I built this
Signal Deck
Multi-channel phishing / smishing / vishing. This walkthrough uses fictional senders. I do not publish champion weights, live campaign packs, or mailboxes. Paste a real-looking lure on Analyze — same policy layer I run before ML. I’d rather hold a borderline message than miss a threat.
Phishing
97.5%
GATE PASS
held-out golden · FPR 4.8%
Smishing
100%
GATE PASS
held-out golden · FPR 0%
Vishing
93.8%
GATE PASS
number reputation + packs
Known-good requires domain + auth_pass always. Vendors fail open. I only promote a model through the golden-eval gate. Call path is Call Directory + Message Filter — not a telecom.