Solver benchmark: ZeitFinder v1611 compared with CP-SAT
In the documented benchmark dated 22 October 2026, ZeitFinder v1611 reached the target before CP-SAT in 13 of 17 cases both solvers solved without additional gaps. The inputs were constructed availability patterns for individual timetables.
What did the benchmark compare?
CP-SAT is a powerful optimisation solver in Google’s open-source OR-Tools toolkit. Its applications include complex scheduling and staff planning.
The supplied report compares ZeitFinder v1611 with CP-SAT for automatically generating individual timetables. The shared measurement target was time to the first complete, valid plan with no additional internal idle time. A scheduled lunch break does not count as an additional gap.
Results of the documented series
| Metric | Result reported |
|---|---|
| Total configurations | 24: 30, 40, 50 and 60 people, three datasets each, with and without a lunch break. |
| Proven infeasible | 6 configurations with 50 people; required time exceeds capacity in a mandatory window. |
| Both solvers reach the zero-additional-gap target | 17 configurations within the 3,000 ms limit per solver call. |
| Which solver reaches the target first? | ZeitFinder: 13 cases. CP-SAT: 4 cases. |
| Paired geometric time ratio | CP-SAT / ZeitFinder: 3.81. Calculated only over the 17 cases where both reached the zero-additional-gap target. |
The PDF shows neither a paired zero-additional-gap result nor individual measurements for the remaining configuration. The 13-of-17 result must therefore not be read as a success rate across all cases.
Three reported individual examples
In three selected individual examples, ZeitFinder found a valid plan without additional gaps roughly 12 to 27 times faster than CP-SAT.
The report highlights one case each for 30, 40 and 60 people: the largest measured advantage at each size. These examples are not averages.
| People / dataset | ZeitFinder time to zero additional gaps | CP-SAT time to zero additional gaps |
|---|---|---|
| 30 people / 1103, with break | 95.6 ms | 2600.8 ms |
| 40 people / 1102, with break | 63.0 ms | 1295.7 ms |
| 60 people / 1102, with break | 169.9 ms | 2066.8 ms |
Measurement method described in the report
Datasets were fixed and hashed before the first solver run. Availability patterns are deliberately constructed variants of a known scheduling structure, not a representative sample of real customer cases.
Identical inputs, serial execution in alternating order and a fresh process per call. CP-SAT used one search thread and search seed 1. There was one measured run per solver and configuration, 48 calls in total.
The 3,000 ms limit includes model building and preparation. Imports, process startup and data transfer are excluded. The figures measure solver computation, not browser waiting time or collecting responses.
According to the report, all recorded plans were separately validated. The comparison covers the zero-additional-gap component; day count, balance and preferences did not form an identical shared overall optimisation objective.
What does the evidence establish?
The figures support the advantage described in this limited test series. A single run per case does not establish statistical reliability or worldwide superiority. The report says CPU type and exclusive host use cannot be verified. Measurements apply to v1611, not to every later ZeitFinder version.
Original report and full case table
The source is the supplied PDF “ZeitFinder – Neue konstruierte Praxisfälle”, dated 22 October 2026. Figures on this page were taken from it and arithmetically checked. The underlying solver runs were not rerun for this publication.
Read the report (PDF, German, 3 pages)Create your own availability poll
Set the days and time ranges, then invite your group to enter their availability.
Create an availability poll