Target Architecture¶
The target architecture is an incremental separation of the existing working system. No rewrite is authorized.
+-------------------+
| Goal Resolution |
+---------+---------+
|
v
+-------------------+
| Candidate Service |
+---------+---------+
|
+---------------+---------------+
| | |
v v v
Taxonomy Graph Evidence
| | |
+---------------+---------------+
v
+-------------------+
| Scoring |
+---------+---------+
v
+-------------------+
| Calibration |
+---------+---------+
v
+-------------------+
| Constraints |
+---------+---------+
v
+-------------------+
| Recommendation |
+---------+---------+
|
+------------+------------+
| |
v v
Learning Path Architecture
| Inference
| |
+------------+------------+
v
Blueprint
Separation strategy¶
- Establish behavioral tests around existing production behavior.
- Extract pure domain models and scoring contracts without changing semantics.
- Extract repositories for taxonomy, graph, benchmark, and calibration data.
- Move recommendation orchestration into an application service.
- Make API, CLI, and MCP adapters consume the application service rather than each other.
- Add versioned provenance to recommendation results.
- Replace static architecture selection incrementally with capability/graph inference while retaining the current mapping as a fallback.
Constraints¶
- Existing public CLI commands remain available.
- Existing API endpoints remain available unless a versioned contract requires otherwise.
- Existing MCP tools remain available.
- Generated graph data remains an artifact of its generator and is never edited manually to fix runtime behavior.
- Deterministic recommendation content is mandatory.
- New dependencies require a demonstrated need.