The sequence shields the consultant’s judgment with custody and mathematics—without publishing the internals of the control plane.
Client telemetry enters a hermetic retention environment and is hashed at the door.
We ingest agreed identity and endpoint telemetry into an isolated retention environment. SHA-256 seals each artifact at intake so the chain of custody starts at the first millisecond—not after a report is drafted.
Analysts work the evidence under dual-control, fail-closed human authority.
Practitioners inspect sealed evidence in an internal clean room. Agentic reasoning, if used, only proposes. Dual control (human-in-the-loop) authorizes. The architecture fails closed by default, so prompt injection or data leakage classes of failure do not become an unsupervised path to a conclusion or an external model.
AETHER applies logical USCM assertions. No language model issues the verdict.
Collected evidence is submitted to AETHER, the notarial evaluation engine. AETHER does not use language models. It applies strict logical assertions over infrastructure state—for example, whether conditional-access policy actually intercepted a failed sign-in—and emits a binary or not-evaluable result.
A human Ed25519 signature binds contract, evidence hashes and the mathematical result.
The client receives a professional opinion with a cryptographic seal (Ed25519). The artifact irrevocably binds the contract, the evidence and the USCM outcome—an auditable record meant to sit beside the client’s own SOC 2, PCI-DSS or NIST CSF 2.0 evaluations, without AegisCore Labs claiming to issue those certifications.
Technical evidence is the primary artifact of truth. Hashes, provenance and gaps are recorded at intake. Policies and interviews are context, not proof. Missing or stale evidence is stated as a gap—never filled with a model’s guess.
A qualified practitioner interprets evidence in operational context: asset criticality, residual risk, and what the organization can actually change. Findings and priorities are reviewed before they are issued. Dual control applies wherever an agentic helper proposes a next step.
Reasoning engines may assist inspection under fail-closed, human-in-the-loop control. They do not score USCM controls. AETHER applies deterministic assertions. The delivery manifesto is signed by a human operator (Ed25519). Language-model text is never the verdict.
The client contracts expert judgment. Cryptographic custody and deterministic evaluation make that judgment auditable.
