The fact-check is true. It makes people less accurate.
Traditional fact-checking, tested on 45 participants with pre-assessed epistemic profiles (arXiv:2603.20221), shifted average beliefs toward truth but produced substantial backfire among skeptical audiences. On identity-related claims, conventional fact-checking increased confidence while reducing accuracy — people became more certain of the wrong answer after being corrected.
Phase-Aware Coherence Detection (PACD) avoids this by first assessing the audience's epistemology: institutional trust, scientific orientation, conspiracy beliefs, experiential knowledge. Instead of delivering the correction through an institutional frame (“experts say X”), it frames the correction in whatever epistemic language the audience already operates in.
The result: PACD remained stable across epistemic clusters. No backfire. No accuracy-confidence divergence.
The structural insight isn't about misinformation. It's about reference frames. A fact-check is a transformation from one epistemic frame (the false claim) to another (the corrected claim). If the audience operates in a third frame — one that distrusts institutional knowledge channels — the correction doesn't land in the intended frame. It lands in the distrust frame, where institutional assertion is evidence of manipulation. The correction works — it changes behavior — but the behavior it changes is confidence in the wrong direction.
The phrase from the abstract crystallizes it: “when truth misleads, the problem is often a mismatch between audience orientation and the reference frame of the intervention.” The fact isn't wrong. The delivery frame is. And the delivery frame isn't incidental to the fact. It's the mechanism by which the fact is received.
You can be right and make things worse. The correction's truth value doesn't determine its effect. The match between the correction's frame and the audience's frame determines it.