PhishNChips Decision Model Comparison
Compare how often decision models miss phishing emails or flag legitimate mail, and how those errors change with the share of phishing in an inbox.
Explore the original evidence
Explore the original evaluation, its methods, and reported findings.
Open evalCoverage
EmDash Cyber tested 12 models on 3,011 unique emails after deduplicating four overlapping files from PhishNChips v5.2 (revision 89afcc3).
Methodology
On October 18, 2026, each model classified the same messages from their text and URL strings. EmDash used its own binary classification question. The dashboard estimates precision at different phishing rates from the observed false positives and misses. The test set contains about twice as many phishing messages as legitimate ones. The run used the dataset’s original labels and left links unopened.