Why correct computation is still not enough: Auditing conclusions and evidence in web application firewall evaluation

Research data and code supporting “Why correct computation is still not enough: Auditing conclusions and evidence in web application firewall evaluation”, by Yuqun Lin. Version 2.0.1 contains 500 frozen historical tasks and reference answers, 5,153 candidate response records with 5,000 selected answers, 48 held-out direct/program comparisons, 48 changed-input checks, the shared 213-cell replay evidence, protocols, analysis code, eight tables, six figures, provenance and integrity manifests. All 3,883 historical response-integrity records are included as public credential-redacted derivatives: 3,391 P1/shared records plus 492 additional records explicitly authorized as integrity evidence. The additional records do not enter the 213-cell primary analysis or constitute new efficacy experiments. All expert-participated blind-review materials remain excluded. The unchanged main results are the 4,930-cell partition, 46/48 versus 48/48 paired scores, and 213/211/141 replay totals. The package passed independent offline extraction, recomputation, 13 original fault tests and five integrity-guard tests. Public and original capture hashes are kept distinct. Historical source-version and truncated-capture limitations are documented. Generated programs were not newly executed in Docker during preparation; archived stdout and Python/SQLite references were checked. Author-controlled project code is MIT; author-controlled data, documentation, tables and figures are CC BY 4.0 within the attached license scope. Third-party software, templates, embedded content and service/model outputs retain applicable original terms. The ZIP includes complete license texts, upstream notices and the redaction record. The earlier GitHub v2.0.0 remains preserved; this archive is the corrected v2.0.1.