Intent and SSIL
Intent records what an institution wants to do and why. Sovereign Source AI stores intent as a typed object. The object links to evidence, requirements, constraints, authority, and measurable outcomes. Sovereign Source Intent Language (SSIL) is the proposed declarative language that describes these links. Later plans and execution steps use the intent as a portable contract. The contract stays valid when models, applications, or infrastructure change.
Intent as a typed object
An intent has these parts:
- an objective
- an accountable owner
- an origin
- the evidence that supports it
- desired outcomes
- guardrails
- a lifecycle status
Its relationships connect it to the requirements and authorizations that the objective needs.
For example, an organization wants more efficient fulfillment. It must also keep accuracy and employee safety. Those outcomes and guardrails go in the intent. A specific programming language, model, or execution environment is a later implementation choice.
The type system keeps these differences clear:
- a proposed objective and an approved objective
- a desired outcome and an observed result
- a constraint and a suggestion
The system keeps these differences across interfaces, workflows, and organizational boundaries.
People must still interpret a typed intent and use judgment. Its structure lets reviewers examine that interpretation. It also gives them a defined object to accept, revise, or reject.
Intent provenance
Intent can come from these sources:
- documents
- contracts
- operational problems
- analytics
- regulations
- events
- ordinary language
These sources can support a candidate intent. They do not approve it.
The provenance record shows each step from the source material to the proposed objective. It includes each observation and interpretation. It also includes each human edit and approval. A later reviewer can find the evidence that supported the decision and the person who accepted the intent.
If an assumption changes, reviewers use this record. They can tell an objective based on an observed operational problem from one based on an inference that someone revised later. The institution can also explain why a system exists. A model does not have to rebuild lost institutional knowledge.
From intent to governed work
An approved intent becomes requirements. The requirements state the conditions that must be true for the objective to be met. Architecture and plans connect those requirements to capabilities and execution choices.
A Sovereign Work Order is the typed, authorized unit of work that carries this context into execution. It can include these items:
- outputs
- policy
- permitted models and tools
- network limits
- execution conditions
- dependencies
- acceptance tests
- evidence requirements
- approvals
- budgets
A large objective can split into nested work orders. Each child work order stays linked to the parent intent. A child work order gets the same authority as its parent, or less. The split itself does not add any permission.
Requirements and work orders also keep the return path. Verification results link to requirements, and outcome evidence links to the intent. Thus the institution can check that the work was delivered and that it met its purpose.
SSIL as the intent language
SSIL describes desired outcomes together with these items:
- requirements
- constraints
- authority
- capabilities
- evidence
- obligations
- approvals
Its proposed execution form is:
Intent → Requirements → Constraints → Capabilities → Authority → Plan → Actions → Evidence → Outcome
The proposed language has four areas:
- SSIL-Core covers foundational objects and authority.
- SSIL-Flow describes dependencies and workflow state.
- SSIL-Market covers offers, contracts, funding, and settlement.
- SSIL-Guild covers domain ontologies, credentials, reputation, and obligations.
These language areas are a proposal. They are not a completed standard. Their common purpose is to keep what the institution means as intent becomes executable work.
Reference build
A reference build has these parts:
- an intent registry
- source and evidence links
- requirement objects
- authority checks
- a planner
- work-order contracts
An intent compiler translates the declarative form into these outputs:
- policy checks
- capability requirements
- approval tasks
- execution obligations
Runtime adapters use those outputs. If you replace an adapter, the organization keeps the original intent and its full lineage.
From the research
- “Intent should survive implementation changes.” Its objective stays the same when the implementation changes. Four-plane architecture notes
- “Compile organizational intent into governed, portable execution.” This is the central proposal of SSIL. Master summary