Useful decisions.
Precisely stated completion.
Eight unchanged historical application originals returned checked feasible answers under source F. A separate flugpl proof and the current-G LA experiment have their own budgets and conclusions.
Keep every outcome.
| Original | Application | Objective | Conclusion |
|---|---|---|---|
| flugpl | Airline planning | 1,201,500 | Global optimum proved separately |
| wachplan | Crew shift planning | −8 | Numerical feasibility; optimum unproved |
| air03 | Airline crew scheduling | 340,160 | Native original feasibility; optimum unproved |
| timtab1 | Public transport scheduling | 780,545.9999997802 | Numerical feasibility; optimum unproved |
| glass4 | Nesting | 1,757,014,465.5020328 | Numerical feasibility; optimum unproved |
| bays29 | City tour | 2,020 | Native tour check; optimum unproved |
| berlin52 | City tour | 7,657 | Native tour check; optimum unproved |
| gasprod1-2 | Gas production planning | 983.0403402505526 | Numerical feasibility; optimum unproved |
The objectives are in model units. Numerical feasibility, native checks and a global optimum are different result states. These are not live customer deployments.
A proof-aware
application experiment.
221 named street-pair locations were organized into 28 model batches. Every batch received a checked global optimum; total exact objective and lower bound were 7,896,074,357. The experiment’s red-exposure proxy changed from 17.067 to 9.101 seconds under its declared equal-split baseline.
A larger batch first returned unknown. After an 8,192-cell proof limit was exhausted, external repartitioning of independent decisions preserved the mathematics and enabled completion. This is a reported workflow finding, not a claim of an already integrated router repair.
Historical 2013–2016 traffic counts and a simplified independent-location model. No field congestion improvement is claimed. Queue, capacity, turn, pedestrian, offset and spillback effects were not modeled.