Repository navigation
Average precision under random ranking - v1.0.0
Average precision under random ranking
The paper and reproducible supplement by Shantanu Singh and Anne E. Carpenter give the exact mean and variance of full-list AP under uniform random ranking, with explicit boundary cases, two asymptotic regimes, and Lean proofs.
Read and reproduce
paper.pdf: the seven-page manuscript, including appendices and references.supplement.zip: complete source, pinned dependencies, Python calculations, Lean proofs, and reproduction instructions.arxiv-source.zip: TeX source with ancillary material arranged for arXiv.CITATION.cffandCITATION.bib: citation metadata for this version.SHA256SUMS.txt: checksums of the downloadable files.
Extract supplement.zip and run bash reproduce.sh to regenerate the paper, figure, tables, verification record, and source archives.
The README also gives Python-only commands and a minimal exact-moments example.
Citation
Singh, S., and Carpenter, A. E. (2026).
Average precision under random ranking: baseline, variability, and formal verification (Version 1.0.0).
Broad Institute of MIT and Harvard.
Versioned release.
Verification
Exact enumeration covers 65 small configurations, including zero positives.
The four second-moment coefficient sums and equivalence to the published variance pass independent checks.
All three project Lean modules build, and the eight audited results use only the recorded foundational axioms.
The manuscript compiles without warnings and has been visually inspected.
The supplement includes the full reader review and verification record.