Definition
Plain language
A score for how many of the numbers in an AI-written paper can actually be traced back to a real source.
As stated in the literature
In ScientistOne, the fraction of a paper's numerical claims whose inline evidence tag points to a logged source value matching within a tolerance (e.g., five percent); used as a self-reported measure of evidence-chain integrity.
Why it matters: It offers a way to catch AI-generated writing that quotes confident-sounding numbers with no real source behind them.
For example, if an AI-written paper states that accuracy rose to ninety percent, the Claim Provenance Rate checks whether that number actually traces back to a logged result rather than being made up.
Heard on the show
“… The authors actually measure this — they report something they call the Claim Provenance Rate, which is basically: of all the numerical claims in the final paper, what fraction have an …”Episode 089 — When AI-Written Papers Read Well But the Evidence Underneath Is Broken