Skip to main content
Frontend Glossary

AI coding · Code quality

AI Evaluation

Also called: Evals, Eval set, Model evaluation, Quality regression

Use a fixed set of cases and scoring rules to see whether a prompt or model change got better or worse—not just two gut-feel tries.

In detail

Evals turn "feels smarter" into comparable numbers or checklists. Keep 20 real user questions, a few tricky UIs, or fields that must be extracted, and rerun them whenever you change prompts or models.

For Vibe Coding products (diagnose, generate, upgrade), evals matter: swapping models, rules, or temperature can silently regress. Scoring can be scripts (exact match, structured fields) or human rubrics.

Tiny or fake eval sets make you overfit the exam. Grow the set from real failures.

Developer info
Term ID
ai-eval
DOM selectors
No DOM cues. This concept isn't detected directly on a page.
Priority
1 · when several match at the same level, the higher priority wins
Version
v1 · updated Oct 5, 2026