ClarusC64/clinical-drug-toxicity-instability-v0.1

clinical-drug-toxicity-instability-v0.1

What this dataset does

This dataset evaluates whether models can detect instability caused by pharmacological load exceeding clearance capacity.

Each row represents a simplified drug metabolism scenario observed across three time points.

The task is to determine whether pharmacological regulation remains stable or is moving toward toxic instability.

Core stability idea

Drug toxicity occurs when drug accumulation exceeds metabolic clearance capacity.

Instability emerges when:

• —drug levels rise rapidly
• —liver clearance declines
• —renal clearance declines
• —sedation or physiological suppression increases
• —drug interactions amplify pharmacologic load
• —intervention occurs too late

The dataset tests interaction reasoning across these signals.

Prediction target

label = 1 → drug toxicity instability label = 0 → stable pharmacologic regulation

Row structure

Each row includes:

• —drug level trajectory
• —liver clearance proxy
• —renal clearance proxy
• —sedation index
• —metabolic rate proxy
• —drug interaction index
• —intervention delay

Decoy variables:

• —lab_noise
• —chart_noise

Evaluation

Predictions must follow:

scenario_id,prediction

Example:

DT101,0 DT102,1

Run:

python scorer.py --predictions predictions.csv --truth data/test.csv --output metrics.json

Metrics produced:

accuracy precision recall f1 confusion matrix dataset integrity diagnostics

Structural Note

This dataset reflects latent stability geometry through observable proxies.

The generator and latent rule structure are not included.

This dataset is part of the Clarus Stability Reasoning Benchmark.

License

MIT