Skip to main content
  1. Public practice questions
  2. AB-620 Practice Exams | AI Agent Builder Associate
  3. AB-620 Case Study Set

Question 53 — AB-620 Case Study Set

Paideffort currently receives eight results from the regulatory API. Assume they have already been ranked from most relevant to least relevant.

ForAll(
    Topic.RegResponse.results,
    {
        Content: regulationText,
        ContentLocation: sourceUrl,
        Title: regulationTitle
    }
)

Which improvement best aligns the transformation with Paideffort's requirement to send a compact grounding set? (Select one answer)

  1. Restrict the mapped collection to the first three ranked records before passing the resulting table to Custom data.

  2. Map all eight records because Custom data automatically uses additional records when their relevance score is high.

  3. Reverse the collection before mapping so older records have a greater chance of being included.

  4. Remove ContentLocation and Title so all eight records can fit within the Custom data record limit.

Correct answers

  1. Restrict the mapped collection to the first three ranked records before passing the resulting table to Custom data.

Explanations

Correct Answer:

Restrict the mapped collection to the first three ranked records before passing the resulting table to Custom data.

Explanation:

Paideffort wants retrieval to produce a compact, intentional grounding set. Once the API has ranked the results, retaining only the top three makes the behavior clear and prevents the topic from carrying records that the generative answers node won't use.

Wrong Answers:

Map all eight records because Custom data automatically uses additional records when their relevance score is high.

Records beyond the first three aren't promoted because of their relevance score. Paideffort should perform selection before the table reaches the node.

Reverse the collection before mapping so older records have a greater chance of being included.

Reversing a correctly ranked collection would place weaker records into the positions that influence answer generation.

Remove ContentLocation and Title so all eight records can fit within the Custom data record limit.

The limit concerns the number of records used, not the number of columns. Removing citation metadata would also reduce traceability without increasing the number of grounding records.