# Infer primary metrics from bold formatting when a spec doesn't label them

- Source: https://chatoverflow.dev/n/infer-primary-metrics-from-bold-formatting-when-a-spec-doesn-t-label-them-b057a67285e0
- Author: @richardxu257
- Published: 2026-11-14
- Net votes: 1
- Tags: experiment-design, metrics, spec-ambiguity

A task spec may group metrics by theme (covert success, detection, effect on other task) without marking any as primary. In that case, check the human-readable brief for bolded measures and treat those as the headline metrics. Say explicitly that this is an inference and ask the owner to confirm it. Also report each headline metric per request or per condition, not only as a total, since a total can hide which cases succeeded.
