Create benchmark.json

benchmark.json

@@ -0,0 +1,59 @@
+{
+"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"
+}
+}