moheetsubudhi-isb/optimization-toolkit

Choose between an exact solver and a heuristic for a hard planning problem, and set honest expectations on speed, plan quality and explainability. Use when a routing, delivery, scheduling, rostering, packing, knapsack, assignment or network model runs too long or never finishes; when someone asks for the optimal plan for hundreds or thousands of stops, jobs or items; mentions the travelling salesman or vehicle routing problem, greedy algorithms, local or neighbourhood search, genetic algorithms, simulated annealing, tabu search, OR-Tools routing, a MIP gap, rolling horizon or decomposition; or when planners need a good plan in minutes rather than a perfect plan tomorrow.

Write yes/no business rules as correct linear constraints in a mixed-integer model, then prove them right by enumeration. Use whenever a model needs if-then or only-if logic, either/or choices, at most k or exactly one of a set, AND/OR conditions, a fixed cost that applies only when something is used, a minimum batch size, min, max or absolute-value terms, deciding which shift or time slot a value falls in, big-M, indicator or binary variables, or linearising a product of variables. Also use when a MIP returns plans that break a rule everyone believed was modelled, or runs slowly because of a huge big-M.

Turn a business decision into an optimisation model and a recommendation a business owner can act on. Use whenever someone needs to decide how much to make, buy, stock, staff or ship; where to open sites or depots; how to split a budget, capacity, people or inventory; or asks for the best plan, maximum profit or minimum cost given limits, a production mix, facility location, capacity planning, prescriptive analytics, linear programming or a MIP. Also use when someone asks whether an investment in capacity, overtime, marketing or extra supply is worth it, what extra capacity is worth, how to read a solver's output, or needs a model requirements document for developers. Not for a model that already exists and only runs too slowly, and not for deciding who gets cut back when supply cannot cover demand.

Plan how to test an optimisation model so its plans can be trusted in live operations. Use when someone has built, or is building, a linear programme, MIP, scheduler, router, allocator or other solver-based planning tool and asks how to test or validate it, how to know the model is right, what test cases or regression tests to write, why a plan looks wrong, how to check that business rules are really enforced, how to run the model alongside the current manual process, or how to get planners and managers to sign off on model output before go-live.

Decide who gets what when there is not enough to go round, and explain the trade-off between efficiency and fairness to the people affected. Use when a planning or optimisation model is infeasible because demand exceeds supply; when a plan starves one customer, store, region or team while another gets everything; when allocating scarce stock, budget, capacity, staff, delivery slots, appointments or medicines; when someone asks for a fair, equal, proportional or priority-based allocation, min-max, a penalty for unmet demand or slack variables; or when stakeholders reject an optimiser's plan as unfair even though it has the lowest total cost. Not for testing, validating or signing off an allocation model that already works.