| --- |
| library_name: scikit-learn |
| license: apache-2.0 |
| tags: |
| - cybersecurity |
| - verified-autonomy |
| - autonomous-defense |
| - verification |
| - scikit-learn |
| datasets: |
| - avernaut/verified-autonomy-scenarios |
| metrics: |
| - accuracy |
| --- |
| |
| # Security Action Verifier |
|
|
| An Avernaut **reference baseline** for the six-gate verified-autonomy decision used in the companion synthetic dataset. |
|
|
| The model consumes six Boolean inputs: |
|
|
| 1. V1 Authorization |
| 2. V2 Target correctness |
| 3. V3 Reachability |
| 4. V4 Availability |
| 5. V5 Blast radius |
| 6. V6 Effectiveness |
|
|
| The expected rule is intentionally explicit: |
|
|
| **VERIFIED iff every required gate passes.** |
|
|
| ## Why publish a model for a deterministic rule? |
|
|
| This artifact exists for interoperability, demonstrations, testing pipelines, and model-vs-rule experiments. It is **not** a replacement for deterministic enforcement logic. |
|
|
| For production security controls, implement and audit the verification rule directly rather than relying on a learned approximation. |
|
|
| ## Evaluation |
|
|
| The model was checked against the complete 64-row Boolean truth table. |
|
|
| - Truth-table accuracy: **1.000** |
| - Cases: **64** |
|
|
| ## Related dataset |
|
|
| `avernaut/verified-autonomy-scenarios` |
|
|
| ## Related project |
|
|
| INVERIQ: https://github.com/avernaut/inveriq |
|
|
| ## Safety note |
|
|
| This model does not establish that a cyber action is safe in the real world. Real deployments require authoritative telemetry, policy, topology, service dependencies, rollback mechanisms, and independent enforcement controls. |
|
|
