Evaluation
Also called: eval · evals
By HubAI editorial team · Source checked: 24 October 2026What does it mean?
A deliberate check of how well a system performs against defined tasks and criteria. Good evaluation covers failure cases and the complete workflow, not only polished example answers.
Picture this
A support team tests 50 representative enquiries, records incorrect policy claims and measures reviewer corrections before deciding whether to expand a trial.
Where you’ll come across it
Choosing tools, testing prompt changes and monitoring a deployed system. Keep the input set, scoring rules and model version so comparisons remain meaningful.
What it doesn’t mean
An automated judge can also be wrong. Use human review where consequences warrant it and do not turn one small score into a universal quality claim.
Keep a tiny scorecard
A short exercise · use fictional or non-sensitive material.
- Choose five representative fictional tasks and define what counts as correct before running them.
- Record success, correction needed and time to a usable result for each attempt.
- Repeat after one change, keeping the task set fixed. Treat this as an exploratory check, not a reliable benchmark.
What failed, and would that failure matter to the person using the result?
Did that make it clearer?
Likes are a private on-page choice, not a public rating. For a correction or complaint, send the pre-filled form to our editorial inbox. Nothing is sent until you submit it.Check the source
Written and explained by HubAI. The reference below lets you go deeper; it is not an endorsement of a particular product.
Technical reference: www.nist.gov ↗