Problem
A named, self-contained problem for benchmarking: fresh problem definitions on demand, fresh per-member evaluation resources on demand, an optional reference solution, and descriptive tags for analysis grouping.
Provisioning a MemberEvaluatorFactoryIfc is what unifies discrete-event and synthetic problems: a DEDS problem supplies a pooled model-backed factory (see ksl.simopt.solvers.concurrent.PooledMemberEvaluatorFactory), a synthetic or static Monte Carlo problem supplies a FunctionMemberEvaluatorFactory — the harness treats them identically.
Both factories must return fresh, independent instances on every call: a benchmark experiment invokes them once per problem run so that no state leaks between experiments, and the returned member-evaluator factory is invoked concurrently on worker threads.
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)