Domain 5 of 6

Evaluating and Testing

What replaces the unit test when the same input can produce a different output. Writing an eval, building a set from real traffic rather than from imagination, using a model to grade a model and checking the grader, catching a regression before a release, the traces that make any of it possible, and judging an agent on its trajectory as well as its answer.

6
Concepts
~20%
Of the exam
18
Practice questions
Concepts in this domain
01Writing an evalThe test set that decides whether a change ships, built backwards from the failure a team is worried about. The four parts of a case, how a grader is chosen honestly, how many cases a set needs and why the held out set stays out of the development loop.02Building an eval setWhere eval cases come from, which is production traffic and never imagination. Reading traces and naming the failures, sampling for coverage with a quota per segment, the three sets a team keeps and how large a set has to be before a difference in the score means anything.03Using a model as a judgeGrading open ended output with a model against a written rubric. A weak judge prompt and the same prompt written properly, the three documented biases including position and verbosity, and the validation against human labels that decides whether the scores are worth anything.04Regression testing a promptRunning a changed prompt against the held out set, comparing it with the version serving production and failing the build on a quality drop. The comparison a pipeline prints, the gate rules that read it, and the segment a passing average hides.05Tracing and observabilityThe structured record of one interaction, covering every input, retrieved chunk, tool call and output. One trace written out in full, the attribute names two tools can agree on, and how replaying a recorded trace against a changed prompt turns yesterday's traffic into today's test case.06Evaluating an agentJudging the path an agent took alongside the answer it reached. A reference trajectory set against a real run, the rules for matching a run against that reference, the scorer that produces four separate numbers, and why an agent that succeeds once is not an agent that succeeds.
Try a question from this domain

A garden equipment retailer is writing its first eval for a refund assistant. One case carries the expected behaviour below, and the reviewer refuses it on the grounds that nobody can grade it. What does the strong version of the same case do?

  • AIt replaces the description with the exact answer the assistant should produce, so an equality check settles the case and no judgement is needed.
  • BIt raises the threshold from a majority of the repeats to all of them, so a loosely worded requirement cannot be satisfied by a partial answer.
  • CIt turns the one vague requirement into separate clauses, covering what the reply must contain, what it must not contain and what it must cite, each of which somebody can check by reading.
  • DIt moves the grading to a model against a rubric, since two reviewers will disagree about helpful and about accurate while a model applies one standard every time.
18 questions on this domain.

One per page, with a worked explanation.

Start the set