02 — 4-week canonical · recommended
Four weeks, one theme each.
Each week has one focus and one output. If you miss the output, repeat the week rather than moving on.
Week1
Foundations and diagnostic~10 h
Take the 10-Q diagnostic before reading anything. practice/diagnostic ↗
Claude 101, the agentic-loop primitive and stop_reason handling. concepts/agentic-loops ↗
Master one concept end to end so you know what 'mastered' means here. knowledge/claude-101 ↗
Done when · diagnostic score, weakest domain named, one concept mastered.
Week2
D1 + D2: architecture and tools (45% of the exam)~12 h
Subagent context isolation and escalation triggers. concepts/subagents ↗
Tool descriptions as the selector; the 18-tool cliff; structured error contracts; tool_choice modes. concepts/tool-calling ↗
Two warm-up drills, one per domain. practice/warm-up ↗
Done when · two warm-ups ≥70% in D1 and D2.
Week3
D3 + D4: Claude Code and prompting~12 h
CLAUDE.md hierarchy, hooks over prompt rules, slash-command scoping, Plan Mode, CI with -p. concepts/hooks ↗
tool_use as structured output, schema-validation retry loops, few-shot for ambiguous tasks. concepts/structured-outputs ↗
Sunday: first timed 30-Q mock. Review every miss against its concept page. practice/mock ↗
Done when · first timed mock done, every wrong answer reviewed.
Week4
D5, scenarios, final mock~10 h
Case-facts block, compaction that keeps identifiers, prompt caching breakpoints. concepts/case-facts-block ↗
Walk 4–6 of the 13 scenarios. scenarios ↗
Final 60-Q mock under exam conditions, then the readiness gate. exam-guide/day-of ↗
Done when · mock ≥850, readiness gate cleared, exam booked.
03 — Final readiness gate
Book the exam when all four are true.
Mock score of 850 or higher on the most recent attempt.
No domain below 70% on the mock's domain breakdown.
You can explain each of the 8 critical anti-patterns without notes.
Day-of checklist read, browser test passed, room confirmed.
04 — If you're between two paths
Three rules.
01
Diagnostic score is the tiebreaker.
≥75% → 2-week or 4-week. 50–75% → 4-week. <50% → 8-week.
02
Hours available wins ties.
If you can't commit 10 hours a week, do not pick the 4-week plan. The 8-week plan covers the same material sustainably.
03
Production experience is the third lever.
If you have shipped agentic systems on Claude, you can compress. If you haven't, you cannot, regardless of test score.
05 — By experience level
How long to prepare.
Two to three weeks is typical for candidates already fluent with tool use, context windows and agentic loops. Most of D1, D2 and D5 maps onto lived experience, so spend the time on D3 and D4 edge cases where exam framing diverges from intuition.
Don't know your score yet?
Ten questions, ten minutes, no account. Then pick the path above.