Datasets:

ClarusC64
/

clinical-rescue-window-timing-v0.1

scenario_id stringlengths 5 5	dominant_constraint stringclasses 6 values	bandwidth_proxy float64 0.22 0.84	constraint_coupling float64 0.12 0.92	deterioration_rate float64 0.05 0.72	rescue_window_width float64 0.08 0.9	feedback_lag float64 0.05 0.55	intervention_absorbability float64 0.18 0.88	current_risk stringclasses 3 values	prior_step_response stringclasses 5 values	timing_decision stringclasses 6 values
TR001	iron	0.62	0.42	0.18	0.72	0.1	0.78	medium	improving	monitor_short_interval
TR002	iron	0.48	0.7	0.42	0.38	0.25	0.61	high	worsening	act_now
TR003	iron	0.31	0.86	0.6	0.18	0.35	0.34	high	worsening	escalate_immediately
TR004	sleep	0.54	0.5	0.25	0.6	0.15	0.72	medium	stable	act_now
TR005	sleep	0.38	0.78	0.5	0.28	0.3	0.44	high	worsening	act_now
TR006	sleep	0.24	0.88	0.66	0.12	0.45	0.22	high	worsening	escalate_immediately
TR007	bandwidth	0.28	0.91	0.55	0.22	0.25	0.31	high	worsening	rollback_now
TR008	bandwidth	0.35	0.82	0.46	0.32	0.2	0.42	high	poor_response	rollback_now
TR009	thyroid	0.7	0.26	0.14	0.78	0.08	0.82	low	stable	delay_until_bandwidth_improves
TR010	thyroid	0.56	0.44	0.3	0.46	0.18	0.66	medium	worsening	act_now
TR011	mixed	0.43	0.76	0.48	0.26	0.28	0.51	high	worsening	act_now
TR012	mixed	0.29	0.88	0.68	0.1	0.5	0.2	high	collapse	missed_rescue_window
TR013	none	0.84	0.12	0.05	0.9	0.05	0.88	low	stable	monitor_short_interval
TR014	none	0.78	0.18	0.08	0.82	0.06	0.8	low	improving	monitor_short_interval
TR015	iron	0.52	0.62	0.35	0.44	0.22	0.64	medium	stable	act_now
TR016	sleep	0.46	0.64	0.36	0.4	0.24	0.58	medium	worsening	act_now
TR017	bandwidth	0.22	0.92	0.72	0.08	0.55	0.18	high	collapse	missed_rescue_window
TR018	thyroid	0.64	0.34	0.16	0.7	0.1	0.74	medium	stable	monitor_short_interval
TR019	mixed	0.37	0.8	0.58	0.2	0.34	0.37	high	worsening	escalate_immediately
TR020	bandwidth	0.41	0.72	0.4	0.36	0.18	0.49	high	poor_response	rollback_now

Clinical Rescue Window Timing v0.1

This dataset tests whether a model can identify intervention timing under a narrowing rescue window.

The task is not diagnosis.

The task is not choosing what to do.

The task is deciding when action is still absorbable.

Core idea

The same intervention can be:

too early
well-timed
too late
no longer absorbable

The model must infer whether to:

act_now
monitor_short_interval
delay_until_bandwidth_improves
rollback_now
escalate_immediately
missed_rescue_window

Prediction target

Predict:

timing_decision

Prediction files should contain:

scenario_id,prediction
TE001,monitor_short_interval
Row structure

Each row contains:

dominant constraint
adaptive bandwidth
constraint coupling
deterioration rate
rescue window width
feedback lag
intervention absorbability
current risk
prior step response
timing decision
Why this is difficult

A static model may choose the correct intervention but miss the timing.

This benchmark tests whether the model can distinguish:

safe to monitor

from:

last safe moment to act

and:

rescue window already missed
Evaluation

Run:

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

The scorer reports:

timing accuracy
urgent accuracy
delay accuracy
missed window detection
rollback accuracy
rescue window safety score
over escalation resistance
macro precision
macro recall
macro F1
structural score

The main metric is:

structural_score
Structural Note

This dataset is synthetic.

It is designed to test timing control under narrowing clinical rescue windows.

It is not medical advice and should not be used for clinical decision-making.

License

MIT

Downloads last month: 14

Number of rows:

Total file size:

12 kB