Reproducibility deposit for: When is a DEA peer composition unique? An exact applicability criterion and a census of its incidence

Résumé fourni par la source

Reproducibility deposit for the manuscript "When is a DEA peer composition unique? An exact applicability criterion and a census of its incidence" (Tengfei Shao and Xu Wang, 2026), prepared for submission to Operations Research Perspectives. The paper asks when the intensity vector a DEA solver returns is the only one reproducing a fixed input-output target, gives the exact dimension of the complete set of such certificates as a count on the target's minimal face, and censuses how often that criterion binds in open data. Contents. The registered structural census and every artefact its pin table binds by content hash; the pre-registration protocol and the gate records for the count-variable search; the post-registration incidence study, including the eleven-population extension, the two counted-coordinate screens, the relative-resolution sweep and its mechanism decomposition, the frontier-size diagnostic, a sensitivity check on the target rule's stage slack, and a demonstration of the external-attribute audit's mathematical core on real non-singleton fibres; the frozen Eurostat GISCO NUTS vintage lists that the regional screen reads offline; the four figures of the paper with the scripts that draw them; and run_all.py. Reproduction. run_all.py verifies all 78 listed files against MANIFEST.json, verifies each source archive by SHA-256, re-executes the analysis and compares what it produces against the deposited results field by field rather than against values it recomputes. The full run reproduces the registered census and every post-registration result from a clean extract. The two figure scripts rebuild the four figures byte for byte. Source data. No source archive is redistributed. The FAOSTAT and Eurostat archives are identified in SOURCES.md by retrieval URL, licence, byte count and SHA-256. One derived third-party file is redistributed, the frozen NUTS vintage list, so that the regional screen can run offline; it retains the European Commission's reuse terms. Arithmetic. Every geometric decision and every reported count is exact, over fractions.Fraction and integers, so no certification depends on floating-point behaviour or on the platform. Quantities derived from those counts for reporting, such as incidence ratios and a log-log slope, are floating point and enter no classification. Scope. The registered external-attribute stage was not authorised by the study's structural gate and was not executed. The deposit therefore contains that audit's specification and a demonstration of its mathematical core on discretised fibres, not an applied attribute result. Licence. MIT for code; CC BY 4.0 for result files, records and documentation; the redistributed NUTS vintage list retains the European Commission's reuse terms. See LICENSE.md. Requires Python 3.11 or later. The results path uses the Python standard library only; the two figure scripts require matplotlib 3.10.9.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.