Reading the file's credentials…
—
—
—
Made by a person, or by a prompt? We check pump.fun's top coins live and score each one out of 100.
or watch the live scanner ↓Reading the file's credentials…
—
—
—
Detection
Each layer is scored on its own. They combine as a weighted geometric mean, so one sloppy layer can't hide behind clean ones, and every verdict lists what it couldn't check.
Many generators sign their images. We read the file's metadata only: OpenAI, Adobe and Microsoft Content Credentials, the IPTC "made by an AI model" tag, prompts left in Stable Diffusion files. A hit is a lookup, not a guess, and it names the generator.
"Revolutionizing the digital landscape." Every stock phrase, em dash and rocket stack in the description gets highlighted, counted in code and weighed against the length of the text.
AI site-builder fingerprints, starter-template leftovers, lorem ipsum, buttons that go nowhere, copy-paste roadmaps, and a site that never mentions its own CA.
The same name, ticker or image reused by other coins trading right now. Five $PERCYs in an hour is a pattern.
A System One model reads the copy and the site and answers in calibrated probabilities: how template is this, does the site promise things it never shows, is the text trying to steer the scanner. Its confidence moves the needle.
Slop signals, not proof. A missing receipt proves nothing, and a human can write like a template. Scores are our read of public data, and none of it is financial advice.