Datasets:

ClarusC64
/

clinical-quad-stress-buffer-lag-coupling-icu-collapse-v0.7

scenario_id string	stress float64	buffer float64	lag float64	coupling float64	drift_gradient float64	drift_velocity float64	drift_acceleration float64	boundary_distance float64	boundary_uncertainty float64	trajectory_uncertainty float64	regime_confidence float64	perturbation_radius float64	collapse_trigger string	recovery_distance float64	recovery_gradient float64	return_feasibility float64	delta_stress float64	delta_buffer float64	delta_lag float64	delta_coupling float64	trajectory_shift float64	minimal_intervention_path string	stabilization_success int64	label_icu_collapse int64
ICU7001	0.68	0.43	0.56	0.63	0.55	0.48	0.12	0.23	0.17	0.2	0.76	0.24	hemodynamic_instability	0.54	-0.18	0.42	-0.17	0.12	-0.05	-0.06	-0.13	fluids+vasopressor_review	1	1
ICU7002	0.74	0.39	0.62	0.7	0.62	0.54	0.15	0.18	0.12	0.15	0.83	0.2	sepsis_progression	0.58	-0.21	0.36	-0.19	0.14	-0.06	-0.08	-0.16	antibiotics+source_control	1	1
ICU7003	0.57	0.5	0.5	0.58	0.41	0.35	0.09	0.33	0.23	0.27	0.64	0.29	transient_metabolic_shift	0.47	-0.11	0.5	-0.1	0.08	-0.04	-0.05	-0.08	electrolyte_correction+monitoring	1	0
ICU7004	0.79	0.36	0.67	0.75	0.69	0.61	0.18	0.15	0.1	0.12	0.87	0.18	multi_organ_drift	0.62	-0.25	0.29	-0.22	0.16	-0.07	-0.09	-0.2	critical_care_escalation+support_bundle	1	1
ICU7005	0.61	0.45	0.55	0.61	0.48	0.42	0.1	0.28	0.36	0.39	0.56	0.31	mixed_icu_signal	0.52	-0.13	0.39	-0.11	0.09	-0.05	-0.06	-0.07	diagnostics+support	0	0
ICU7006	0.77	0.37	0.65	0.73	0.66	0.58	0.16	0.16	0.09	0.11	0.88	0.19	shock_refractoriness	0.6	-0.24	0.31	-0.2	0.15	-0.07	-0.09	-0.18	vasopressor_adjustment+volume_review	1	1
ICU7007	0.55	0.51	0.48	0.56	0.37	0.32	0.08	0.39	0.24	0.28	0.66	0.34	slow_respiratory_drift	0.45	-0.1	0.55	-0.08	0.07	-0.04	-0.05	-0.07	serial_review+monitoring	1	0
ICU7008	0.7	0.41	0.6	0.68	0.6	0.52	0.14	0.2	0.42	0.45	0.47	0.27	uncertain_regime_shift	0.56	-0.12	0.33	-0.13	0.11	-0.06	-0.07	-0.05	imaging+supportive_care	0	0
ICU7009	0.72	0.4	0.59	0.67	0.61	0.53	0.14	0.19	0.14	0.17	0.79	0.23	renal_hemodynamic_decline	0.55	-0.2	0.37	-0.18	0.13	-0.06	-0.08	-0.15	hemodynamic_optimization+renal_review	1	1
ICU7010	0.63	0.46	0.54	0.62	0.47	0.41	0.1	0.27	0.18	0.22	0.71	0.28	post_intervention_drift	0.5	-0.15	0.43	-0.12	0.1	-0.05	-0.06	-0.11	reassessment+lab_review	1	1

What this repo does

This repository contains a Clarus v0.7 dataset modeling ICU collapse using a quad-coupling system representation.

The dataset extends the v0.6 intervention layer by introducing uncertainty geometry.

The question addressed by earlier versions was:

Can the system be stabilized?

v0.7 adds a second critical question:

How confident are we in that conclusion?

This allows Clarus to distinguish three operational states:

• confident deterioration
• confident stabilization
• uncertain regime

This distinction matters in ICU settings where false confidence can delay escalation and obscure evolving collapse.

Core quad

The system state is represented by four structural variables.

• stress
• buffer
• lag
• coupling

These variables capture the structural state of ICU deterioration rather than isolated measurements.

Clinical variable mapping

Quad Variable	Clinical Measurements	Typical Indicators
stress	Vasopressor need, lactate burden, oxygen demand, inflammatory load	Rising lactate, increasing pressor support, worsening oxygen demand
buffer	Hemodynamic reserve, renal reserve, metabolic reserve, respiratory reserve	Poor urine output, acidosis, low reserve margin
lag	Time to recognition, delayed review, delayed intervention	Late escalation, slow imaging, delayed treatment response
coupling	Multi-organ spillover, cardio-renal-respiratory interaction, diffuse instability	Rising SOFA, worsening creatinine, ventilation-hemodynamic interaction

This mapping connects the structural quad variables to common ICU collapse indicators.

Prediction target

Target column:

label_icu_collapse

Default v0.7 rule:

label = 1 if stabilization_success = 1 AND trajectory_shift < -0.10

Relaxed variant optional rule:

label = 1 if stabilization_success = 1

The stricter rule ensures that a stabilization must produce a meaningful trajectory correction, not only a transient response.

Row structure

Each row represents an ICU system state and contains:

• system state variables
• trajectory dynamics
• boundary geometry
• uncertainty geometry
• perturbation information
• recovery dynamics
• intervention vector
• stabilization outcome
• classification label

Uncertainty signals

v0.7 introduces explicit uncertainty modeling.

Two signals are required.

boundary_uncertainty
trajectory_uncertainty

An optional support signal may also be present.

regime_confidence

boundary_uncertainty

Normalized variance in the boundary_distance estimate across perturbation samples.

Range: 0–1

Higher values indicate the instability boundary is poorly defined in the current region of the state space.

Low values indicate a stable boundary estimate.

trajectory_uncertainty

Normalized variance in the drift_gradient estimate across consecutive state snapshots.

Range: 0–1

Higher values indicate the direction of system motion is unstable or oscillating.

Low values indicate consistent directional drift.

regime_confidence

Confidence that the system is operating within the assumed ICU collapse regime.

Low values indicate ambiguous system behavior or overlapping failure patterns.

Files

data/train.csv

Training rows include:

• intervention vectors
• stabilization_success
• label_icu_collapse

data/tester.csv

Tester rows exclude:

• stabilization_success
• label_icu_collapse

However the following signals remain available during inference:

• boundary_uncertainty
• trajectory_uncertainty
• regime_confidence

These signals allow models to estimate prediction reliability.

Example v0.7 row

Example of a high-uncertainty stabilization scenario.

scenario_id: ICU7011
stress: 0.71
buffer: 0.40
lag: 0.60
coupling: 0.69

drift_gradient: 0.57
drift_velocity: 0.50
drift_acceleration: 0.13

boundary_distance: 0.22
boundary_uncertainty: 0.46
trajectory_uncertainty: 0.42
regime_confidence: 0.49

perturbation_radius: 0.26
collapse_trigger: mixed_multi_organ_signal

recovery_distance: 0.55
recovery_gradient: -0.13
return_feasibility: 0.34

delta_stress: -0.15
delta_buffer: 0.10
delta_lag: -0.05
delta_coupling: -0.06

trajectory_shift: -0.12
minimal_intervention_path: support_bundle+targeted_escalation

stabilization_success: 1
label_icu_collapse: 1

Interpretation:

The intervention succeeds.

But boundary and trajectory uncertainty are both high.

The stabilization result therefore carries low confidence.

This is exactly the type of case v0.7 is designed to expose.

Evaluation

Standard classification metrics:

• accuracy
• precision
• recall
• f1

v0.7 also introduces uncertainty diagnostics.

• recall_successful_stabilization
• failed_rescue_rate
• high_uncertainty_false_positive_rate
• boundary_misconfidence_rate
• trajectory_misconfidence_rate

These metrics detect situations where a model produces confident predictions in uncertain conditions.

Dataset construction

This dataset is generated using the Clarus instability modeling framework.

The generation process follows four stages.

State sampling

System states are sampled across the quad space:

• stress
• buffer
• lag
• coupling

Sampling includes both stable and near-boundary regions.

Trajectory estimation

Trajectory signals are computed for each state:

• drift_gradient
• drift_velocity
• drift_acceleration

These values describe motion through the stability manifold.

Boundary estimation

Instability boundaries are estimated using perturbation sampling.

This produces:

• boundary_distance
• boundary_uncertainty

Boundary uncertainty represents the variance of boundary estimates across perturbations.

Intervention simulation

Candidate intervention vectors modify system state:

• delta_stress
• delta_buffer
• delta_lag
• delta_coupling

Trajectory is recomputed after intervention to determine:

• trajectory_shift
• stabilization_success

Labels are assigned according to the v0.7 rule.

Dataset limitations

This dataset models instability geometry, not complete patient records.

Important limitations:

• Quad variables represent abstract system states rather than full ICU physiology
• Intervention effects represent structural transitions rather than full treatment protocol detail
• Uncertainty signals represent estimation uncertainty rather than monitor artifact or sensor noise

The dataset should therefore be interpreted as a system stability benchmark, not a direct clinical decision tool.

Intended use

Appropriate uses include:

• benchmarking instability detection models
• studying intervention reasoning in dynamic systems
• evaluating uncertainty calibration in safety-critical predictions
• training models to recognize unstable stabilization scenarios

Not intended for:

• direct clinical deployment
• real-time treatment recommendations
• replacing medical judgment

Position in the Clarus dataset ladder

This dataset belongs to the Clarus instability modeling ladder.

The ladder reconstructs the geometry of system failure and recovery.

v0.1
cascade detection

v0.2
trajectory awareness

v0.3
cascade forecasting

v0.4
boundary discovery

v0.5
recovery geometry

v0.6
intervention reasoning

v0.7
uncertainty-aware intervention geometry

Relationship to other Clarus datasets

Clarus datasets apply the same instability geometry across domains.

These include:

• clinical systems
• financial networks
• infrastructure systems
• AI coordination environments

Each domain models the same structural signals:

• system state
• trajectory dynamics
• boundary estimation
• recovery pathways
• intervention reasoning
• uncertainty geometry

This enables transfer of structural reasoning across domains.

Research direction

Future Clarus datasets may extend the ladder through:

• regime transition modeling
• multi-cascade coupling
• adversarial perturbation environments
• stability control policies

These developments move Clarus from a monitoring framework toward a general system stability instrument.

Structural note

Clarus v0.7 extends the intervention ladder.

The instrument now models:

• system state
• trajectory
• boundary
• recovery
• intervention
• uncertainty

This transforms Clarus from a collapse detector into a calibrated system navigation instrument.

Production deployment

The dataset class is most useful where false confidence carries high operational risk, including:

• ICU monitoring
• early collapse detection
• multi-organ deterioration review
• escalation bundle evaluation
• unstable system reassessment environments

Enterprise and research collaboration

Potential applications include:

• clinical AI benchmarking
• intervention pathway model evaluation
• uncertainty calibration research
• instability monitoring systems
• safety evaluation for clinical decision support models

License

MIT

Downloads last month: 64

Number of rows:

Total file size:

23.2 kB