Label agreement misses key authorization and completeness issues
Label Agreement Does Not Measure Authorization
Artificial IntelligenceMachine Learning
Summary
Label agreement checks if two labels match, but it doesn’t show if the label was allowed, complete, or sensitive to changes in evidence. The authors built a pipeline using large language models to label neuroimaging data and discovered that while label agreement scores looked good, other important measures revealed many failures. For example, sometimes required data fields were missing or the label did not update when the input changed, and these problems were hidden by usual agreement metrics. The authors suggest measuring these aspects separately and involving humans when needed to catch errors.
What this means in practice
- •For data engineering teams: Augment data labeling pipelines with checks for label entitlement, completeness, and sensitivity to input changes to improve error detection beyond agreement scores.
- •For medical imaging analysts: Use independent metrics besides label agreement to ensure diagnostic labels in neuroimaging data are authorized and complete before clinical decisions.
Authors
Amir Sabbaghziarani, Bradley Thomas Baker, Theodore J. LaGrow, Sergey Plis
Abstract
Many groups now delegate label ontology and metadata harmonization to agentic LLM pipelines. We built one and audited it. Our aggregate scores looked healthy, but the pipeline kept failing in ways they did not show, so we set out to find what they hid. Label agreement asks whether a proposed label matches a reference. It does not ask whether the agent was entitled to propose it, whether the output was complete enough to act on, or whether the label moved when the evidence moved. We measured those three separately on COBRE and FBIRN, two schizophrenia and control neuroimaging cohorts from different consortia, and they come apart, from label agreement and from each other. Showing the agent an upstream proposal barely moves label agreement, 0.857 to 0.870, while agreement on the chosen action doubles, 0.409 to 0.830. Output that parses as JSON still drops a required field on 10% of one model's cases and 33% of the other's. And an agent that replays its first answer scores perfectly on original cases and zero once we change the evidence that decides them. Downstream, a row-order error that none of these metrics reports erases most of the diagnostic signal. So we measure these properties apart, pair each with a control, and gate commitment on the result, which makes failures visible and easy to route to a person. None of this prevents failure. Our reference labels are rule-derived, so agreement with them means consistency, not correctness. Code is available at https://github.com/amir-sbg/Label-Agreement-Does-Not-Measure-Authorization.