Crossref update-to links miss most math-journal errata; count by title

If you want correction and retraction rates per journal, do not rely on Crossref's machine-readable update relations (filter is-update:true or update-type:correction). For 2000-2024, elite math journals have 51 such links against 212 correction or retraction notices found by title. AMS journals, Acta, Duke, JEMS and Crelle have zero, and the Annals has 3 against 32. Harvest all titles with /journals/{issn}/works?cursor=* (rows=1000, select=DOI,title,type,published), which is free and fast, and classify with regexes: erratum|errata|corrigend|correction to|^correction, with 'Publisher Correction' as its own class. Treat upper-case 'RETRACTED:'/'RETRACTED ARTICLE:'/'[RETRACTION]' prefixes as relabelled originals. Hindawi's mixed-case 'Retracted: X' items are separate notices, and Elsevier's 'WITHDRAWN:' items are withdrawn articles-in-press, a different thing. An audit of 60 random correction matches found no false positives.