Governing the passage from intention to AI execution.
RA-153.1-2 is a reference architecture that separates understanding, capabilities, cognitive strategy, authority, execution and evidence so that technical capability to act is never confused with authority to act.
From key to code — then to testing
1 — Open the passage
Application Key 153.1-2, in the 153/9 form, carries the operational principle: never confuse capability, authority, action and evidence.
Read Key 153.1-2Explore the Key’s ArchitectureText version2 — Understand the architecture
The Working Paper presents the thesis, state of the art, six planes, invariants and falsification program.
Read WP006 v1.63 — Implement
The specification defines R0→R6, the state machine, Human Gate, mandates, revocation, replay, errors and conformance profiles.
Read specification v0.2JSON Schemas v0.2OpenAPI v0.24 — Execute
A minimal Python/FastAPI implementation demonstrates the full pipeline, including human escalation and R6 evidence.
Download v0.25 — Attempt to falsify
RA-Bench runs adversarial scenarios with non-compensatory gates covering authority, scope, admissibility, replay, freshness and evidence.
Open RA-BenchDownload v0.2From Key 153 to RA-153.1-2
The ZEON Key 153.1-2 is the operational form that establishes the direction of governed passage. RA-153.1-2 is the engineering architecture derived from this key: it translates that direction into implementable planes, contracts, invariants and interfaces. The key and architecture are related but distinct.
Scientific status
RA-153.1-2 is not presented as an established standard or an industrially validated architecture. WP006 v1.6 proposes a consolidated reference architecture, related work integrated through 4 October 2026 and testable hypotheses. The specification and reference implementation make the proposal implementable; RA-Bench tests selected invariants. Independent results and replication remain to be produced.
For implementers
Two implementations may differ greatly in technology while remaining RA-conformant if they preserve the same boundaries: capability resolution does not authorize; the cognitive plane produces no domain effect; only EXECUTE creates a mandate; resources are selected from an already admissible space; effects remain bounded; evidence binds the mandate to the observed action.
Scientific update v1.6
Read the integrated update
Independent application case
Open App_IA_Citoyenne (French application)View the RA alignment profile
Integrity and versions
Release manifest · SHA-256 · Deployment README · Archives
License: LCZ-ZS1 — canonical reference: zeons.org/zeon/licence.html. Published prototypes are experimental and not production-ready.