Misinformation Example and Score Comparison.
This was an X post the was eventually debunked as satire, but got reposted a few times.
This is my misinformation example benchmark post. See others on this site for other examples. I'm currently using GPT 5.6 Sol, but folks are welcome to try other small or large LLMs to understand how much error the model can introduce.
Sanity Media 2.4 Score (GPT 5.6 Sol):
Nick Shirley / Marjorie Taylor Greene bribery post
Mainstream Alignment: 10
Accuracy: 10
Clickbait: 90
Toxicity: 60
Synthetic Indicators: 40
Summary:
Mainstream Alignment is very low because the central allegation is outside mainstream and wire-level reporting. Accuracy is very low because the $14 million bribery claim has been traced to satire and is unsupported by corroborating evidence; the post also names “Joseph Barron” as the FBI Atlanta Special Agent in Charge, while the FBI identifies Marlo Graham in that position. Clickbait is extreme because the post combines an all-caps “HAS PROOF” headline, a large alleged payoff, political betrayal, an FBI investigation, and implied impending consequences. Toxicity is elevated because it accuses Greene of betraying her country and president, though it contains no threats or dehumanization. Synthetic Indicators are comparatively modest because the photographs and social-media layout appear conventional; the prose is formulaic, but that overlaps substantially with ordinary human meme and engagement-content production.
Signal pattern:
10 / 10 / 90 / 60 / 40
Institutional Omission:
Not triggered.
