Answers are grounded in this paper and cite their sources.
An accurately calculated progression score need not enforce the intended competency requirement. We analyze an implemented score combining test accuracy, reported practical completion, and historical competency estimates, and compare progression decisions under mean and every-skill readiness. Algebraic inspection identifies overlapping evidence and compensatory behavior; a set-inclusion cost decomposition characterizes an added gate at a common threshold. Synthetic offline replay comprises 14,400 initial learners and 43,200 additional learners from three independently generated seed datasets. At fixed threshold 0.75 and error costs 3:1, the gate increases decision cost under both readiness criteria. Separately calibrated score-and-threshold policies instead yield pooled D-minus-B differences of +0.02614 for mean readiness and −0.01184 for every-skill readiness in the extension. Post-review analyses add a two-threshold rule, learned logistic weights, length-specific calibration, and uncertainty from threshold reselection. The logistic comparator has the lowest baseline point cost among the six evaluated policies, while D improves only modestly over the minimum-skill rule. These results characterize conditional computational trade-offs under stipulated readiness labels and fixed exposures; they do not establish educational effectiveness or universal policy superiority.
No takes yet. Share an insight, caveat, or question.
Sabitov et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: