Single words are noisy. Rules look for constructions that resemble an actual script.
Open rulebook
Read the entire taxonomy
The complete rule library behind PrivateCue: nine scam archetypes, five behaviour areas and every weighted indicator, with its reasoning.
11 Script families5 Behaviour areas79 Indicators
Open rulebook
Read the entire taxonomy
This is the actual product; everything else on the page is a wrapper around it. How the score is built.
Changelog
- 2026.09Baseline release. Nine script families, five behaviour areas, 65 weighted indicators and 6 protective patterns — assembled from published fraud guidance (FBI IC3, FTC, City of London Police, AARP) plus survivor reports, and calibrated against samples of healthy conversation.
New versions appear here the day they ship. The library is edited by hand, and every change is tested against healthy conversation before it reaches a report.
Every rule has to earn its place.
11 script families, 5 behaviour areas and protective patterns that can lower a score. Each indicator carries a weight from 1 to 3 and a written reason for existing.
01Published sourceOfficial guidance or survivor report
→
02Repeated phraseA pattern, not a single word
→
03Written reasonWhy the behaviour matters
→
04Healthy-text testChecked against ordinary conversation
If the reason cannot be written plainly, the rule does not belong.
Ordinary relationship talk and plausible behaviour actively reduce the score.
No model is trained on conversations. The library is edited from published evidence.
View the principal source set
FBI IC3 Internet Crime Report 2025 · FTC Consumer Sentinel and the April 2026 romance-fraud release · City of London Police (May 2026) · AARP Fraud Watch Network (February 2026), together with survivor reports.
A library of known patterns—not certainty.
×Not a verdictIt measures resemblance to scripts, not guilt.
×Not a probabilityNo invented precision claim or certainty percentage.
×Not exhaustiveA low score is not evidence that somebody is safe.
ENWritten for EnglishTranslated text can understate risk; link analysis still works.
Analyse a conversation against it
The report names the rule that fired and shows you the phrase from your own conversation that triggered it.