Initially normal is not normalised: registered re-analysis of the open KoSS septic-shock dataset — protocol, analysis code, outputs, supplementary tables and verification record

Analysis record for a transparent, preregistered re-analysis of the Korean Shock Society (KoSS) septic-shock registry dataset published as the S1 File of Ryoo et al., PLoS ONE 2019;14(6):e0217857. The study examines how lactate normalisation was defined (most patients coded as normalised within 6 h had a normal first lactate), conditioning on the first lactate, the role of worst-24-h SOFA, and the stability of data-driven lactate cut-points. Contents: registered protocol v1.0 (OSF 10.17605/OSF.IO/6HUVG); pre-registration pilot code and output; integrity checks; the registered analysis script and its output; post hoc analyses (labelled as such); code for Table 1 and Figures 1–3; supplementary tables S1–S5 and the STROBE/RECORD checklist; and the verification log comparing the registered run with an independent human re-implementation. A SHA-256 manifest lists every file. The dataset itself is not re-deposited: it is the S1 File of the source article (doi:10.1371/journal.pone.0217857.s001, CC BY). Scripts check its SHA-256 (12683d34aa68fd161dd95d1378a6d142411bc541bbb7df004d5fe007ddccbe9b). Code and drafts were written by an AI system (Claude, Anthropic) under the direction of R. Ama; the independent re-implementation was done by a human statistician. This deposit is an archive, not a registration.