Architecture Health

Run — 2026-07-13T10:59:28Z

2026-07-13T10:59:28Z

2Candidates

0Strong

2Worth Exploring

src/gsd_lean/state/__init__.py

Cluster: gsd_lean.state
Why shallow: 73 lines of pure re-export — zero implementation, an __all__ of 30+ names forwarded verbatim from state.config, state.detector, state.planning; interface surface (the name list a reader must scan) exceeds implementation depth (none).
Deletion test: The only runtime consumer is cli/app.py (its single from gsd_lean.state import (...) block). Internal modules bypass the facade already — state/config.py and workflow/engine.py import from gsd_lean.state.planning directly, and the test suite imports submodules directly. Deleting the facade forces app.py to point its import block at .planning / .config instead: a mechanical edit, no logic reconcentrates. Complexity vanishes for the sole caller; all other code is unaffected because it never used the facade.
Verdict: candidate for deletion
Dep class: in-process
Confidence: Worth-exploring
Rationale: Contested classification — a package __init__ re-export is a legitimate public-API idiom, but here it is used by exactly one caller and bypassed by internal modules and tests, so its value as an API boundary is unproven rather than clearly zero.

src/gsd_lean/workflow/__init__.py

Cluster: gsd_lean.workflow
Why shallow: 19-line pure re-export of 7 names from workflow.engine; no logic, interface = implementation ≈ 0.
Deletion test: Only consumer is cli/app.py (from gsd_lean.workflow import PreconditionError and ... import transition as workflow_transition). tests/test_workflow.py already imports gsd_lean.workflow.engine directly, bypassing the facade. Deleting it forces app.py to import from gsd_lean.workflow.engine: trivial rename, no complexity migrates. Pass-through.
Verdict: candidate for deletion
Dep class: in-process
Confidence: Worth-exploring
Rationale: Same idiom-vs-shallowness ambiguity — single caller, tests bypass it; deleting is a judgment call about intended public surface, not an unambiguous win.

Note: the two flagged facades are the same pattern (single-caller re-export bypassed internally). If acted on, treat them as one cleanup decision about whether the package public API is worth an explicit facade at all — not two independent refactors.