The last stage asks the question the whole program turns on: do the scores carry the concept? Stance Is Not a Construct runs that test on the field’s newest instrument — large language models as annotators — and finds they catch surface stance while missing the theoretical structure beneath it. Political Geometry goes one step further and opens the instrument itself, asking how ideology is organized inside a model’s representations. And throughout the substantive work, validation is done the old way first: hand annotation, multiple coders, reliability checks, before any model is allowed to scale.
The habit underneath: test the instrument before trusting the number. Prior stages: conceptualization · measurement.