{
 "id": "https://redfinchadvisory.com/lexicon/regression-eval",
 "slug": "regression-eval",
 "prefLabel": "Regression eval",
 "altLabel": [
  "Regression test"
 ],
 "hiddenLabel": [],
 "definition": "An eval designed to protect a working AI feature against unintended change.",
 "scopeNote": "Classic software: a regression test, a deterministic check that known-good output stays the same.\nAI: checking a working feature still performs as well after a change, including the vendor updating or retiring the model.\nRegression eval: Example: a working chatbot whose tools or documents are rewritten to be more efficient; you need to confirm it works as well as before. Because the output isn't deterministic, exact-match tests aren't enough for the content, so you use eval machinery: a frozen test set, scored, with a threshold you don't drop below, run before and after.",
 "tier": "core",
 "group": "Evaluation",
 "broader": [
  "https://redfinchadvisory.com/lexicon/eval"
 ],
 "narrower": [],
 "related": [
  {
   "id": "https://redfinchadvisory.com/lexicon/verified-test-set",
   "type": "related"
  },
  {
   "id": "https://redfinchadvisory.com/lexicon/reward-hacking",
   "type": "related"
  }
 ],
 "links": [
  {
   "id": "https://redfinchadvisory.com/lexicon/deterministic",
   "label": "alone is not enough for",
   "direction": "in"
  },
  {
   "id": "https://redfinchadvisory.com/lexicon/model-agnostic",
   "label": "relies on",
   "direction": "in"
  }
 ],
 "modified": "2026-10-21",
 "markdown": "https://redfinchadvisory.com/lexicon/regression-eval.md",
 "inScheme": {
  "id": "https://redfinchadvisory.com/lexicon",
  "version": "1.0",
  "currentAsOf": "2026-10-21"
 }
}