smartroute-ai

Extend the BaseSolver class in apps/api/solvers/, register it in SOLVER_REGISTRY, and implement get_distance_matrix() so asymmetric matrix problems work. Validate the route with validate_route() and test against both the 10-city and matrix_tsp fixtures before any database write.

For under 20 nodes use OR-Tools for exact solutions; for 20-100 nodes use GA, ACO, or Tabu Search; for 100-500 nodes use ACO, SA, or Tabu. For multi-objective needs use NSGA-II, and always run Benchmark Mode to let the comparison engine decide.

The rule-based engine computes a weighted score: 40% solution quality, 20% runtime, 15% stability across three seeds, 15% scalability by solver class, and 10% constraint fit. No LLM is used inside the recommendation engine.

Yes, asymmetric TSP is supported through the matrix_tsp problem type. When distanceMatrix is provided in the canonical schema, solvers use it directly and ignore x/y coordinates, requiring no schema changes.

All stochastic solvers must accept a seed parameter. Run twice with seed=42 to verify reproducibility, and check that convergence is monotonically non-increasing and validate_route() passes before returning results.

Never advance a phase without human confirmation that all acceptance criteria pass. Each phase has a hard gate, such as valid Hamiltonian cycles for Phase 003 or 80% pytest coverage for Phase 006.