Glossary

Eval suite

A set of graded test cases that score an agent's quality and gate releases against regressions.

category ▸ Operations

audit/verify
#101hash ✓#102hash ✓#103hash ✓#104hash ✓#105hash ✓
head 0x9f3a…c1ok: true
verifyChain ▸ chained SHA-256 · signed receipts
What it means

Eval suite, in plain language.

An eval suite is to an AI agent what a test suite is to ordinary software: a curated set of cases with known-good expectations, run automatically to measure quality and catch regressions before they ship. For agents, evals typically grade dimensions like faithfulness to source, citation coverage, and safety, not just exact-match correctness.

Evals are how AI quality stops being a matter of opinion. A green demo proves nothing about the next thousand runs; a passing eval suite, run as a release gate, gives you a defensible, repeatable basis for saying an agent is ready.

In Cortex

How Cortex implements it.

This term isn't abstract here — it maps to a real capability in the runtime. Here is exactly how Cortex enforces or relates to it.

Operations

Cortex runs gold eval suites as release gates and scores quality on every run — faithfulness, citation coverage, and safety — with live runs returning a quality { overall } reading that traces back to the run that earned it. A simulation lab lets you A/B prompts and models on real traffic before promoting a change.

Eval pass rate feeds the reliability score, so failing evals don't just show on a chart — they pull the score down and can hold a release at the publish gate.

See Eval suite enforced, not just defined.

Book a walkthrough and watch the controls in this glossary return real verdicts, seal real evidence, and trace every fact back to its source.