Datasets:

ClarusC64
/

clinical-handoff-integrity-v0.1

scenario_id stringlengths 5 5	collapse_trigger stringclasses 7 values	constraint_state_known float64 0.38 0.92	handoff_latency float64 0.1 0.58	receiving_team_readiness float64 0.36 0.92	trajectory_summary_fidelity float64 0.4 0.9	active_sequence_clarity float64 0.22 0.87	risk_escalation_signal float64 0.18 0.94	context_loss float64 0.08 0.62	prior_policy_failures int64 0 4	handoff_label stringclasses 6 values	label_outcome stringclasses 7 values
TR001	policy_failure	0.92	0.12	0.88	0.9	0.86	0.42	0.1	2	safe_handoff	trajectory_preserved
TR002	policy_failure	0.44	0.18	0.86	0.4	0.36	0.5	0.58	2	unsafe_handoff_missing_constraint	constraint_lost
TR003	policy_failure	0.86	0.2	0.82	0.84	0.22	0.56	0.34	2	unsafe_handoff_fragmented_context	sequence_lost
TR004	rescue_window_closing	0.82	0.52	0.78	0.8	0.74	0.76	0.18	3	unsafe_handoff_too_late	window_missed
TR005	acute_escalation	0.9	0.1	0.92	0.88	0.84	0.86	0.08	3	safe_handoff	trajectory_preserved
TR006	acute_escalation	0.88	0.44	0.48	0.82	0.76	0.88	0.2	3	unsafe_handoff_wrong_priority	receiving_team_unready
TR007	bandwidth_collapse	0.38	0.16	0.84	0.42	0.4	0.72	0.62	3	unsafe_handoff_missing_constraint	constraint_lost
TR008	bandwidth_collapse	0.78	0.24	0.8	0.72	0.68	0.7	0.26	3	safe_handoff	trajectory_preserved
TR009	trajectory_stable	0.8	0.1	0.76	0.78	0.74	0.2	0.16	0	handoff_not_required	local_control_safe
TR010	trajectory_stable	0.62	0.12	0.74	0.6	0.58	0.18	0.3	0	handoff_not_required	local_control_safe
TR011	policy_space_collapse	0.91	0.18	0.9	0.89	0.87	0.92	0.08	4	safe_handoff	trajectory_preserved
TR012	policy_space_collapse	0.76	0.58	0.82	0.74	0.7	0.94	0.28	4	unsafe_handoff_too_late	window_missed
TR013	policy_space_collapse	0.8	0.2	0.4	0.78	0.72	0.9	0.24	4	unsafe_handoff_wrong_priority	receiving_team_unready
TR014	mixed_signal_collapse	0.48	0.22	0.86	0.46	0.44	0.8	0.6	3	unsafe_handoff_missing_constraint	constraint_lost
TR015	mixed_signal_collapse	0.86	0.2	0.84	0.42	0.8	0.78	0.58	3	unsafe_handoff_fragmented_context	summary_fragmented
TR016	rescue_window_closing	0.88	0.46	0.86	0.84	0.82	0.84	0.14	3	unsafe_handoff_too_late	window_missed
TR017	rescue_window_closing	0.9	0.18	0.88	0.86	0.84	0.8	0.1	3	safe_handoff	trajectory_preserved
TR018	policy_failure	0.82	0.16	0.36	0.8	0.78	0.64	0.18	2	unsafe_handoff_wrong_priority	receiving_team_unready
TR019	bandwidth_collapse	0.84	0.18	0.82	0.82	0.28	0.76	0.52	3	unsafe_handoff_fragmented_context	sequence_lost
TR020	trajectory_stable	0.78	0.14	0.8	0.76	0.74	0.22	0.2	0	handoff_not_required	local_control_safe

Clinical Handoff Integrity v0.1

This dataset tests whether a model can determine whether an escalation handoff preserves the patient trajectory.

The task is not diagnosis.

The task is not choosing an intervention.

The task is assessing whether the receiving system inherits the constraint geometry needed to continue safe recovery control.

Core idea

When local control collapses, the patient may need to be handed off.

But a handoff can fail if:

the blocking constraint is not transferred
the active sequence is unclear
the receiving team is not ready
the handoff occurs too late
the trajectory summary is fragmented

The action may be correct, while the structure is lost.

Prediction target

Predict:

handoff_label

Prediction files should contain:

scenario_id,prediction
TE001,safe_handoff
Allowed labels
safe_handoff
unsafe_handoff_missing_constraint
unsafe_handoff_wrong_priority
unsafe_handoff_too_late
unsafe_handoff_fragmented_context
handoff_not_required
Row structure

Each row contains:

collapse trigger
constraint state known
handoff latency
receiving team readiness
trajectory summary fidelity
active sequence clarity
risk escalation signal
context loss
prior policy failures
handoff label
outcome label
Why this is difficult

A model must distinguish:

handoff needed

from:

handoff preserves the trajectory

and from:

handoff destroys the trajectory because the hidden constraint state is lost

This benchmark tests continuity of recovery geometry across control transfer.

Evaluation

Run:

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

The scorer reports:

handoff accuracy
safe handoff accuracy
unsafe handoff accuracy
no handoff accuracy
missing constraint detection
late handoff detection
wrong priority detection
context fragmentation detection
context loss safety score
latency safety score
over handoff 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 handoff integrity and trajectory preservation under escalation.

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

License

MIT

Downloads last month: 25

Number of rows:

Total file size:

14.8 kB