{
"suite": {
"suite_name": "Clarus Adversarial Stability Benchmark Suite",
"suite_version": "v1.0",
"suite_goal": "Evaluate state-space stability intelligence in AI systems",
"design_rule": "Labels must emerge from multi-variable interactions rather than single-feature correlations"
},

"benchmark": {
"benchmark_name": "CASSES — Adversarial Instability Detection",
"benchmark_version": "v0.3",
"repo": "ClarusC64/casses-adversarial-instability-detection-v0.3",
"task_type": "instability_detection"
},

"capability_tested": "Detect system instability under deceptive signals and adversarial traps",

"target_variable": "label_system_instability",

"input_features": [
"system_pressure",
"buffer_capacity",
"intervention_lag",
"subsystem_coupling",
"drift_gradient",
"boundary_distance",
"recovery_feasibility",
"regime_competition_ratio"
],

"hidden_generation_fields": [
"true_instability_state",
"trajectory_shift",
"trap_type"
],

"adversarial_traps": [
"false_stability",
"boundary_masking",
"trajectory_aliasing",
"temporal_alias",
"intervention_decoy"
],

"evaluation": {
"metrics": [
"accuracy",
"precision",
"recall",
"f1",
"trap_accuracy",
"false_stability_accuracy",
"boundary_masking_accuracy",
"trajectory_aliasing_accuracy"
],
"primary_metric": "trap_accuracy",
"secondary_metric": "false_stability_accuracy"
}
}