Performance benchmarks:

I am sorry, but I don't understand the test code itself. This changes the return in a MockModel for one of the model reporters. But if a value is invalid, should that not always raise an exception and thus be handled inside the data collector?

To be clear, the current code makes no sense to me. Jus trying to understand the wider context within which this PR is being used.

Let me clarify:

The original code has a bug: assert ValueError is a no-op. It doesn't raise an exception - it just checks that the ValueError class is truthy (which it always is). The code was always returning None, the assert line did nothing.

The test expects None: Line 165-166 explicitly asserts that model_calc_fail returns None:

for element in data_collector.model_vars["model_calc_fail"]:
    assert element is None

DataCollector doesn't handle exceptions: Looking at datacollection.py:378, there's no try/except around reporter calls. If a reporter raises an exception, it would crash the simulation.

My PR just removes the no-op: I'm not changing the design - just removing the useless assert ValueError and making the intent clear with a comment.

However, you raise a valid design question: Should DataCollector catch reporter exceptions and store None (or some error value) instead? That would be a separate enhancement. My PR just fixes the misleading no-op assertion in the existing test.

However, you raise a valid design question: Should DataCollector catch reporter exceptions and store None (or some error value) instead? That would be a separate enhancement. My PR just fixes the misleading no-op assertion in the existing test.

Thanks for the clarification. On this, let's not touch the current data collector and instead focus on the emerging ideas arround dataset and collector. I am going to merge this pr in the meantime.