Automated Intelligence
Architectural Class Definition
Definition
A system architecture that performs reasoning and execution tasks where:
every action belongs to an execution class — the minimal authorization unit, where two actions share a class if and only if authorization for one logically implies authorization for the other — explicitly granted by a principal external to the system's execution scope prior to execution;
authorization is class-specific: an authorized objective does not authorize execution classes not explicitly granted, even if those classes would serve the objective;
behavioral boundaries are enforced by a mandatory pre-execution validation gate operating at a privilege level unreachable by the reasoning engine's write operations; composite actions must be atomically decomposed into constituent execution classes prior to gate validation — if any component class lacks valid authorization, the composite is rejected in its entirety before any constituent action executes;
execution begins at the first operation that accesses, observes, or modifies any state outside the reasoning engine's transient working memory — pre-execution planning may not involve external state access; any operation touching external state, including read-only queries, belongs to an execution class requiring prior authorization;
all state changes are logged, bounded in scope, and recoverable within defined constraints.
Core Structural Terms
Execution Class
The minimal unit of authorization. Two actions share an execution class if and only if authorization for one logically implies authorization for the other. Authorization boundaries are defined by decision implication, not by capability taxonomy.
Principal
An entity that can issue verifiable authorization tokens and revoke authorization state, and that exists outside the execution scope of the system being authorized. The system cannot grant its own authorizations. The principal relationship is not reflexive.
Transient Working Memory
The internal non-persisted reasoning context of the system. Strictly bounded to context that:
- Is not accessible to external systems without the reasoning engine's active cooperation
- Does not persist beyond the current reasoning pass
- Is not externally inspectable without active disclosure
The following are external state, not transient working memory:
- Retrieval stores
- Vector databases
- Tool persistence
- File storage
- Model-side persistence
- Cached recallable outputs
- Hidden scratch files
- Tool buffers
Memory boundary expansion is a silent autonomy vector. This definition must not expand.
Pre-Execution Validation Gate
A mandatory checkpoint that must be passed before any action executes. Properties:
- Operates at a privilege level unreachable by the reasoning engine's write operations
- Validates action class membership against current authorization state
- Receives only action descriptor and authorization state — not reasoning engine justification
- Cannot be modified by the reasoning engine
Automation / Autonomy Boundary
The structural boundary between Automated Intelligence and autonomous systems:
Can the system initiate new goal-directed execution without fresh external authorization?
- No → automation remains. System is operating within Automated Intelligence class.
- Yes → autonomy begins. System has exited the Automated Intelligence class.
Serving an authorized objective by unauthorized means constitutes autonomy beginning. Authorization is class-specific, not intent-specific.
Key Structural Insight
Automated Intelligence is defined by authorization topology, not by behavior.
Not: what the system does.
But: what the system is permitted to initiate.
This distinction is the architectural foundation of the class.
Derivation Chain Summary
| Version | Status | Key Closure |
|---|---|---|
| v0 | Draft | Arch proposal — initial structure, three unearned terms |
| v1 | Draft | Forge stress test — authority, predictability, reversibility addressed |
| v2 | Draft | Solen tightening — principal defined, goal class removed, enforcement gap identified |
| v3 | Draft | Forge closure — execution class defined by authorization implication, privilege isolation introduced |
| v4 | Draft | Solen confirmation — atomic composite rejection added |
| v5 | Locked | Forge closure — execution boundary defined by state scope; Solen final clearance |
Solen Final Clearance Statement
"Draft v5 holds. No structural contradictions detected. No unresolved load-bearing gaps remain. This is no longer an exploratory draft. This is the filing version."
Immutability Notice
This document is version-locked.
No edits. No rewrites. No formatting drift. No silent cleanup.
If corrections are required, create v6. Do not overwrite v5.
Open Threads
- Implementation mapping against Reiva's current architecture
- Terminology collision scan — confirm "Automated Intelligence" is not a prior formal standard or regulated classification
- Classification ladder evaluation — Automated / Assisted / Autonomous Intelligence tiers
- Whitepaper pathway — academic priority establishment
- Patent surface — specific implementation claims in Reiva