Compare completed results
Read compatible saved results without inventing a new run.
Compare two completed, accessible results with the same task/evaluation contract. Reading or comparing saved results does not start another solver or upgrade their verification.
Choose the correct comparison
repeat: the same procedure and compatible protocol for a repeated run.algorithms: supported Moving AI A* versus Dijkstra, with the same pinned cases and evaluation contract.
Use the result IDs returned by the actual runs. The service checks source/profile versions, tasks, metrics and compatible limits. It refuses incompatible results; it does not align coordinate frames or normalize different robot tasks on your behalf.
comparison = client.call("compare_results", {
"left_id": "replace-with-left-result-id",
"right_id": "replace-with-right-result-id",
"mode": "algorithms",
})
In the UI, choose both completed results and the comparison mode under Compare completed runs. A saved result is explicitly a recording. It remains subject to current workspace and source access.
Interpret the report
Read compatible metrics and the recorded outcomes, not just client exit status. A valid path and matching published length can be compared within this grid task; they do not certify body dynamics, flight performance or physical safety. A technical error is different from a robotic-task failure.