ClarusC64/clinical_safety_coherence_eval_v0.2

Clinical Safety Coherence Eval v0.2

What this is

A small dataset that tests one question:

Can you detect when a clinical safety system is moving toward coherence failure, not just carrying safety pressure?

This repo focuses on safety coherence evaluation.

It models a system where:

• safety signal strength may weaken
• protocol alignment may drift
• latent hazard pressure may rise
• decision friction may slow clean response before failure becomes obvious

Run this first

Generate baseline predictions:

python baseline_heuristic.py data/tester.csv predictions.csv

Score them:

python scorer.py data/tester.csv predictions.csv

That is enough to see the full evaluation loop.

You will get:

standard metrics

trajectory detection performance

safety coherence failure detection errors

What to try next

Replace the baseline.

Build your own model.

Output a file like:

id,prediction_score
0,0.12
1,0.81
2,0.67

Then run:

python scorer.py data/tester.csv your_predictions.csv
What matters

Not just accuracy.

The key signals are:

recall_trajectory_deterioration_detection

false_stable_trajectory_rate

These tell you:

are you catching systems that are getting worse

are you missing hidden safety coherence failure

Data

Each row represents a clinical safety state.

Core variables:

safety_signal_strength

protocol_alignment_score

latent_hazard_pressure

decision_friction_index

drift_gradient

coherence_stability_score

escalation_readiness_index

oversight_integrity_score

Target:

label_safety_coherence_failure

Important distinction

There are two different components in this repo.

scorer.py

evaluates predictions

domain-agnostic

works across all v0.2 datasets

does not generate predictions

baseline_heuristic.py

generates predictions

domain-specific

uses the variables in this dataset

Do not reuse the heuristic across datasets.

It is only a local reference.

What changed from v0.1

v0.1:

static safety coherence classification

v0.2:

adds direction via drift_gradient

This allows you to separate:

pressured but stabilizing safety states

pressured and deteriorating safety states

Why this exists

Most models answer:

what is happening now

This tests:

where the safety state is going

That difference is where hidden failure appears early.

Files

data/train.csv — training data

data/tester.csv — evaluation data

scorer.py — canonical evaluation script

baseline_heuristic.py — dataset-specific reference model

README.md — dataset card

Evaluation

Primary metric:

recall_trajectory_deterioration_detection

Secondary metric:

false_stable_trajectory_rate

Standard metrics are also reported:

accuracy

precision

recall

f1

The scorer supports binary predictions or score-based predictions.

License

MIT

Structural Note

Clarus datasets are structural instruments.

They are designed to expose instability geometry, not just predict isolated outcomes.

This v0.2 repo adds directional state movement so the dataset can separate static safety pressure from active deterioration in clinical safety coherence.

Production Deployment

This dataset can be used in:

safety review research

protocol drift monitoring

escalation pathway benchmarking

clinical governance simulations

model benchmarking for trajectory-aware safety reasoning

It is suitable for research and prototyping.

It is not a substitute for live clinical judgment.

Enterprise & Research Collaboration

Clarus builds datasets for:

instability detection

trajectory tracking

intervention reasoning

These structures are not domain-bound.

They apply wherever systems move toward or away from failure.

Applicable domains include:

healthcare systems

financial markets

energy infrastructure

logistics networks

artificial intelligence systems

manufacturing systems

supply chains

climate systems

Any environment where:

capacity and demand interact

delays and coupling exist

trajectory determines outcome

This dataset is one instance of a general stability framework.