Exact truth

Problems whose optimum can be derived rather than estimated, used to grade methods against ground truth: the Bayes-optimal score of a reader of residues, the least height of an integer polynomial vanishing at 1 to a given order, and set-valued prediction with an exact posterior.

• RULER — evaluations that know their optimum
• FLATTEN — the least height flattened to a given depth
• SETS — set-valued prediction against its exact optimum