write the rules down.
get the plan and its gap.
Get the best plan the solver finds within your rules, with the gap that says how far from best it can be. In whitebox optimization the goal and the limits become a mathematical model. A solver searches the possible plans systematically until the gap to the best possible plan is zero or time runs out.
your planners already know the rules.
They say them out loud every day. kint's decision agent writes them down as a model, line by line, in a form your experts can check.
| Your planner says | The model reads | Type |
|---|---|---|
| Every ship leaves on time. | finish(vessel) ≤ departure(vessel), for every vessel | Hard rule |
| A driver works at most eight hours per shift. | sum of assigned hours(driver, shift) ≤ 8 | Hard rule |
| Two cranes never work the same bay. | at most one crane per bay and hour | Logic rule |
| Only certified drivers on the big cranes. | assign(driver, crane) = 0 unless certified | Qualification |
| Use as few machines as possible. | minimize cranes + carriers in use | Goal |
two numbers that meet.
While it searches, the solver tracks the best plan found so far and a bound that no plan can beat. When the two meet, the gap is zero: no better plan exists for the model as stated, within the solver's tolerance.
- Best plan found. Gets better as the search goes on.
- Bound. What the best possible plan could reach at most.
- Gap zero. No better plan exists for the model as stated, within the solver's tolerance.
If you stop the search early, the gap tells you exactly how far the plan can be from the best one at most. Rules of thumb and heuristics cannot give your board that number.
a plan, and the reasons behind it.
The solver answers more than the question you asked. It tells you which rule decides the outcome and what changing it would be worth.
- the planConcrete assignments, sequences and quantities. After release it goes back only through an API you control.
- the optimality gapAt 0.0 %, no better plan exists for the model as stated, within the solver's tolerance. Any other value states the largest possible distance to the best plan.
- the binding limitThe rule that holds the result back. That is where negotiation, investment or a policy change pays off.
- sensitivityWhat one more unit of a resource is worth: one more crane hour, one more euro of budget, one more truck.
- what-ifChange a limit, keep the rest, solve again. The difference is shown next to the original plan.
- bound and gapStored with the run, next to data version, model version and who released the plan, in the record.
whitebox is the right choice when
- the goal is clear: cost, time, machines, emissions
- the limits can be written down: capacities, shifts, budgets, laws
- you must show that the plan is the best one, not only a good one
- the same decision comes back every day, week or season
kint picks the formulation with your team
| Method | Structure | Typical decisions |
|---|---|---|
| LP | linear programming | Blending, allocation, flows through a network. |
| MIP | mixed-integer programming | Shift plans, routes, which sites to open. |
| CP | constraint programming | Sequencing, timetables, job shops. |
| QP | quadratic programming | Portfolios, balancing loads, smooth curves. |
bring your rules. leave with the plan and its gap.
Describe one decision in your own words or bring an existing model. We show you what the solver finds and why.
- Methods
- LP · MIP · CP · QP
- Gap
- Bound and gap, stored with the run
- Solvers
- Open-source LP, MIP and CP solvers included · commercial solvers supported · technology
- Contact
- hello@kint.pro