Experimental Demo — Synthetic Data
One risk score. A separate ML signal.
Explore how an experimental phishing warning can appear alongside the existing risk classification. Every example is synthetic. This page does not scan a URL or make a payment decision.
Use the arrow keys to explore. The warning never changes the supplied risk score or classification.
Existing risk assessment
/ 100
—Existing boundaries: SAFE 0–30 · SUSPICIOUS 31–60 · UNSAFE 61–100. This synthetic score is supplied directly; no hybrid scoring formula is recalculated.
Experimental ML signal
—Synthetic candidate phishing score: —
Advisory only. This signal never automatically changes the risk classification to UNSAFE or blocks an action.
Try Experimental ML Prediction
Experimental ML Only — Not a Full QR Shield Scan
Analyze URL text with the isolated local candidate. The destination is never visited. No production risk score or classification is calculated.
Experimental ML Only — Not a Full QR Shield Scan
Candidate phishing score:
Advisory only. A warning does not confirm phishing; no warning does not guarantee safety. Existing production risk classification has not been evaluated.
How to read this demo
A warning does not confirm phishing. No warning does not guarantee safety. A SAFE risk classification is also not a guarantee of safety.
The exact experimental warning threshold is 0.40144790130467084, inclusive. It applies to the candidate phishing score, not to the 0–100 hybrid risk score. Candidate scores are not independently calibrated probabilities of harm.
At this validation-selected threshold, measured phishing recall was about 90%, with a 19.40% false-positive rate: approximately one in five benign validation examples received a warning. About one in ten phishing examples was missed. These are validation results, not guarantees for future scans.
Dataset version and upstream licensing remain unverified. The optional prediction form connects only to the isolated local experimental API. The production scanner remains separate.