Zhijun He / Data and code
Data and code
Replication material for the empirical work. Each package is archived with a DOI, licensed for reuse, and documented to the point where a reader can check a reported number without taking my word for it.
Who Fails the Filter? Replication package
For "Who Fails the Filter? Size, Verification and the Incidence of an Unchecked Minimum Capital Requirement," under review at the European Journal of Law and Economics.Processing code, a codebook, and a SHA-256 manifest for a study of Brazil's minimum capital requirement for the single-owner limited-liability company (EIRELI), computed on the Receita Federal's open CNPJ register.
The package holds 56 analysis scripts, 54 run in batch and 2 by hand; a codebook documenting every variable read and constructed and the coded values the analysis depends on; a manifest of all 308 source archives, of which the 58 archives from the July 2026 and May 2023 extractions carry the reported results; the derived tables behind each figure; and the captured output of every script, so that any reported number can be checked without downloading the 10 GB of source data.
The source data are not redistributed. They are the Receita Federal's public extractions, held on the official share on a rolling thirty-nine-month window. The manifest hashes the archives as published, so it identifies a byte-identical copy and will not match a mirror that has re-compressed the same contents.
The codebook also lists six questions this register cannot answer. A source that is used honestly has to be described by what it excludes as well as by what it contains, and those six limits bound what the paper is entitled to claim.
Two further packages are in preparation: the machine-learning corpus of UN Security Council resolutions built for the Office of the Special Representative on Sexual Violence in Conflict, and the custody timeline of Rwanda's gacaca archive assembled for the Genocide Studies Program. Both will be posted here when the underlying work is ready to be checked.
ORCID 0009-0001-0062-4062. · Last updated August 2026.

