A harness that works for one model is a harness that has to be rewritten for the next. The durable version talks to a named system through one interface, so swapping the model is a config change, not a code change.
About 1 min of reading
Run one harness against two systems
Write the run file header
Before you run it, say out loud what it will print. Then open the answer.
Append a run for system A
Before you run it, say out loud what it will print. Then open the answer.
Compare two systems in one pass
Before you run it, say out loud what it will print. Then open the answer.
The harness is shared by every project that has a model. Each project supplies a set and a bar, and inherits the loop.
The grade is never cached, the output is. Caching the grade freezes a grader that may have been fixed since.
The harness calls answer(system, input). Every model that can satisfy that signature can be scored by the same set.
The run writes one line per case: system, case, grade, reason, version. Every later comparison is a read of that file.
Grading inside the call means the grader cannot be changed alone. Separating the two lets a grader be fixed without touching the model.
Each one is defined here on this site. Open the entry for the full definition and the other words it points at.