What multi-model consensus catches that single models miss

A single model doesn't fail loudly. Its failures are fluent, specific, and wrong. It will invent a plausible statistic, misattribute a quote, or date an event two years off, and it packages the error in the same confident prose as its correct answers. If one model is your fact-checker, its blind spots are your blind spots, and nothing in the system can see them.

The engineering answer is old, which is why we trust it: independent redundancy. Get several estimates from systems that fail differently, and compare.

Every claim we check goes to a panel of frontier models from different providers, with different training data, different cutoffs, and different failure habits. Each votes independently, without seeing the others' answers. When one model asserts something the rest refute, that's not noise to average away. It's the most useful signal in the pipeline. A confabulation is rarely shared. The models that never saw the invented statistic outvote the one that "remembers" it, and the disagreement itself lowers the consensus confidence, visibly, in the report, with the dissenting vote on the record.

Knowledge cutoffs work the same way. A model that stopped learning in 2024 will describe a 2025 state of the world with the same confidence. On a panel, its vote collides with models that know better. On live-research checks, it collides with models that just looked it up and have to cite what they found.

Here's the part most "ensemble" pitches skip. Models trained on overlapping internet data can all inherit the same popular misconception. On those claims the panel is unanimous, confident, and wrong together. Agreement is evidence, not proof.

So unanimity gets the most suspicious treatment in our pipeline, not the least. Before we confirm a unanimous, confident verdict, an adversarial challenger tries to refute it. The challenger is a cheap, independent model that never sits on any panel, and its prompt asks for one thing: a refutation, not a re-derivation or a summary. If the challenger finds a flaw, the case escalates to a full judgment with that dissent on the record. If it can't, you get a verdict that survived an attempt to break it, which is a stronger claim than "four models agreed."

When the panel splits, or the challenger succeeds, we don't take a majority vote. Counting heads treats a well-evidenced minority the same as an outvoted hallucination. An independent judge model reads every vote with its full rationale and cited evidence, then writes a verdict on a 7-point scale. On that scale, "misleading" is a first-class outcome rather than a rounding error toward true or false. Misleading means literally true but missing the context that changes the takeaway. Three shallow "seems true" votes can lose, and should lose, to one vote that traces the claim back to a primary source.

The confidence score you see at the end is computed, not self-reported. It starts from the judge's confidence and is dampened by how much the panel agreed, how authoritative the cited sources are, and whether those citations support the claim. The full formula is on our methodology page.

Here's a claim from our public sample report. The article asserts that every French Fifth Republic president except de Gaulle and Macron has a major corruption scandal to their name. The panel split three ways. One model ("sonnet") called it mostly true, because the pattern is broadly defensible. Two ("gemini" and "gptmini") called it false, working president by president. One ("deepseek") said unverifiable, because the source article only evidenced two of the six presidents it indicted.

Each model, alone, would have shipped a different answer, and each answer is reasoned. The judge's synthesis resolves it. The claim holds for Chirac and Sarkozy, is weaker for Giscard and Mitterrand, and fails for Hollande and Pompidou. A single exception falsifies an "every president" claim. Verdict: mostly false, at low confidence, with 40% panel agreement on the record. The judge notes that the overreach fails, not the underlying pattern. No single vote contained that nuance. The mostly-true model saw the pattern, the false models saw the exceptions, and the verdict keeps both.

None of this asks for trust. Every report keeps its audit trail: each model's vote and reasoning, the challenger's attempt, the judge's synthesis, and the citations behind it. Run a document with a claim you already know is wrong in a small way and watch where the panel splits. The disagreement is the product.