Governed State Contracts

An agent can assemble context, recognise a pattern and suggest what should happen next. None of that tells you whether the proposed action is allowed now, for this customer, under this policy, with this evidence. That decision belongs to the system that owns the state.

That is the idea behind governed state contracts. A user interface, integration or agent may propose an action. The operational authority checks current state, permissions, policy and invariants; it either executes the transition or returns a structured reason why it cannot. The audit and resulting events are part of the same decision, not an afterthought.

The model does not argue against agents or broad analytical context. It gives them a safe place to operate. Context can be distributed: reports, documents, graphs, search indexes and vector stores all help a person or model understand what might be true. Authority remains with the capability that can validate and execute the change against current truth.

The accompanying Designing the Stack series develops the architectural argument. This site holds the durable reference model behind it.

The reference model

• The model explains the separation between broad reasoning and authoritative state change.
• Operational authority defines the system that decides whether an action may proceed.
• Analytical context explains why useful read models do not become competing sources of authority.
• Governed actions describe how intent, validation, rejection, audit and events fit together.
• Capability contracts set out what an action contract must say beyond the shape of a payload.
• Agents as proposers applies the same rule to agents, user interfaces, integrations and automation.

The machine-readable form of a capability contract is being developed separately. This reference model deliberately describes the architecture before prescribing a particular implementation.