From Public Predictions to Auditable Judgments: Operationalization and Evidence Obstacles in 120 Cybersecurity Conference Predictions, 2005–2018

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

Version 1.6 (September 23, 2026). Substantial methodological correction of version 1.1: this work is now presented as an exploratory case study of the auditability of 120 included predictions, not a measure of speakers' or conferences' predictive skill. The reframed question was formulated after the original results were examined. The frozen extraction and outcome records were not altered. The original coding, attributed in full to Codex (AI), classified 54 predictions as fulfilled, 35 as unfulfilled, and 31 as indeterminate. The 60.7% figure (54/89 resolvable cases) describes that coding and matches the trivial rule of calling every resolvable case 'fulfilled'; it does not establish predictive ability. Of the 120 units, 115 required operationalization. Subsequent review by Codex, Claude, DeepSeek, and Gemini was used to locate disagreements. The author adjudicated 48 focused decisions after seeing their votes; six of 18 outcomes selected for AI disagreement differed from the original coding. This is neither a population error rate nor independent human validation. The original protocol was internal, with no prior external preregistration. The corrected manuscript, preserved original records, review matrices, focused author adjudication, code, and a SHA-256 manifest are included. The PDF and original research contributions are offered under CC BY 4.0; original software is MIT-licensed. Verbatim third-party quotations in the package are not sublicensed. This is an author preprint, not a peer-reviewed paper. See the change note and RIGHTS.md in the ZIP.