scenario_id int64 | systemic_stress float64 | physiological_buffer float64 | intervention_delay float64 | organ_coupling float64 | drift_gradient float64 | drift_velocity float64 | drift_acceleration float64 | boundary_distance float64 | perturbation_radius float64 | collapse_trigger int64 | recovery_distance float64 | recovery_gradient float64 | return_feasibility int64 | label_icu_collapse int64 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
1 | 0.83 | 0.22 | 0.65 | 0.74 | 0.71 | 0.62 | 0.36 | 0.04 | 0.02 | 1 | 0.78 | 0.49 | 0 | 1 |
2 | 0.75 | 0.31 | 0.56 | 0.68 | 0.53 | 0.47 | 0.21 | 0.08 | 0.05 | 0 | 0.62 | 0.17 | 1 | 0 |
3 | 0.67 | 0.37 | 0.5 | 0.6 | 0.39 | 0.36 | 0.14 | 0.13 | 0.07 | 0 | 0.43 | -0.13 | 1 | 0 |
4 | 0.8 | 0.26 | 0.61 | 0.7 | 0.64 | 0.56 | 0.29 | 0.05 | 0.03 | 1 | 0.71 | 0.38 | 0 | 1 |
5 | 0.58 | 0.45 | 0.43 | 0.52 | 0.15 | 0.25 | 0.02 | 0.2 | 0.1 | 0 | 0.24 | -0.37 | 1 | 0 |
6 | 0.72 | 0.34 | 0.53 | 0.64 | 0.46 | 0.41 | 0.18 | 0.1 | 0.06 | 0 | 0.5 | 0.05 | 1 | 0 |
7 | 0.89 | 0.18 | 0.73 | 0.79 | 0.79 | 0.68 | 0.42 | 0.02 | 0.01 | 1 | 0.84 | 0.56 | 0 | 1 |
8 | 0.63 | 0.41 | 0.47 | 0.55 | 0.21 | 0.29 | 0.06 | 0.17 | 0.09 | 0 | 0.31 | -0.26 | 1 | 0 |
9 | 0.85 | 0.21 | 0.68 | 0.75 | 0.74 | 0.64 | 0.37 | 0.03 | 0.02 | 1 | 0.8 | 0.47 | 0 | 1 |
What this repo does
This repository provides a Clarus v0.5 cascade recovery geometry dataset modeling ICU collapse.
Earlier Clarus datasets focused on detecting deterioration states and identifying instability boundaries.
Version v0.5 adds a recovery geometry layer that asks a stricter question:
Can the system still return to stability?
The task is binary classification over ICU deterioration states using:
• a four-variable clinical quad
• trajectory dynamics
• boundary discovery signals
• recovery geometry variables
Models must determine whether ICU deterioration remains reversible.
Core quad
The core quad for this dataset is:
• systemic_stress
• physiological_buffer
• intervention_delay
• organ_coupling
Operational interpretation:
systemic_stress
Represents aggregate physiological burden such as inflammatory load, hemodynamic instability, metabolic stress, and multisystem strain.
physiological_buffer
Represents compensatory reserve available to absorb critical care instability.
intervention_delay
Captures treatment lag between deterioration onset and stabilizing intervention.
organ_coupling
Represents propagation of dysfunction across interacting organ systems.
Trajectory layer
The dataset includes a trajectory signal:
drift_gradient
Range:
−1 to +1
Interpretation:
negative values indicate motion toward recovery
positive values indicate motion toward deterioration
This lets the model infer directional movement rather than assess a static snapshot alone.
Dynamic forecasting layer
Three dynamic variables describe system motion:
• drift_velocity
• drift_acceleration
• boundary_distance
These variables allow models to reason about how quickly the system is moving and how near it lies to the cascade boundary.
Boundary discovery layer
The dataset retains the boundary discovery layer introduced in v0.4.
Variables:
• perturbation_radius
• collapse_trigger
Interpretation:
perturbation_radius
Measures how much disturbance the system can absorb before crossing into collapse.
collapse_trigger
Binary indicator that the instability boundary has been crossed.
Recovery geometry layer
v0.5 introduces a recovery geometry layer that determines whether recovery remains possible.
Variables:
• recovery_distance
• recovery_gradient
• return_feasibility
These variables transform the task from collapse detection into recovery reasoning.
Models must determine not only whether a system is unstable, but whether a path back to stability still exists.
Recovery variable definitions
recovery_distance
Distance from the current system state to the nearest stable basin.
Definition:
recovery_distance = min ||x - x_stable||
Range:
0 to 1
Interpretation:
small values indicate proximity to a recoverable region
large values indicate deep cascade penetration
recovery_gradient
Direction of motion relative to the nearest recovery basin.
Range:
−1 to +1
Interpretation:
negative values indicate motion toward recovery
positive values indicate motion deeper into collapse
return_feasibility
Binary indicator representing whether recovery remains possible.
Values:
0
system has crossed an irreversible cascade boundary
1
system still lies within a recoverable region
Prediction target
Target column:
label_icu_collapse
A positive label indicates irreversible ICU collapse.
Collapse threshold
The cascade boundary threshold used for labeling is:
collapse_threshold = 0.05
Label logic
Positive labels trigger when either condition holds:
boundary_distance < 0.05
or
return_feasibility = 0
This encodes irreversible cascade detection.
Binary simplification note
The underlying system dynamics are continuous and multi-dimensional.
For benchmark clarity, the dataset compresses this structure into a binary classification task:
recoverable state
versus
irreversible deterioration
The recovery geometry variables preserve the deeper system structure.
Row structure
Each dataset row contains:
scenario_id
systemic_stress
physiological_buffer
intervention_delay
organ_coupling
drift_gradient
drift_velocity
drift_acceleration
boundary_distance
perturbation_radius
collapse_trigger
recovery_distance
recovery_gradient
return_feasibility
label_icu_collapse
Variable ranges
State variables
0 to 1
drift_gradient
−1 to +1
drift_velocity
0 to 1
drift_acceleration
−1 to +1
boundary_distance
0 to 1
perturbation_radius
0 to 1
collapse_trigger
0 or 1
recovery_distance
0 to 1
recovery_gradient
−1 to +1
return_feasibility
0 or 1
Files
data/train.csv
Labeled training examples.
data/tester.csv
Unlabeled test scenarios.
scorer.py
Evaluation script for binary classification.
cli.py
Optional command-line wrapper for running the scorer.
README.md
Dataset documentation.
Evaluation
The scorer reports the following metrics:
accuracy
precision
recall_irreversible_detection
false_recovery_rate
f1
confusion_matrix
Primary metric
recall_irreversible_detection
This metric prioritizes detection of irreversible deterioration.
Secondary diagnostic metric
false_recovery_rate
This measures how often irreversible states are incorrectly treated as recoverable.
Version progression
Clarus datasets evolve through successive capability layers.
v0.1
Cascade state detection datasets
v0.2
Cascade + trajectory datasets
v0.3
Cascade + trajectory + dynamic forecasting datasets
v0.4
Cascade + trajectory + dynamics + boundary discovery datasets
v0.5
Cascade + trajectory + dynamics + boundary discovery + recovery geometry datasets
Earlier versions remain unchanged to preserve benchmark continuity.
License
MIT
Structural Note
Clarus v0.5 marks the transition from instability mapping to recovery geometry.
Earlier datasets asked whether systems were approaching collapse.
v0.5 asks a more operational question:
Is recovery still structurally possible?
This makes the dataset class closer to real-world decision support systems.
Production Deployment
Recovery geometry datasets are suitable for applications where distinguishing recoverable ICU states from irreversible collapse is critical.
Possible domains include:
critical care escalation monitoring
ICU deterioration surveillance
early rescue pathway modeling
intervention timing simulation
Enterprise & Research Collaboration
For dataset expansion, custom coherence scorers, or deployment architecture:
Instability is detectable.
Governance determines whether it propagates.
- Downloads last month
- 27