Re-score by corpus
Score the same questions against different slices of the past-examination corpus. A cohort can only have been exposed to the papers that preceded its own sitting, so validating the rubric against student performance needs the difficulty as it stood at that sitting. The model is not called: Table 3 is counted in code and the eleven Table 2 scores are reused from the run.
Run
Any finished run whose questions were calibrated.
No runs yet
Generate some questions first.
Variants
One scoring per cut-off, plus the corpus-free and whole-corpus ends.
This run has no calibrated questions.