Replication package for "Perfect prediction makes a poor prescription for curriculum retention"
Replication package for the manuscript "Perfect prediction makes a poor prescription for curriculum retention", version 2.1, prepared for the Journal of the Operational Research Society. No analysis has changed since version 1.0. The analysis scripts, the checking scripts, the tools and every derived result table are byte-identical across all three versions, so the version 1.0 audit trail applies unchanged. Version 2.1 carries the manuscript as revised after an internal review round: the abstract now states the theoretical result, a jackknife over institutions was added, the declaration of generative AI use separates language editing from code drafting, Section 7.3 notes how an institution could elicit the objective weight and the floor, and a paragraph duplicated in Appendix J was removed. Contents: 82 analysis scripts, 7 checking scripts, 5 tools, 59 derived result tables, 9 figures at 600 dpi in PNG and PDF, the manuscript with its Online Resource 1, title page and statement of contribution, and the 33 scripts that perform and guard the shortening. 246 files in all. Verification: 46 reported numbers are audited automatically against their source tables and all 46 match, with none mismatching and ten patterns skipped as needing re-anchoring to the rewritten text; 44 load-bearing claims are pinned against loss from the main text. Both audits are reproducible from the archive. The primary dataset is a third-party release and is not redistributed here; the README gives its citation and access details, together with a warning about two institutions whose enrolment column is empty at source. Random seeds are fixed and no reported value is transcribed by hand.