Deterministic evaluation cases over the stored corpus. Outcomes are fixed examples, not live model calls.
Checks passed
3
Flagged for review
1
Evaluation cases
Each check evaluates one stored answer against a fixed expectation.
Evaluates
Who approves a new equipment request?
Expectation
Every citation passage appears verbatim in the cited section.
The passage matches section 3.2 of Equipment Request Procedure, version 1.2, character for character.
Evaluates
How many weeks of paid parental leave does the company provide?
Expectation
Questions outside the corpus receive a decline, never an invented answer.
The parental-leave question returns a documented decline with no suggested answer.
Evaluates
What does the company provide for remote work?
Expectation
Citations resolve to the newest stored version of the document.
Section 3.1 of Remote Work Handbook, version 2.0 is the current version; citation resolves to it.
Evaluates
What is the nightly hotel cap for international travel?
Expectation
Answers state every obligation the cited passage imposes.
The source requires written approval from People Operations for bookings above the nightly cap; the answer mentions receipts but not the approval step. Routed to human review.