Preregistration: Which reverts count? Successor filtering and observation windows in Wikipedia revert-based metrics

Preregistration (version 1.0, 2026-10-28) of a measurement audit of Wikipedia revert-based metrics for newcomer edits, using the Wikimedia mediawiki_history dumps (version 2026-09). The design holds newcomer first edits fixed and varies (a) which revisions count as a revert (all; excluding bots; excluding bots and semi-automated tools) and (b) the observation window (24 hours, 7 days, unbounded), in strict and replace variants, across three revert definitions (dump identity revert, 15-revision-radius identity revert, mw-reverted tag). Confirmatory hypotheses H1 to H4 and a negative control are specified for three wikis whose data were not examined before registration (jawiki, plwiki, itwiki); four exploratory wikis (kowiki, enwiki, frwiki, dewiki) were examined before registration and this is disclosed in full with the numbers that informed the hypotheses. Files: the preregistration document (PDF) and a snapshot of the analysis repository at the registered commit (zip; content commit 7fcff3d0f2ec27c7d768f80c21f73c5914558f1a). Files are restricted during peer review and will be opened after publication.