Module 10 opening video · ~3 minutes · the paper, the clock, the flag habit, the trap, and the gate that says when to book
| Item | Value |
|---|---|
| Questions | 60, multiple choice, one correct of four |
| Time | 120 minutes |
| Structure | 4 scenarios drawn from a bank of 6 |
| Scoring | Scaled 100 to 1,000; 720 passes |
| Delivery | Online proctored or a test center |
| Fee | $125 |
| Valid | 12 months |
| Unanswered | Impossible: the platform requires an answer before you advance |
| Domain | Weight | Of 60 |
|---|---|---|
| 1 Agentic Architecture and Orchestration | 27% | about 16 |
| 2 Tool Design and MCP Integration | 18% | about 11 |
| 3 Claude Code Configuration and Workflows | 20% | 12 |
| 4 Prompt Engineering and Structured Output | 20% | 12 |
| 5 Context Management and Reliability | 15% | 9 |
The question counts are the weights applied to 60 and rounded; the guide publishes the weights, and the colored sheet on the left is one plausible spread of them.
Scenario 1: Customer Support Resolution Agent
You are building a customer support resolution
agent using the Claude Agent SDK. It handles
returns, billing disputes, and account issues
through custom MCP tools (get_customer,
lookup_order, process_refund, escalate_to_human).
Target: 80%+ first-contact resolution while
knowing when to escalate.
Q1 agent skips get_customer in 12% of cases ...
Q2 both tools have minimal descriptions ...
Q3 55% first-contact resolution ...
Read the stem once, properly. Every question
under it reuses the same four tools and the
same 80 percent target.120 minutes / 60 questions = 2 minutes each
Checkpoints:
question 15 by minute 30
question 30 by minute 60
question 45 by minute 90
last 30 minutes: the flag pile
Ahead of the line: relax, read the stems well.
Behind it: speed up on the short questions.
Never on a scenario stem; the stem pays out
four times.Q7. Two options are defensible. You have read
it twice.
[ ] A [ ] B [x] C (best guess) [ ] D
[F] flagged for review
Pick the best guess, because the platform
will not let you advance without one. Flag it.
Move on without guilt. Momentum is worth more
than any single question, and question 7 will
still be there at minute 95.Flags: 7, 19, 33, 41
You have now seen the whole paper. Question 33
asked about structured error responses, and
question 41's stem described exactly the
tool_choice setup question 7 was about.
Questions answer each other.
Change an answer only when you can name the
reason. A vague feeling that the first pick
was right is a myth with a literature; a
specific reason you missed is evidence.Q: Both tools have minimal descriptions and
the agent picks the wrong one a fifth of the
time. First step?
A) Eight few-shot routing examples
B) Rewrite the descriptions: inputs, example
queries, edge cases, when to use which
C) A keyword router in front of the model
D) Merge the two tools
A, C, and D all work in some shop. B is what
the documentation recommends, and B is the
guide's own answer. When two options are
defensible, pick the documented pattern over
the clever one. This exam rewards orthodoxy.| Domain | Score | Gate |
|---|---|---|
| 1 | 78% | NO GO |
| 2 | 91% | GO |
| 3 | 88% | GO |
| 4 | 86% | GO |
| 5 | 90% | GO |
Average 86.6, and it does not matter. Domain 1 is 27 percent of the paper and it is under the line. A strong average hiding a weak domain is how people fail a weighted exam. Drill domain 1, then book.