Definition
Plain language
A three-step loop where one AI checks each claim against its source, another flags weak ones, and a third fixes or drops them.
As stated in the literature
ScientistOne's verification cycle: a grounder validates inline evidence tags, a critic flags unsupported or overclaimed statements, and a resolver edits or drops claims until the tagged representation is fully backed.
Why it matters: It catches the fabrications and overclaims that creep into long AI-generated documents, where a single unchecked sentence can poison an otherwise good output.
For example, the grounder checks that a paper section's claim about a 30% improvement actually traces to its cited table, the critic flags that the table reports 20%, and the resolver edits the claim to match.
Heard on the show
“And there's a verification loop on top of it that the authors call Ground-Critic-Resolve.”Episode 089 — When AI-Written Papers Read Well But the Evidence Underneath Is Broken