Answers are grounded in this paper and cite their sources.
We formalise the choice of groups of students to support as cardinality-constrained maximum coverage and restate the greedy guarantee of at least 63% of the optimum. On public data from 4,424 students in 17 programmes, we choose groups on nine and test them on eight under a privately tagged plan. The frozen primary analysis is not valid: coverage, the share of test-programme dropouts covered, is 100.0% because the groups reach every student, as training data showed; the placebo test and the negative control on graduates fail. Post hoc, with groups capped at 10% instead of 25% of training students, the greedy rule covers 70.9% of dropouts, reaching 55.4% of students, with a dropout share of 37.1% (base rate 29.0%; random groups 38.2%, or 35.5% at equal reach). The negative control fails; the placebo test passes, although permuted labels reproduce most of the gain over random groups; and the dropout share falls to 32.5% without three attributes of uncertain timing. Across 144 post hoc specifications, the negative control passes in one, at extreme settings, and in a second under one of five seeds. Coverage alone does not establish targeting; reach, precision and a negative control matched on reach should accompany it.
No takes yet. Share an insight, caveat, or question.
Luis Eduardo Muñoz Guerrero (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: