Guardrails (Authoritative)
Deterministic Core / Probabilistic Edge
- The core engine is deterministic and authoritative.
- AI is probabilistic and advisory.
- The boundary must never blur.
Deterministic core (authoritative)
cassian test verdict + exit code semantics- scenario execution + fault orchestration
- results.json and resolved topology
- fail-fast validation rules
Probabilistic edge (advisory)
cassian ai- evidence summarization
- coverage suggestions
- human guidance
Gate-first UX (locked)
cassian test:- runs from clean state
- destroys any existing lab
- executes deterministically
- returns binary verdict
- produces authoritative artifacts
cassian run:- explicitly non-authoritative
- for exploration/debug only
No drift / no silent changes
- Defaults must not change without an explicit milestone update and verification updates.
- Any ambiguity must error before execution.
- Any schema addition must be deterministic and validated.
What Cassian Gate is NOT
- not a general-purpose lab tool
- not chaos engineering
- not a packet analyzer / Wireshark replacement
- not AI-driven validation or auto-remediation