Milestones
List view
Introduce optional AI assistance while maintaining deterministic control. Goal: Treat the LLM strictly as a suggestion generator, never as a source of truth. Scope: - strict suggestion contract - settings_sha256 binding between suggestions and configuration - deterministic merge rules (Heuristics > LLM) - suggestion verification pipeline - closed-world validation for IDs - year-range and filename validation - invalid suggestion detection - minimal JSON repair (JSONFix) Exit criteria: - suggestions are verifiable, schema-bound, and configuration-bound - invalid suggestions are rejected deterministically - heuristics remain authoritative over probabilistic outputs
No due date•0/9 issues closedImplement the deterministic backbone of docflow. Goal: Ensure that identical inputs and identical configuration always produce identical results. Scope: - strict settings loader with required configuration files - Pydantic validation with extra=forbid - closed-world configuration rules - stable configuration fingerprint (settings_sha256) - deterministic document type heuristics - doc_type → archive area routing - normalization and deterministic tie-breaking logic Exit criteria: - deterministic document analysis works without probabilistic components - settings and routing are validated strictly - configuration-dependent behavior is reproducible
No due date•0/8 issues closedEstablish the strict testing baseline for docflow before implementing core functionality. Goal: Ensure that architectural invariants are enforced by tests rather than by convention. Scope: - pytest test infrastructure - CLI contract tests - deterministic heuristics tests - suggestion pipeline governance tests - audit logging verification - integration smoke tests for CLI stability - testing guide documentation Exit criteria: - the core architectural assumptions are covered by tests - CLI behavior is protected by contract and smoke tests - regression risk is reduced before further implementation phases
No due date•6/7 issues closed