# Reproduction scope and limits

The compact archive supports several bounded checks and reanalyses of files it contains. It does not provide a fresh end-to-end reconstruction from every original raw source, and the published summary should not be described as universally reproducible from raw inputs.

The archive README lists the supported commands and prerequisites. Its tiers include standard-library verification of delivered-file hashes, reintegration of the saved likelihood for the headline, native-family refits from supplied contrast inputs, and selected diagnostics refit from supplied analysis tables. Some diagnostics have a separate portable substate tier. These procedures verify or recompute only the supplied numerical inputs and specified model components.

The compact bundle deliberately omits some original workbooks, large raw datasets and caches, restricted materials, and copyrighted article PDFs. Some likelihood grids are already profiled from underlying cells, and some source-acquisition or full-project entry points therefore cannot be rerun from the compact bundle. The bundle manifest checks the files it contains; it does not make omitted sources available or certify every upstream transformation.

The 72,585-row evidence register is an inventory, not a count of independent studies. The 675 report checks are completion checks for this report, not a guarantee of full raw-source replay. For the current synthesis, the headline summary is based on 17 contexts, 21 contrasts, and 16 likelihood blocks. The shared-mean result is conditional on the specified evidence and model assumptions, not a universal individual-level risk ratio.