duncan-clark/PPDisentangle: v0.2.0 — current major-revision analysis and paper reproduction
Abstract
This release brings the public software archive up to the current major-revision analysis and provides the matching Oklahoma paper-reproduction workflow. The companion results v2 contains Oklahoma job 8804859 plus its ATE backfill: 818 post-treatment events, a 250-day temporal cutoff, 512 attempted bootstrap replicates, and the 100-day observed-AOI-versus-none contrast. Expected saved counts are -51.508 for the naive fit and +175.200 for SEM. The main simulation and 67-scenario robustness results remain frozen. Explicitly select observed-versus-none for Oklahoma figures and tables, rejecting unavailable contrasts. Regenerate count and partition-sensitivity tables directly from saved fits, with full-precision CSV summaries. Freeze county geometry, resolve archived robustness paths locally, and support paths containing spaces. Regenerate publication assets during archive staging; validate canonical results and include a source manifest and per-file checksums. Document fitting settings, bootstrap recentering, stationary-rate reporting and the different simulation/application intervention conventions. Reproduction commands and limitations are in inst/zenodo/README.md. The archived univariate all-treated simulation omits control-history descendants, and its reference truth follows that same convention; the Oklahoma bivariate implementation retains the supplied pretreatment history. This release documents the frozen calculation and does not change the scientific model or rerun the studies. Validation: full figure/table reproduction from a staged archive; checks of canonical Oklahoma counts, bootstrap retained counts and all 67 scenarios; focused contrast-selection, bivariate ATE dispatch/KDE-cache and truncated-Omori tests. Source was tested with the existing compiled library; a fresh local native build was unavailable because this machine's Xcode licence has not been accepted.
