AriaAICompany/phish-structural

1---2license: apache-2.03library_name: scikit-learn4pipeline_tag: text-classification5language:6  - en7  - fa8tags:9  - cybersecurity10  - phishing-detection11  - text-classification12  - multilingual13datasets:14  - AriaAICompany/phish-messages15---16 17# phish-structural18 19Small calibrated reviewers used by the Phish Space.20 21- `tfidf_lr.joblib` — `TfidfVectorizer(char_wb 3-5)` + logistic regression, sigmoid-calibrated22- `lightgbm.joblib` — LightGBM on structural features only, sigmoid-calibrated23 24Trained on the development campaigns of `AriaAICompany/phish-messages` (seed 3). Test campaigns are held out by campaign, sender group, and template.25 26## Intended use27 28Decision support for a human reviewing a **synthetic or already-shared** message. Scores are not a safe/unsafe certificate and are not a probability of account compromise.29 30## Out of scope31 32- Fetching or rendering live URLs33- Executing attachments34- Judging malice from a logo screenshot35- Treating copied SPF/DKIM/DMARC headers as a trusted receiver verdict36- Production mailbox enforcement37 38## Metrics39 40See `metrics.json` written by `scripts/build_artifacts.py` and `evaluations/evaluation.json` in the Space. Lab PR-AUC on synthetic fixtures is not operational accuracy.41