The Incident Intelligence Layer

While the platform provides the infrastructure, the Incident Intelligence Layer provides the governance and verification. This layer is responsible for the interpretation of signals and the enforcement of safety.

1. Incident Intelligence Taxonomy

We categorize incident verification into five distinct levels of verification, ensuring the AI-driven OODA loop is both accurate and safe.

Level	Scope	Verification Mechanism
Contract	API Schema Alignment	Pydantic validation of Finding and ActionProposal objects.
Synthetic	RCA Accuracy	Replaying historical incident logs into the SafetyAnalyzer.
Property	Logic Resilience	Hypothesis property-based tests for quorum and nonce logic.
Adversarial	Security Hardening	Red-team suites for signature forgery and replay attacks.
Recovery	Post-Remediation Verification	Rollback gates evaluated based on telemetry regression signals.

2. Test Suite Mapping

We maintain a flat, verified test suite mapped directly to our verification taxonomy:

Test File	Verification Level	Scope & Objective
test_safety.py	Contract & Synthetic	Verifies standard Pydantic models (contract) and runs end-to-end simulation pipelines to verify root-cause analysis (synthetic).
test_property_based.py	Property	Uses Hypothesis to prove quorum logic remains resilient under random nonce generation and varying system clock drifts.
test_adversarial.py	Adversarial	Evaluates response of safety modules against simulated attacks, including signature tampering, nonce reuse, and invalid agent keys.
test_recovery.py	Recovery	Simulates post-actuation telemetry regressions (e.g. error rates) to verify that rollback gates trigger correctly.
test_kms_signer.py	Integration	Verifies the Cloud KMS asymmetric signing adapter and its safe fallback to local in-memory keys.

3. The Verification Workflow

Incident validation is not a post-facto audit; it is a pre-actuation gate.

1. Static Analysis: Ruff and Pydantic enforce schema integrity before an agent is allowed to sign a proposal.
2. Logic Proof: Property-based tests prove that the VotingValidator is mathematically sound under extreme clock-skew (±28s).
3. Adversarial Gate: The test suite includes intentional forgery attempts. If the system fails to block a forged signature, the CI/CD pipeline is automatically frozen.

4. Failure Domain: Incident Intelligence Plane

If the Incident Intelligence plane fails (e.g., Gemini API outage or Consensus timeout), the system defaults to the Platform Safety Boundary.

• Intelligence Plane (Soft Fail): Returns a RETRY_WITH_BACKOFF or HUMAN_ESCALATION status.
• Platform Plane (Hard Fail): The SafetyGate blocks all automated state-changes until the intelligence plane is recovered.

This separation ensures that a hallucination or failure in the AI layer never results in an unverified infrastructure change.