Derived suites
A suite is a fixed set of public real-world datasets, taken as a snapshot and run through the standardized harness, so that Ufinq and the other methods see the same data with the same time. It is the only place where a real-world observation becomes a comparison.
No suite exists yet. The first one is proposed from the public base once enough eligible datasets are there; its datasets, its rules and its results will be listed here.
How a suite is made
- Admission. A dataset enters a suite only from a public project whose owner granted the rights to use it this way, and never from one under a non-commercial licence. Each dataset records its origin analysis, the snapshot's hash, the rights basis and the date.
- Dev and holdout. Half the datasets are held out: no experiment and no tuning run sees them. Only a public run of a release does.
- Versions. A suite version is fixed once a run exists on it; adding or retiring a dataset makes a new version, and a retired dataset stays listed with its reason.
- Results come from the same harness as the standardized benchmark: every method gets the same data, time and seeds. The verdict reads as there.