Execution Architecture · KYE State Engine

Agent Governance for the AI execution stack — KYE owns 2 layers, integrates with 8.

KYE Protocol sits at exactly two layers of the AI execution stack: Authority Resolution (deciding whether a proposed action is admissible) and Evidence Finality (proving the decision is replay-derivable from public keys alone). KYE CONSUMES sensed events, capability profiles, hardware attestations. KYE INTEGRATES with policy authoring bodies. KYE PRODUCES recurring-failure feedback. KYE PROVIDES the substrate for contestability and legal accountability. KYE never moves money, never grades models, never runs hardware, never authors policy, never owns dispute UI. The 10-named-layer moat, made explicit.

Anti-dilution lock · for CISOs and procurement

KYE Protocol is NOT — and exactly what KYE is.

Every regulated-buyer conversation begins with a positioning question: is this an observability tool, a GRC platform, a model-eval lab, a dispute-resolution product? The answer is: none of them. KYE is the Authority Resolution + Evidence Finality layer in a 10-layer execution stack — one layer it OWNS, one layer it OWNS, eight it INTEGRATES with.

KYE is NOT an observability platform. KYE CONSUMES sensed events from Datadog, Honeycomb, Grafana, OpenObserve.
KYE is NOT a model evaluation company. KYE CONSUMES capability profiles from Apollo Research, Anthropic Evals, OpenAI Evals, METR.
KYE is NOT a compute-governance platform. KYE CONSUMES hardware attestations via the Machine State Adapter from AMD-SEV, Intel SGX/TDX, AWS Nitro, GCP Confidential VM.
KYE is NOT a GRC tool. KYE PRESERVES and VERSIONS policy across customer + regulator authoring bodies.
KYE is NOT a dispute resolution system. KYE PROVIDES the Evidence Pack + Replay Proof substrate that makes contestability possible.
KYE is NOT a legal accountability framework. KYE PROVIDES the named-person reasonable-steps Evidence Pack that liability frameworks consume.
10-layer architecture · canonical diagram

The 10 layers, KYE's role at each, the integration partners.

Every arrow into Authority Resolution is a KYE-CONSUMED signal; every arrow out is either a decision envelope (to enforcement) or an evidence substrate (to contestability, liability, assurance). The convergence point is the KYE State Engine.

flowchart LR
  subgraph KYE_OWNED["KYE OWNS"]
    L3["3 — Influence State"]
    L7["7 — Response/Enforcement (DECIDES)"]
    L10["10 — Assurance (KYE Seal)"]
  end
  subgraph KYE_CONSUMES["KYE CONSUMES"]
    L1["1 — Sensing"]
    L2["2 — Capability State"]
    L4["4 — Machine State"]
  end
  subgraph KYE_INTEGRATES["KYE INTEGRATES"]
    L5["5 — Policy Evolution"]
  end
  subgraph KYE_PRODUCES["KYE PRODUCES"]
    L6["6 — Learning/Feedback"]
  end
  subgraph KYE_SUBSTRATE["KYE PROVIDES SUBSTRATE"]
    L8["8 — Appeal/Contestability"]
    L9["9 — Liability/Accountability"]
  end
  L1 --> L7
  L2 --> L7
  L3 --> L7
  L4 --> L7
  L5 --> L7
  L7 --> L6
  L7 -->|Evidence Pack + Replay Proof| L8
  L7 -->|Evidence Pack| L9
  L7 -->|Attestation window| L10
  L7 -->|Enforcement instruction| ENF[("Gateway / Rail / Engine ENFORCES")]
    

Canonical source: constitution/55-EXECUTION-LAYERS-ARCHITECTURE.md · layer manifest at internal.

Per-layer breakdown

The 10 named layers — role · partners · contract.

1 · Sensing Layer

KYE role · CONSUMES

Observability of model invocations, tool calls, latency, error rates, drift signals. KYE pulls structured trace as kye.evidence.tool_call.v1.

Partners: Datadog, Honeycomb, Grafana, OpenObserve.

2 · Capability State Layer

KYE role · CONSUMES

Model evaluations, red-team results, capability monitoring. KYE asserts the capability profile is in-window at decision time.

Partners: Apollo Research, Anthropic Evals, OpenAI Evals, METR.

3 · Influence State Layer KYE OWNS

KYE role · OWNS

Distinguishes rubber-stamped human approval from substantive review. Native KYE primitive: AI Influence State. The most important new primitive identified in the 2026-05-30 execution-layers analysis.

Schema: kye.ai_influence_state.v1.

4 · Machine State Layer

KYE role · CONSUMES

Compute governance + hardware attestation via the Machine State Adapter — validates TEE attestations and binds the attested code-hash into the decision-bound Evidence Pack.

Partners: AMD SEV-SNP, Intel SGX/TDX, AWS Nitro Enclaves, Google Confidential VM.

5 · Policy Evolution Layer

KYE role · INTEGRATES

Governance bodies author and version policy. KYE preserves the version graph and replays decisions against the policy state at decision time.

Partners: Customer policy teams, FCA, MHRA, European Commission.

6 · Learning / Feedback Layer

KYE role · PRODUCES

Recurring-failure detection across Evidence Packs. KYE emits the structured feedback signal; downstream MRM / fine-tune loops consume it.

Partners: Customer MRM teams, fine-tune aggregators.

7 · Response / Enforcement Layer KYE DECIDES

KYE role · DECIDES; partners ENFORCE

API gateways, payment rails, workflow engines, agent orchestrators carry out the side effect. KYE emits the Action Admissibility decision envelope; the partner enforces.

Partners: Stripe, Modulr, Temporal, Claude Agent SDK.

8 · Appeal / Contestability Layer

KYE role · PROVIDES SUBSTRATE

KYE serves the Evidence Pack + Replay Proof that make a decision contestable. A separate product owns the dispute UI / case management.

Partners: Customer case-management UI, ombudsman portals.

9 · Liability / Accountability Layer

KYE role · PROVIDES SUBSTRATE

Legal accountability frameworks (UK SM&CR, EU AI Act accountable-officer schemes, insurance) consume the named-person reasonable-steps Evidence Pack.

Partners: UK SM&CR regime, EU AI Act schemes, insurance underwriting.

10 · Assurance / Certification Layer KYE OWNS

KYE role · OWNS

Periodic third-party assurance via KYE Seal — binds the operating control set to declared regulatory frameworks for the attestation window.

Partners: ISO/IEC 42001 conformity assessment bodies, SOC 2 auditors.