Sprint 14 - Minimal factory and registry¶

Objective. Enable dynamic component instantiation only after concrete specs and examples justify it.

Deliverable. Components can be instantiated from a small registry without introducing a plugin system.

—

Decisions already made¶

Registry and factory stay local and explicit.
No plugin loading is introduced in this sprint.
Dynamic creation must not hide component ownership or lifecycle registration.
If specifications are involved, they are already-parsed data structures, not a new schema system.

Scope¶

The likely scope is a small registry plus a thin factory that can create one component, or a deterministic batch, from registered component types and constructor data.

Spec loader boundary¶

If a loader exists in this sprint, keep it minimal. Do not add schema validation or a Pydantic dependency here.

To decide during sprint planning¶

Exact registry API and duplicate-registration behavior.
Whether batch creation is necessary in the first version.
Whether any spec loader belongs in this sprint at all.

—

Validation¶

[ ] Component types can be registered and retrieved.
[ ] Duplicate registration behavior is documented and tested.
[ ] Factory creates components with constructor kwargs.
[ ] Unknown component types raise clear errors.
[ ] Batch creation is deterministic.

—

Risks and mitigation¶

Risk: premature schema binding. Keep specs free-form until real examples show the shape.

Risk: plugin-system creep. Registry and factory stay local and explicit; no dynamic plugin loading.

—

Dependencies¶

Requires: stable component constructors.
Requires: examples proving repeated component instantiation is painful.
Benefits from: parameter/configuration work if it exists.

—

Scope boundaries¶

In scope:

local component registry
thin factory
clear errors

Out of scope:

plugin loading
schema validation
full application spec model
runtime component removal

—

Success signal¶

[ ] Dynamic creation reduces boilerplate without hiding component ownership.
[ ] The registry remains a simple tool, not a second abstraction layer.