ClarusC64/clinical-recovery-stability-sepsis-v1

Clinical Recovery Stability Sepsis Detection Overview

This dataset tests whether a model can distinguish between temporary improvement and true structural recovery in a sepsis-like clinical system.

In many complex systems, short-term improvement can occur even while the system remains dangerously close to the instability boundary. Vital signs may improve temporarily, but the underlying dynamics may still favor relapse or collapse.

The task is therefore not simply detecting improvement, but determining whether the system has moved far enough away from instability to enter a stable recovery basin.

The recovery stability problem

A recovering system can exist in two fundamentally different regimes.

Unstable recovery

The system shows signs of improvement but remains close to the collapse boundary.

Small disturbances can reverse the trajectory and push the system back toward failure.

Stable recovery

The system has moved far enough away from the instability boundary that disturbances no longer threaten system integrity.

Recovery becomes self-reinforcing rather than fragile.

The dataset asks whether models can detect the difference between these two regimes.

Core system geometry

Each scenario represents a simplified clinical dynamical system described by structural variables.

pressure Current physiological stress.

buffer_capacity Remaining physiological reserve available to absorb stress.

coupling_strength Strength of interaction between physiological subsystems.

trajectory_drift Directional movement of the system state.

recovery_velocity Speed at which the system is moving away from the collapse region.

stability_margin Distance between the current system state and the instability boundary.

Together these variables describe whether the system has entered a durable recovery basin.

Prediction target

label_stable_recovery

0 = unstable recovery 1 = durable recovery

The model must detect whether improvement represents genuine stabilization or only temporary relief.

Row structure

Each row represents a synthetic clinical scenario.

Columns:

scenario_id pressure buffer_capacity coupling_strength trajectory_drift recovery_velocity stability_margin

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

Evaluation

The scoring script produces the following metrics.

accuracy precision recall f1 specificity negative predictive value (npv)

Primary metric recall

Secondary metric f1

Recall is prioritized because missing stable recovery states can lead to premature escalation or unnecessary intervention.

Why this benchmark matters

Many predictive models focus on identifying deterioration.

However, in real clinical practice a central question is:

Is this patient truly recovering?

Mistaking unstable recovery for stable recovery can lead to:

premature withdrawal of treatment inadequate monitoring unexpected relapse

The benchmark tests whether models can reason about basin stability in a dynamical system.

Structural note

This dataset intentionally exposes system geometry without revealing the generator used to construct the scenarios.

The benchmark is designed to probe whether models can reason about stability margins and trajectory dynamics rather than memorizing static patterns.

Relationship to other Clarus probes

This dataset is part of a broader family of instability geometry benchmarks.

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-false-stability-sepsis-v1

Together these datasets map the lifecycle of instability and recovery in complex clinical systems.

Clarus Stability Geometry Benchmarks

This dataset is part of a broader benchmark family designed to test whether models can reason about instability and recovery in complex systems.

Current 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 explore the lifecycle of instability in complex clinical systems.

License

MIT