ICU oxygen-target reanalyses: analysis record (code, data and derived results underpinning the clinical reanalysis and the methodological evaluation)

Analysis record for two related manuscripts: a reanalysis of published aggregate data from randomised trials of oxygenation targets in critically ill adults, including seven sepsis datasets; and a methodological evaluation of how a fitted optimum from a continuous-exposure network meta-analysis should be reported, with an ADEMP simulation of 20,000 datasets plus 36,000 further datasets with off-centre generating truths. This is a retrospective analysis record, not a prospective registration. The clinical reconstruction was data-aware throughout and is exploratory. The simulation and the optimum-decomposition functionals were specified in a dated local plan before they were run, but after related results were known. The README states the chronology as a dated table. The record contains: the two dated analysis documents as written; the complete analysis code (estimators written out explicitly rather than called from a meta-analysis package); the derived arm-level inputs with the source table or figure identified for each number; and every derived numerical output the manuscripts quote, as machine-readable JSON, CSV and run logs, including a leave-one-trial-out sensitivity of the sepsis dose-response curve, a heterogeneity-prior sensitivity and the off-centre simulation, all added on 15 September 2026 after an internal review found two supplement figures that could not be traced to a results file. Posterior draw arrays are omitted for size and are regenerated deterministically by the code. A SHA-256 manifest covers every file. No individual participant data were used. The deposit contains no publisher PDFs, no extracted article text, no reproduced abstracts and no journal author instructions; primary reports must be obtained from their publishers. Licence: text and data CC BY 4.0; code (the code/ directory) MIT. AI tools assisted with source extraction, statistical implementation, cross-checking and drafting throughout; this is disclosed in the manuscripts. The named authors take responsibility for the content. The analyses have not yet had independent human statistical review, and one numerical correction has already been needed after internal review.