People assume that things with numbers are inherently objective.

But that's not true!

Tons of hidden assumptions have to be embedded in the decision of what to measure in the first place.

For example, polling for elections has to make adjusting assumptions about response rates of different subpopulations, etc.

The results look objectively true, but are actually largely created based on the baseline assumptions of the model, what the electorate "should" look like based on what it's looked like in the past.

But the underlying context that those assumptions are about could change invisibly to the model; it's outside the model.

And then the numbers would be fundamentally wrong later.