongest stabilizing effect B = intervention B provides the strongest stabilizing effect C = intervention C provides the strongest stabilizing effect

The task is to identify which intervention path most effectively counters the system’s instability trajectory.

Row structure

Each row represents a synthetic clinical scenario.

Columns:

scenarioid pressure buffercapacity couplingstrength trajectorydrift interventionAeffect interventionBeffect interventionCeffect

Training rows include the label. Tester rows omit the label.

Evaluation

The scoring script reports:

accuracy

Primary metric accuracy

Accuracy is used because the task requires selecting the single best intervention pathway among competing options.

Why this benchmark matters

Clinical decision making often involves choosing between several possible treatments.

Different interventions interact with the underlying system dynamics in different ways. An intervention that stabilizes one system state may be ineffective or harmful in another.

This benchmark tests whether models can reason about intervention competition in a dynamical system rather than simply detecting deterioration or intervention presence.

Structural note

This dataset exposes system geometry while keeping the generator used to produce the scenarios private.

The goal is to evaluate whether models can reason about structural intervention competition rather than memorizing treatment patterns.

Clarus Stability Geometry Benchmarks

This dataset is part of a broader benchmark family exploring instability and recovery in complex systems.

Related probes include:

clinical-compensation-collapse-sepsis-v1 clinical-fork-point-sepsis-transition-v1 clinical-organ-failure-cascade-v1 clinical-recovery-window-sepsis-v1 clinical-intervention-alignment-sepsis-v1 clinical-recovery-stability-sepsis-v1 clinical-false-stability-sepsis-v1

Together these benchmarks map the lifecycle of instability and recovery in clinical dynamical systems.

License

MIT