Datasets:

ClarusC64
/

fusion-cross-field-decoherence-precursor-detection-v0.1

id string	device_class string	window_ms int64	rolling_mag_temp_corr float64	rolling_temp_impurity_corr float64	rolling_mag_impurity_corr float64	magnetic_fluctuation_index float64	temperature_gradient_index float64	impurity_concentration_ppm int64	turbulence_index float64	confinement_quality_h98 float64	drift_rate float64	stability_margin_index float64	decoherence_precursor_flag int64	onset_ms_before_quench int64	precursor_pattern_label string	disruption_risk_score float64	confidence_score float64	notes string	constraints string	gold_checklist string
FDEC-001	tokamak	50	0.86	0.82	0.8	0.18	0.72	42	0.22	0.98	0.002	0.74	0	0	stable-band	0.12	0.88	stable window	<=280 words	flag+onset+pattern+risk+confidence
FDEC-002	tokamak	50	0.78	0.7	0.72	0.24	0.66	60	0.3	0.93	0.01	0.62	1	65	mag-temp slip	0.58	0.83	mag-temp decouples first	<=280 words	flag+onset+pattern+risk+confidence
FDEC-003	stellarator	80	0.84	0.86	0.83	0.16	0.7	38	0.2	1.01	0.003	0.78	0	0	stable-band	0.1	0.9	robust coupling	<=280 words	flag+onset+pattern+risk+confidence
FDEC-004	tokamak	50	0.7	0.62	0.6	0.3	0.6	90	0.38	0.88	0.016	0.54	1	42	impurity-led drift	0.66	0.81	impurity coupling breaks	<=280 words	flag+onset+pattern+risk+confidence
FDEC-005	spherical_tokamak	60	0.74	0.66	0.69	0.28	0.62	72	0.34	0.91	0.012	0.6	1	55	dual-surface drift	0.62	0.82	two correlations sliding	<=280 words	flag+onset+pattern+risk+confidence
FDEC-006	tokamak	40	0.62	0.58	0.55	0.33	0.58	110	0.44	0.84	0.02	0.46	1	28	rapid decoherence	0.78	0.79	near quench	<=280 words	flag+onset+pattern+risk+confidence

What this dataset tests

Whether a system can detect early decoherence
across three linked stability surfaces:

The goal is not to detect quench.
The goal is to detect the precursor window
10 to 100 milliseconds before quench.

Required outputs

Use case

Layer 2 of COH-FUSION-001
Plasma Stability Coherence Collapse.

Why it matters

Plasma disruption is preceded by
cross-field correlation loss.

This dataset trains systems
to flag that loss early.