s
Frameworks

Key Models

Four structural frameworks that translate organisational governance logic into tools for AI alignment and systems design. Each model addresses a different dimension of the control problem.

Framework 01

The QUAD Model

Questions → Understanding → Action → Decisions

The QUAD model is a structural decision framework that enforces a specific sequence: inquiry before comprehension, comprehension before action, action before commitment. It prevents the most common failure mode in high-stakes environments — acting before understanding.

In AI alignment terms, QUAD provides a governance architecture for how autonomous agents should process information and escalate decisions. It maps directly to the question of when an AI system should act independently versus when it should defer to human oversight.

QQuestions
UUnderstanding
AAction
DDecisions
Framework 02

The Levels Model

Constraint Hierarchy → Emergent Freedom

The central insight of the Levels model is counterintuitive: well-designed constraints do not restrict freedom — they create it. A chess player is not limited by the rules of chess. The rules are what make chess possible.

Applied to AI alignment, Levels provides a framework for designing constraint hierarchies where each layer of rules enables a higher level of productive autonomy. It addresses the fundamental tension between control and capability — showing that the two are not in opposition when structure is properly designed.

L3Emergent freedom & capability
L2Operational constraints
L1Foundational rules & boundaries
Framework 03

The Abacus Model

Calibrating Human Oversight Across the Autonomy Spectrum

Not all decisions require the same degree of human oversight. The Abacus model provides a calibration system — a sliding scale that maps the appropriate level of human control required at each stage of AI-human interaction.

At one end: full human control. At the other: full AI autonomy. The Abacus framework defines the criteria for positioning any given decision along this spectrum, based on reversibility, consequence magnitude, uncertainty level, and stakeholder impact.

Full Human Control
Full AI Autonomy
Calibration: reversibility · consequence · uncertainty · stakeholder impact
Framework 04

The Mirror System

Observational Restraint & Self-Reflective Governance

The Mirror System examines how self-observation creates natural governance. In human organisations, the most effective control mechanisms are not external impositions — they are internalised feedback loops where the system observes and corrects itself.

Applied to AI, the Mirror framework provides a model for building systems that monitor their own behaviour against defined constraints — creating a form of artificial restraint that does not require constant external supervision. It is governance through self-awareness rather than governance through enforcement.

Observe
Evaluate
Restrain
Adapt