Problem
Parameters
a unique (within an experiment), stable name for the problem; flows into cell labels and result records
creates a fresh problem definition on each call
creates the per-member evaluation-resource factory for one run of the problem. The supplied problem definition is the instance the harness created for the run and will hand to every cell's solver — the member evaluators must be built against that same instance (solutions validate their problem identity), which is why it is an argument rather than something the provider creates for itself
the reference the problem's runs are gapped against; null when no reference exists (runs are then gapped against the best found in the experiment)
descriptive key-value pairs for analysis grouping (e.g. dimension, family, noiseLevel, constrained)