Optimization, feasibility, and solver quality

The deterministic engine first normalizes and validates the model. Missing required capacities, demands, objective scores, broken references, or unsupported numeric inputs can make analysis invalid. An infeasible result means the evaluated constraints cannot all be satisfied in that run. A heuristic search that finds no incumbent does not prove infeasibility.

Manual scenarios receive an evaluated result. For optimization scenarios, a bounded exact search may return optimal only when it completes its search envelope and reports optimalityProven: true. If the candidate envelope or search-node budget is exceeded, the engine may return a deterministic heuristic feasible incumbent. Call it “highest modeled” only when the exact proof is present; otherwise say it is a candidate portfolio that may be improved.

Optimization is conditional and risk-neutral under the encoded scoring policy. It does not discover omitted projects, validate estimates, infer resource calendars, solve detailed construction scheduling, establish legal compliance, or authorize funding. Preserve source fingerprints and rerun after any material edit.

Next: allocation, utilization, and bottlenecks.