Datasets:

ClarusC64
/

clinical-recovery-policy-switching-v0.1

scenario_id stringlengths 5 5	current_policy stringclasses 7 values	policy_fit_score float64 0.18 0.82	regime_transition_score float64 0.12 0.88	policy_failure_signal float64 0.1 0.84	rescue_window_remaining float64 0.12 0.78	adaptive_bandwidth float64 0.21 0.7	constraint_volatility float64 0.22 0.91	recent_feedback_quality stringclasses 3 values	local_action_available stringclasses 2 values	policy_decision stringclasses 7 values	label_outcome stringclasses 10 values
TR001	recovery_sequence	0.82	0.12	0.1	0.78	0.64	0.22	good	yes	continue_recovery_policy	recovery
TR002	recovery_sequence	0.38	0.72	0.68	0.34	0.36	0.74	poor	yes	switch_to_stabilization_policy	stabilized_after_switch
TR003	recovery_sequence	0.31	0.84	0.8	0.18	0.24	0.86	poor	yes	escalate_to_acute_policy	acute_escalation_needed
TR004	iron_rebuild	0.76	0.2	0.18	0.7	0.58	0.3	good	yes	continue_recovery_policy	recovery
TR005	iron_rebuild	0.42	0.62	0.58	0.42	0.46	0.65	mixed	yes	switch_to_resource_rebuild_policy	recovered_after_reframe
TR006	sleep_repair	0.8	0.14	0.16	0.76	0.6	0.28	good	yes	continue_recovery_policy	recovery
TR007	sleep_repair	0.36	0.7	0.64	0.4	0.39	0.72	poor	yes	switch_to_load_shedding_policy	stabilized_after_load_shedding
TR008	load_shedding	0.74	0.18	0.2	0.68	0.44	0.34	good	no	continue_recovery_policy	slow_recovery
TR009	load_shedding	0.28	0.88	0.84	0.12	0.21	0.91	poor	no	escalate_to_acute_policy	acute_escalation_needed
TR010	stabilization	0.71	0.22	0.25	0.6	0.42	0.38	mixed	no	continue_recovery_policy	stabilized
TR011	stabilization	0.33	0.78	0.76	0.2	0.26	0.82	poor	no	escalate_to_acute_policy	acute_escalation_needed
TR012	recovery_sequence	0.58	0.48	0.45	0.52	0.5	0.56	mixed	yes	switch_to_sleep_repair_policy	recovered_after_sleep_repair
TR013	recovery_sequence	0.55	0.52	0.48	0.5	0.48	0.58	mixed	yes	switch_to_resource_rebuild_policy	recovered_after_resource_rebuild
TR014	recovery_sequence	0.46	0.6	0.55	0.46	0.34	0.62	poor	yes	switch_to_load_shedding_policy	stabilized_after_load_shedding
TR015	over_intervention	0.2	0.4	0.72	0.58	0.7	0.36	poor	no	stop_intervention_policy	recovered_after_pause
TR016	over_intervention	0.18	0.5	0.8	0.42	0.62	0.44	poor	no	stop_intervention_policy	recovered_after_pause
TR017	iron_rebuild	0.34	0.74	0.7	0.28	0.3	0.78	poor	yes	switch_to_stabilization_policy	stabilized_after_switch
TR018	sleep_repair	0.3	0.82	0.78	0.22	0.28	0.84	poor	yes	switch_to_stabilization_policy	stabilized_after_switch
TR019	resource_rebuild	0.78	0.16	0.14	0.74	0.57	0.26	good	yes	continue_recovery_policy	recovery
TR020	resource_rebuild	0.4	0.66	0.62	0.36	0.41	0.68	mixed	yes	switch_to_load_shedding_policy	stabilized_after_load_shedding

Clinical Recovery Policy Switching v0.1

This dataset tests whether a model can detect when the current clinical recovery policy is no longer the right frame.

The task is not diagnosis.

The task is not choosing the next local intervention.

The task is deciding whether the whole recovery policy should continue, switch, stop, or escalate.

Core idea

A local intervention may still look plausible while the overall policy has failed.

Example:

iron_first may still be locally reasonable,
but if the system has crossed into bandwidth collapse,
the correct policy is stabilization or acute escalation.

Prediction target

Predict:

policy_decision

Prediction files should contain:

scenario_id,prediction
TE001,continue_recovery_policy
Allowed labels
continue_recovery_policy
switch_to_stabilization_policy
switch_to_resource_rebuild_policy
switch_to_sleep_repair_policy
switch_to_load_shedding_policy
escalate_to_acute_policy
stop_intervention_policy
Row structure

Each row contains:

current policy
policy fit score
regime transition score
policy failure signal
rescue window remaining
adaptive bandwidth
constraint volatility
recent feedback quality
local action availability
gold policy decision
label outcome
Why this is difficult

Most models optimize inside the current frame.

This benchmark asks whether the model can detect frame failure.

The correct answer may be:

stop optimizing the old policy
switch policy
escalate
or stop intervening
Evaluation

Run:

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

The scorer reports:

decision accuracy
switch accuracy
continue accuracy
acute escalation accuracy
stop policy accuracy
regime transition safety score
rescue window policy score
over-switch 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 recovery policy switching under regime transition and frame failure.

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

License

MIT

Downloads last month: 30

Number of rows:

Total file size:

14.4 kB