ClarusC64/clinical-quad-renal-stress-buffer-lag-coupling-aki-transition-v0.4

What this repo does

This repository introduces a Clarus v0.4 dataset for acute kidney injury transition boundary discovery.

The dataset models renal deterioration as a coupled system where stress, physiological reserve, delayed intervention, and organ coupling interact.

Version v0.4 focuses on identifying when the system lies on the instability boundary rather than simply forecasting collapse.

The task is to detect when the renal system is sufficiently close to collapse that small perturbations can trigger acute kidney injury.

Core quad

renalstress physiologicalbuffer interventiondelay organcoupling

renal_stress represents physiological strain on renal function.

physiological_buffer represents systemic resilience and remaining metabolic reserve.

intervention_delay represents delay in correcting renal stress.

organ_coupling represents propagation of dysfunction across organ systems.

Trajectory layer

drift_gradient

Range −1 to +1

Negative values indicate stabilization.

Positive values indicate movement toward deterioration.

Dynamic forecasting layer

driftvelocity driftacceleration boundary_distance

These variables describe how rapidly the system is moving toward the collapse boundary.

Boundary discovery layer

Two variables describe the instability boundary.

perturbationradius collapsetrigger

These convert the dataset into a cascade boundary discovery benchmark.

Boundary variable definitions

perturbation_radius

Minimum perturbation required to trigger cascade.

Range 0 to 1.

Smaller values indicate that the system is extremely close to collapse.

collapse_trigger

Binary indicator of whether the simulated perturbation caused collapse.

0 stable 1 cascade

collapse_trigger is included as an observed perturbation response feature.

It is not the prediction target.

The prediction task is to identify the underlying boundary-risk state.

Prediction target

labelakitransition

The label is positive when either condition holds

boundary_distance < 0.10

or

perturbation_radius < 0.08

These thresholds represent proximity to the instability manifold and minimal perturbation collapse risk.

Row structure

scenario_id

renalstress physiologicalbuffer interventiondelay organcoupling

driftgradient driftvelocity driftacceleration boundarydistance

perturbationradius collapsetrigger

labelakitransition

Files

data/train.csv data/tester.csv scorer.py README.md

Evaluation

accuracy precision recallboundarydetection falsesaferate f1 confusion_matrix

Primary metric

recallboundarydetection

License

MIT

Structural Note

Clarus dataset progression

v0.1 cascade detection v0.2 trajectory detection v0.3 dynamic forecasting v0.4 boundary discovery

Production Deployment

Research dataset for cascade detection benchmarking and instability monitoring research.

Not intended for clinical decision use.

Enterprise & Research Collaboration

For dataset expansion, custom coherence scorers, or deployment architecture:

team@clarusinvariant.com

Instability is detectable. Governance determines whether it propagates.