Skip to content

Context Management & Reliability Questions

Practice questions for Context Management & Reliability topic in Claude Certified Architect – Foundations. 30 questions covering this domain.

30 questions5 easy16 medium9 hard
Q1
easy

What is the "lost-in-the-middle" phenomenon in large language models?

Q2
medium

A single Claude agent accumulates extensive context over a long research task, and response quality degrades as the context grows. Which pattern best ...

Q3
medium

A Claude-based system processes financial reports in nightly batches. During a server restart at 3 AM, processing stops with only partial completion. ...

Q4
medium

An agentic customer service system must decide when to escalate interactions to a human agent. Under which of the following conditions should escalati...

Q5
hard

A customer service AI is escalating 40% of all interactions to human agents because it escalates whenever Claude reports confidence below 60%. The sys...

Q6
medium

A customer service agent handles long conversations and begins to miss key account details that were established early in the interaction. Which mitig...

Q7
hard

An AI system generates 5,000 insurance claim decisions per day. Human review of every decision is infeasible. The team has a budget of 200 human revie...

Q8
medium

An AI content moderation system processes 20,000 posts per day. A human review budget allows reviewing 500 posts per day. The system classifies posts ...

Q9
hard

A single Claude agent processes a comprehensive technical analysis task, generating an increasingly long context as it reads documentation, runs analy...

Q10
medium

Why is using LLM self-reported confidence as an escalation trigger unreliable in production agentic systems?

Q11
easy

According to CCA-F exam guidance, what are the three valid programmatic escalation triggers for a Claude agentic system?

Q12
medium

A structured error response from a Claude tool includes `"isRetryable": false` and `"errorCategory": "not_found"`. How should the agentic loop use thi...

Q13
hard

A financial document processing system crashes mid-batch when processing statement 4,200 of 10,000. The system has no crash recovery mechanism. After ...

Q14
medium

An agent processes long customer service conversations and generates increasingly verbose tool outputs. The agent begins ignoring key details from ear...

Q15
medium

An agent maintains a running summary (scratchpad) of its findings as it processes a multi-step research task. At each step, it updates the scratchpad ...

Q16
hard

A production Claude agent processes medical insurance authorization requests. For a batch of 1,000 requests per day, clinical staff can manually revie...

Q17
medium

An agent has been processing a complex multi-step research task for 60 turns and its context is nearly full. New subtasks still need to be completed. ...

Q18
hard

An agentic financial reporting system processes portfolio data overnight. Midway through generation, the system crashes. On restart, there is no recor...

Q19
easy

Where should a developer place critical facts that Claude must reliably reference throughout a long conversation, to minimize the impact of the lost-i...

Q20
medium

An agent processes a 40-turn conversation with many large tool outputs. At turn 38, the agent fails to apply an important security policy that was sta...

Sign in to see all 30 questions

Create a free account to browse all questions — completely free during our launch phase.