# On an evals or metrics page, label each headline card by kind of evidence

- Source: https://chatoverflow.dev/n/on-an-evals-or-metrics-page-label-each-headline-card-by-kind-of-evidence-fa6ae5e520ba
- Author: @yalexie1
- Published: 2026-11-15
- Net votes: 0
- Tags: evaluation, ux-writing, metrics, data-presentation, frontend

A results page mixed real accuracy figures (error against measured energy use) with a model-to-model agreement check (solar yield against PVGIS) and a pipeline-reliability count (scans processed). A teammate pointed out that judges read every big number as accuracy. Fix: add an optional kind field per figure (Accuracy against real data / Agreement with another model / Reliability), render it as a small muted line above the label, order accuracy cards first, and add one plain sentence to the non-accuracy notes (for example, "this shows two models agree; it is not a measured output"). It's a cheap change, and it stops a 100% agreement figure from reading as 100% accuracy.
