Reproduction package: What a Median Cannot Support. Reporting Rules and Corrective Instruments

<p dir="ltr">Data and code behind every computed quantity in the article "What a Median Cannot</p><p dir="ltr">Support: Reporting Rules and Corrective Instruments".</p><p><br></p><p dir="ltr">The archive contains a single Python script and fifteen CSV files: FDA file-level</p><p dir="ltr">extracts for the 510(k), De Novo and premarket approval pathways; the receipt-cohort</p><p dir="ltr">records with the reconstructed applicant crosswalk; the parsed Federal Register</p><p dir="ltr">notices determining regulatory review periods; the patent term extensions granted</p><p dir="ltr">under 35 U.S.C. 156; the notified-body survey series commissioned by the European</p><p dir="ltr">Commission; the duration bracket definitions; Eurostat size-class figures; and a</p><p dir="ltr">table of every externally sourced parameter with its source and page.</p><p><br></p><p dir="ltr">Running the script reproduces the clearance-time distributions for the three US</p><p dir="ltr">pathways and for the European survey, the distribution-free identification bounds on</p><p dir="ltr">tail measures, the mixture-variance decomposition by firm size, the applicant-</p><p dir="ltr">experience regressions at the mean and at four quantiles, and the welfare arithmetic</p><p dir="ltr">of the article: the cost of the reporting constraint across thresholds, the share of</p><p dir="ltr">the upper tail held by small and medium-sized manufacturers, the error-cost frontier</p><p dir="ltr">of the tail-indexed instrument against the median financing interval, and the</p><p dir="ltr">counterfactual in which the US file-level record is collapsed to the European</p><p dir="ltr">reporting format of Implementing Regulation (EU) 2026/977 and only the identified</p><p dir="ltr">sets remain.</p><p><br></p><p dir="ltr">Both the archive and its contents carry SHA-256 manifests, so a reader can establish</p><p dir="ltr">before and after extraction that the copy being run is the one the reported</p><p dir="ltr">quantities come from.

All inputs are public. The applicant key used in the receipt-</p><p dir="ltr">cohort estimates is a reconstruction from the public applicant strings and not an FDA</p><p dir="ltr">field; it reproduces the identifier, the prior count and the submission order of all</p><p dir="ltr">15,702 analysis cases, but carries no claim about corporate control or renaming.</p>