This is an aid, not a guarantee. It explains which written rules matched. It cannot prove a message is safe, and it cannot tell you what VA, a bank, or a court will do. If you already sent money or personal information, contact the organization using a number you look up yourself, then consider ReportFraud.ftc.gov and IC3.gov.

Screenshot reading uses local Tesseract files in this folder. It needs a normal web address (a local server or GitHub Pages), not a double-clicked file, because browsers block the OCR worker from file://. No screenshot is loaded yet.

Results appear here after you analyze a message. A short note is not enough to judge, and a low result is not a clearance.

How the rules were chosen

The rules are original to this toolkit. They look for tactics described in public VA, VA OIG, FTC, and FBI IC3 material: benefits impersonation, pension poaching, claim fees, gift cards and wires, romance or emergency money requests, lookalike links, prize lures, and remote-access support. Each hit shows the phrase that matched and a link to a public page. The page does not download those sites when you analyze a message. It does not inspect a face, a voice, or a video, and it does not detect deepfakes.

A second, optional hint comes from a small naive Bayes model trained on the SMS Spam Collection v.1 (Almeida, Gómez Hidalgo, and Yamakami, 2011). The raw text messages are not included. At a 90% cutoff, one 20% holdout of that collection had about 88% precision and 99% recall. That is promotional text-message spam from around 2011, so the hint is weak on veteran-benefits fraud. The rule score does not change because of the model.

A second reading uses a point rubric stored with this page. It does not change the written-rule level. OCR uses Tesseract.js stored in vendor/tesseract/ and runs on this device. An optional smaller Gemma model can be downloaded later on this page. It is not required. Use of that model is subject to the Gemma Terms of Use and the Gemma Prohibited Use Policy, which the download panel links before any file is fetched.