ADR-0001: Functional core and explicit typed dataflow

Context

Ordinary Java applications often let services reload known data and directly own repositories, external clients, retries, and workflow state. In TPF this hides data coupling and prevents the framework from owning boundary semantics.

Decision

TPF uses a functional core with a framework-owned imperative shell. Authored business steps are small transformations over explicit typed inputs. If the pipeline already knows a fact, it carries that fact forward immutably instead of discarding and querying it back. I/O, persistence, retries, correlation, replay, telemetry, transport, and deployment integration stay in framework-owned boundaries.

This decision governs framework/api, framework/runtime-core, framework/deployment, examples, and application authoring guidance.

Rationale

Visible typed dataflow lets the compiler validate what each decision knows and lets the runtime apply consistent durability, replay, and operational policy around pure domain logic.

Consequences

• Authored services do not inject infrastructure merely for convenience.
• Pipeline state may become richer because it preserves facts needed downstream.
• Simple local code need not become a pipeline when TPF semantics add no value.