sync — Genesis / I-GEN Surrogate Detector & Validation Workspace

This repository hosts the Genesis Surrogate validation workspace for the C3-ICM / LRAI Project Genesis research program. It implements and calibrates the I-GEN (Inertia-Generation) hard-detector stack used to score grid-telemetry risk in the Connotative Governance and Zero Trust Architectures white paper (Part IV).

The repository pairs a surrogate enterprise/grid environment (IAM, telemetry, ticket metadata, adversarial injection harness) with a calibrated detector layer driven by the Sajadi et al. (2022) frequency-response dataset.

Latest update — Calibrated I-GEN detector

The I-GEN detector weights have been recalibrated against the frequency-response heatmap dataset using a grouped-holdout protocol (group = source_file) to prevent leakage across simulation traces. The fixed-weight scoring rule is now the production interpretation; the gradient-boosted variant was set aside despite a near-perfect cross-validation AUC because it tracked synthetic-rule artifacts and is harder to audit in deployment.

Key artifacts:

• genesis_surrogate/hard_detectors.py — H_DEEP evaluation and I_GEN scoring (uses the calibrated weights and threshold).
• genesis_surrogate/calibrated_igen_weights_final.json — exported weights, threshold, and holdout metrics.
• genesis_surrogate/README_VALIDATION.md — calibration methodology and leakage diagnostics.

Validation summary

Grouped holdout (held-out simulation traces, 20% of groups, random_state=42):

Metric	Value
Logistic regression (grouped-holdout) ROC AUC	0.9355
Fixed-weight (grouped-holdout) ROC AUC	0.8997
Fixed-weight PR AUC	0.9116
Fixed-weight accuracy	0.8050
Precision	0.7799
Recall	0.8407
F1	0.8091
I_GEN_THRESHOLD (raw)	0.190566

See genesis_surrogate/README_VALIDATION.md for the full calibration write-up, leakage/duplicate diagnostics, and the production interpretation of the threshold.

Repository layout

.
├── genesis_surrogate/             # I-GEN detector, calibration, IAM + injection harness
│   ├── hard_detectors.py
│   ├── calibrated_igen_weights_final.json
│   ├── weight_calibration.ipynb
│   ├── telemetry_emitter.py
│   ├── iam_role_layer.py
│   ├── adversarial_injection_test.py
│   ├── README.md
│   └── README_VALIDATION.md
├── freq_response_data.zip         # Sajadi et al. (2022) frequency response data
├── freq_response_code.zip
├── heatmap_data.zip               # Calibration dataset for I-GEN weights
├── heatmap_code.zip
└── mech_system_code_data.zip

White paper

The accompanying Connotative Governance and Zero Trust Architectures white paper (C3-ICM / LRAI, Project Genesis) is intentionally frozen as a citable reference document. White-paper files are not part of this documentation cleanup and should not be edited as part of repository maintenance.

Citation

Frequency-response simulation data: Sajadi, A. et al. (2022). Nature Communications 13, 2490. doi:10.1038/s41467-022-30164-3.