Skip to main content
Adapt-1 can carry evidence and constraints across calls through the domainless State gateway, or work within a Domain that gives a task a stable contract and a scope for its evidence and learned state. Use State when retaining and revising evidence adds useful context to later requests. Its reconstructive retrieval recovers context from accumulated evidence. An application can combine this with a language model for prose interpretation and generation, and optional RAG for original source passages. The continuity guide explains the State workflow. Choose a Domain when observations and measured outcomes should improve a defined prediction or decision. Examples include numerical forecasting, learning which action works in a given situation, and acquiring control sequences. One application can combine these learning tasks with its language workflow.

​
The objects in your integration

A Domain retains its applicable state across environment episodes until an explicit lifecycle operation changes it. This includes parameters learned by neural and other trained models. RLAS and learned structure explains how model fitting and structural revision contribute to later decisions.

​
Decide what teaches the result

Completed observations can teach a predictor. To revise action values from outcomes, preserve decision-time context and the executed policy. Learning from evidence and outcomes covers the full set of configurations.

​
Structure and learning schedule

Discovery forms eligible learner-facing structure from evidence. Authored structure makes selected task relationships explicit before use. Both support an empty task-specific learned state at run start. The learning schedule determines whether earlier acquired state is loaded and whether learning continues during operation. Choose the schedule separately: define the starting state and whether writes are allowed during the run. Frozen use computes actions from current observations using unchanged learned state. See Learning schedules.

​
Inspect what happened

Check the request status and learner-admission diagnostics. Inspect the later result to see whether it used the applicable state. Use the route’s result status, evidence, and learning diagnostics. Confidence fields have route-specific meanings; use the documented support and abstention behavior when deciding whether to act. See Explain a result.

Mechanisms

See how observations and outcomes change Core’s learned structure.

​
Contract and application boundaries

The public task meanings, event schema, result contract, available actions, outcome semantics, and boundaries stay stable within the declared workflow. Supporting observations, discovered input and representation structure, learned relationships, trained model parameters, active alternatives, policy or transition state, uncertainty, and support can change through eligible writes.State can form during a zero-start run or a separate acquisition phase. Decide whether later use permits further learning or only reads retained state. See Choose a learning setup.
A Domain defines the public task boundary. Discovery can form eligible input projections, fields, combinations, temporal relationships, state-dependent action values, and other learner-facing structure from evidence.Use authored structure when it must be fixed or reviewed in advance.
Your application supplies the external environment, final commitment, execution, safety checks, and consequence measurement. It also supplies perception, format translation, or candidate generation when the selected route requires them. An adapter should expose only information available at that point and withhold the answer or mapping the learner is expected to form.Keep the returned result, the application’s committed value, and the operation actually executed separately when validation or overrides can change them.
Reuse a Domain when public meanings, result units and schema, consequence semantics, and state lifecycle remain compatible. Use separate Domains when those boundaries differ, even if observations come from the same product or dataset. Each Domain operation addresses a domain_id under the authenticated owner.Keep current observations and external state in the application. Use State-gateway continuity for evidence and constraints that can be queried without a Domain contract.
Predicting an observable consequence, judging a committed decision’s utility, scoring movement toward a bounded objective, assigning delayed credit, and comparing typed hypotheses are different relationships. A Domain can combine compatible learning relationships. Keep their targets and feedback meanings distinct.Choose how a Domain learns describes the distinctions. Explain a result describes support, uncertainty, and learner diagnostics.