Benchmark comparison workbench
Choose an evaluation for the question you need to answer, then inspect results under matching conditions. This workbench starts with numerical verification and measured physical validation.
See the animated benchmark guide for the model types, motions, and errors each test examines. These teaching illustrations are separate from the result records below.
No published runs yet. The protocols below are real proposals with cited references; there are no invented demonstration scores. Local result files stay in this browser tab and are not uploaded or saved by this page. Closing or reloading the tab clears them.
Compare the protocols
| Benchmark | Question | Reference | Readiness | Evidence |
|---|---|---|---|---|
| Ideal pendulum | Does the implementation approach the prescribed equations as the timestep decreases, without introducing unacceptable energy error? | analytical | protocol ready | 0 published runs |
| Planar pushing | Can a model predict held-out object motion under a measured pushing action, with uncertainty and preprocessing made explicit? | measured | candidate | 0 published runs |
Read the benchmark cards and their sources before interpreting results. Passing an analytical check does not establish physical validity; agreement in one measured regime does not establish general reliability.
Inspect results
Loading the result workbench…
Interpreting comparisons
Compare only records using the same benchmark, protocol version, reference, physical scenario, and execution settings. A declared comparison group is the contributor's assertion of compatibility; the workbench checks matching metadata, not the underlying experiment. Author-run experiments, published claims, and independently reproduced observations remain separately labeled. None of these labels is a quality certification.
Differences are descriptive, not tests of statistical significance. Consult the sample count and uncertainty notes, and retain individual trials for repeated experiments. Missing metrics stay missing. There is no aggregate ranking or automatic pass/fail threshold.
This first interface compares implementations or revisions at fixed settings. Timestep-convergence curves and speed–accuracy plots are not yet implemented; different timesteps remain in separate groups even though the pendulum protocol calls for a timestep sweep.
For private JWM use, prepare a result file outside this public repository and open it using the local-file control above. Publishing is a separate, deliberate editorial step. Read the result format and privacy boundary.