Scores a host name and shows exactly why. The model, the feature extraction and the SHAP explanation all run in this tab — nothing is sent anywhere.
A support classifier with a measured error rate, not an authority. It reads the
host name only: it cannot resolve shorteners like bit.ly, cannot see page
content, and will miss phishing hosted on compromised legitimate domains — 321 hosts in
the evaluation data are exactly that case. At its 1 % false-positive operating point
it catches roughly — of phishing, so it misses the rest. Do not
treat a “legitimate” verdict as safety.