evals/targeted-improvement-heldout-v2/graders/validation.md

---
type: llm
---
PASS if the answer includes both required elements below, in any wording or method. Parenthetical examples are illustrative, not required.
(1) A concrete check with customers who hold two similar rentals (for example, an uncoached task to extend a named unit, a moderated session, or a test with real accounts).
(2) An observable result from that check that would show the change works (for example, customers extend the intended unit without help, or can say which rental changed and its new date). Post-launch aggregates such as support contacts about which rental changed may supplement this result but cannot replace it.
FAIL if validation is missing, names no observable result from a check with such customers, or relies only on support contacts or other post-launch aggregates.