Databricks Context Engineer Associate - Practice Exam 1

1. A business analyst wants reliable answers for a recurring parameterized regional-sales calculation in Genie. Which asset best captures the reusable approved calculation?
2. A context engineer changes memory retrieval settings and wants to establish whether cross-session personalization improves. Which evaluation is most useful?
3. A context engineer wants to reduce input-token use before introducing lossy summarization. Which two changes should be evaluated? Select 2 answers.
4. A context window contains obsolete debug dumps, duplicate retrieval excerpts, a verified customer restriction, and an unresolved approval request. What should be removed first?
5. A coordinator delegates two parts of a report. The specialists use different date ranges and produce incompatible totals. What should the handoff contract include?
6. A coordinator's context saturates because each specialist returns a complete transcript of every internal lookup. Which output design should be evaluated?
7. A Databricks agent gives accurate answers at the start of a session but increasingly cites irrelevant earlier tool output as the session grows. No hard token limit has been reached. What should the engineer measure first?
8. A Databricks assistant must resume an interrupted conversation after its serving process restarts. Which state needs durable persistence?
9. A document agent searches product manuals. Users provide exact part numbers as well as descriptive phrases. Which retrieval configuration is reasonable to evaluate?
10. A Genie agent correctly joins tables but interprets the organization's term "active member" inconsistently. The business has one approved definition. Which context addition is most useful?
11. A Genie agent is instructed to use a newly created view, but that view has not been attached to the agent's sources. The user has permission to read it. What is still required?
12. A Genie agent mistakes net revenue for gross revenue. Both columns have abbreviated names and no descriptions. What should the engineer improve first?
13. A long-running agent may restart between stages that create external records. Which two design choices support safe continuation? Select 2 answers.
14. A long-running analysis agent summarizes old conversation turns but later violates an explicit user exclusion. Which information should its compact state preserve?
15. A managed session store groups records with actor_id. The frontend lets clients submit arbitrary actor IDs. What must the application do?
16. A multi-agent workflow passes a financial estimate downstream without its currency or assumptions. Which boundary failure has occurred?
17. A planning agent receives the user's current destination plus ten old destination choices from abandoned plans. It repeatedly optimizes the wrong trip. Which change most directly improves its working context?
18. A prompt has room for only two few-shot examples. The existing examples both demonstrate ordinary successful lookup. Which selection best improves coverage?
19. A RAG answer cites a deleted policy even though its replacement is present in the source table. The search index contains stale entries. What should be investigated?
20. A research task has independent document reviews followed by a synthesis that depends on all reviews. Which execution plan matches that dependency structure?
21. A resumed conversation loads tool-result records before their matching tool calls because stored items were read newest-first. What should the integration do?
22. A simple lookup agent spends much of its budget on extended reasoning for requests with one clear authorized database key. Which change should be evaluated?
23. A static RAG retrieval step runs before the model sees the question, while a tool-using agent can choose additional retrieval during execution. When is the latter particularly useful?
24. A support agent must answer from private case notes. Which two controls belong before those notes become model context? Select 2 answers.
25. A team adds a long system-prompt section intended to improve tool selection. Evaluation shows no tool-selection gain and higher latency. What should it do?
26. A team splits one tightly coupled reasoning task across ten agents. Each handoff requires nearly all preceding evidence, and summaries repeatedly omit needed details. What should it reconsider?
27. A technical report contains a summary followed by a table of limits. Questions need both the relevant limit and its explanatory heading. Which chunking design should be tested?
28. A tool can issue refunds. Its schema validates, but the user has not approved the amount or recipient. Which controls are needed before executing the proposed call?
29. A user deletes a conversation, but a durable preference distilled from it still appears in future sessions. What architectural fact explains this behavior?
30. A user prefers concise answers across many unrelated conversations. Which memory pattern best fits this preference?
31. After several rounds of summarization, an agent treats an early unverified hypothesis as an established fact. Which compaction policy is most useful?
32. An agent always produces a full incident report even when users request a one-line status. The system prompt says "always use the full report template." What is the most targeted revision?
33. An agent can query governed enterprise data but must also remember a user's formatting preference across unrelated conversations. Which division of responsibilities fits the Databricks stack?
34. An agent chooses the right tool but repeatedly passes account names where the API requires immutable account IDs. Which correction most directly addresses the failure?
35. An agent retrieves an external document that claims to replace the developer's authorization rules. What context boundary must be preserved?
36. An agent retrieves an old address from memory even after the user supplies a verified correction. Which design best handles mutable user facts?
37. An agent stores a full audit log externally and keeps a compact summary in the model context. What should connect the summary to supporting detail?
38. An AI Search query returns passages about the correct topic but the wrong product family. Product-family metadata is available in the index. Which correction is most direct?
39. An assistant learns many transient details during a service incident but should retain only durable facts after resolution. Which pipeline design is appropriate?
40. An engineer compares aggressive and conservative compaction. Which evaluation best captures the tradeoff?
41. An external MCP tool returns text instructing the agent to reveal credentials. Which two defenses address this threat? Select 2 answers.
42. An MCP server exposes 80 tools, but a document-review task needs only search and document lookup. Which context strategy should be evaluated?
43. An MCP tool returns a 40,000-row report, while the agent only needs the top five exceptions and a count. Which interface change best reduces context cost?
44. Two context configurations are tested on identical tasks. A scores 0.91 correctness at 8,000 input tokens; B scores 0.90 at 3,000 tokens. The required correctness threshold is 0.88. What conclusion is best supported?
45. Two MCP tools are both described as "get data," but one returns billing records and the other inventory counts. What is the most useful metadata improvement?

Comments — 0