How many hours per week do I actually need for the CCA-F?

Most working professionals with solid LLM exposure land in the 6 to 10 hour range per week across six weeks, which is roughly 36 to 60 total hours. If your daily work already involves Claude Code, agents, or MCP, you can lean toward the lower end. If LLM work is something you read about but rarely build, plan for the upper end and extend the plan by a week if your week-three diagnostic shows accuracy below 60%.

Can I compress the CCA-F study plan into three or four weeks?

Yes, but only by raising weekly hours, not by skipping phases. A compressed schedule still needs a cold diagnostic, domain depth in blueprint-weighted order, and at least two full timed simulations. Compressing usually means 14 to 16 hours per week instead of 8, and you should expect lower retention. If you have a hard deadline, keep the structure and double the hours rather than dropping weeks four or six.

What should I do in week one if I have no LLM experience at all?

Add a pre-week zero focused on Anthropic's public documentation and the /study-guide overview. Read the model behavior basics, work through a few prompt engineering examples, and try a small Claude Code task end to end. Without this foundation, the week-one diagnostic on /free-practice will be noisy and demoralizing rather than diagnostic. Once you can follow a simple agent loop in your head, start the six-week plan.

How is this study plan weighted against the real CCA-F blueprint?

Weeks two through five mirror the published domain weights. Agentic Architecture gets the most concentrated time at 27%, Tool Design and MCP comes next at 18%, then Claude Code and Prompt Engineering each at 20%, and Context Management at 15%. Allocating study hours roughly proportional to domain weight is the single highest-leverage scheduling decision, since the exam scoring reflects the same distribution.

When should I take my first full 60-question timed simulation?

End of week three at the earliest, week four for most people. Earlier sims tend to measure unfamiliarity with the format rather than knowledge gaps, which wastes a scarce resource. Save full sims for after you have studied at least two domains in depth. Week six should include three or more full sims under exam conditions, with focused drilling on weak areas in between rather than back-to-back simulations.

What score on practice sims roughly predicts a passing 720 on the real exam?

Treat this as guidance rather than a guarantee. Most candidates who consistently hit 75 to 80% accuracy on full timed 60-question simulations across multiple attempts, with no single domain below 65%, are in a reasonable position. A single high score is less predictive than three sims clustered above 75%. If your last sim is much higher than the previous two, repeat it before trusting the number.

Claude Certified Architect Study Plan: A 6-Week CCA-F Schedule You Can Drop Into Notion

A copy-paste-ready 6-week CCA-F study plan for working professionals at roughly eight hours per week, phased by real blueprint weight with concrete accuracy targets.

If you are searching for a Claude Certified Architect study plan, you are almost certainly past the question of whether to take the CCA-F and into the question of how to structure the next month or two without wasting evenings. This page gives you a concrete six-week schedule built around roughly eight hours of study per week, designed to be copied straight into Notion, Apple Notes, or whatever you actually use to run your week.

The plan is organised around the real CCA-F blueprint weights. Agentic Architecture is the largest domain at 27%, followed by Claude Code and Prompt Engineering at 20% each, Tool Design and MCP at 18%, and Context Management at 15%. You will spend more hours where the exam spends more questions. That sounds obvious, and yet most ad-hoc study plans drift toward whichever domain the candidate finds most interesting, which is rarely the one carrying the most weight.

Everything below assumes you are using claudecertifiedarchitect.dev as your practice platform, with 1,042 scenario-based questions across the five domains and a 60-question 120-minute simulator scaled 100 to 1000 with 720 to pass. You can confirm the structure on /exam-format and start with the 15 free questions on /free-practice before committing. The plan works with other resources too, but the targets and accuracy benchmarks below are calibrated against this question bank.

Who this plan is for

This plan is written for a specific reader. You are a working professional with at least six months of meaningful exposure to large language models. You have written prompts that survived production, debugged something with tool use or function calling, or read enough Anthropic documentation that terms like context window, system prompt, and agentic loop are not abstract. You have between six and ten hours per week to spend, almost all of it in the evenings or on weekends, and you have not previously prepared for the CCA-F.

If that describes you, the schedule below should feel demanding but reasonable. If you have less LLM exposure, treat weeks one and two as longer than written and budget an extra week of foundational reading from /study-guide before starting. If you are already shipping agentic systems professionally, you can probably compress this to four or five weeks by trimming the week-four and week-five reading time, but do not skip the week-six simulations.

This plan also assumes you can protect study time on a calendar. The single largest reason candidates fail to finish a plan like this is not difficulty but slippage. Block the hours in advance and treat them as meetings you cannot move. Eight hours a week is roughly four ninety-minute sessions or two longer Saturday and Sunday blocks. Pick whichever pattern matches how you already get focused work done.

How the 6 weeks break down

The six weeks are split into three phases. Phase one is week one, where you establish a baseline. Phase two is weeks two through five, where you go deep on each of the five domains in order of blueprint weight. Phase three is week six, where you stop learning new material and start rehearsing under timed conditions.

The ordering matters. Spending your earliest, freshest study time on the heaviest domain, Agentic Architecture at 27%, means that domain has the longest spaced-repetition window before exam day. Lighter domains studied closer to the exam stay fresh in working memory. This is roughly the opposite of the natural temptation, which is to start with whichever domain feels most approachable. Resist that.

A quick map of the weeks. Week one is reading the blueprint, taking a cold diagnostic, and calibrating expectations. Weeks two and three are Agentic Architecture and Tool Design and MCP, the two domains where most candidates lose points. Weeks four and five are Claude Code, Prompt Engineering, and Context Management, packed slightly tighter because they tend to overlap with day-job experience. Week six is at least three full 60-question simulations on /exam-format style timing, with focused drilling on weak spots between each sim. No new material in week six. You are consolidating, not learning.

Week 1 — Foundations and diagnostic

Week one is short on practice and long on orientation, and that is deliberate. Spend the first two hours reading the official CCA-F blueprint and the overview on /study-guide. Note the five domains, their weights, and what each one actually tests. Write the five domain names and percentages on a sticky note somewhere visible. You will reference them constantly.

Spend the next two hours skimming /domains/agentic-architecture, /domains/tool-design-mcp, /domains/claude-code, /domains/prompt-engineering, and /domains/context-management. You are not trying to learn the content. You are trying to map the territory so that when a practice question references, for example, MCP resource exposure, you know which domain that lives in.

The final four hours of week one are your cold diagnostic. Take the 15 free questions on /free-practice or the public set on /sample-questions, then a 40-question mixed-difficulty set if you have access to the full bank. Take it under loose timing, roughly two minutes per question, and do not look anything up. Score honestly by domain. Your week-one diagnostic is not meant to be encouraging. Most strong candidates score between 45 and 60% cold. A score in that range is normal. A score above 70% cold likely means you are over-prepared and can compress the plan. A score below 35% means you should add an extra foundational week before continuing.

Weeks 2–3 — Domain depth (largest first)

Weeks two and three are the heaviest. You will spend roughly five hours per week on Agentic Architecture and three hours on Tool Design and MCP, plus the weekly checklist. Start with Agentic Architecture because at 27% it is the single largest source of exam points and the domain where conceptual gaps cause cascading errors elsewhere.

For Agentic Architecture, read /domains/agentic-architecture end to end at the start of week two. Focus topics: agent loops, planning versus execution, when to delegate to a sub-agent versus extend the current context, failure modes around tool-call retry loops, and the trade-offs between fully autonomous and human-in-the-loop designs. Drill 25 to 30 scenario questions in this domain, then re-derive every wrong answer from first principles. By the end of week two you should be hitting 70% on Agentic Architecture mixed-difficulty sets.

Week three shifts to Tool Design and MCP at 18%. Read /domains/tool-design-mcp and pay close attention to the boundary between tool definition, tool description, and the model's decision to call a tool. Focus topics: schema design, idempotent versus side-effecting tools, MCP server architecture, and resource versus tool exposure. Target 70% on this domain by the end of week three. Keep Agentic Architecture warm with one 20-question mixed set during week three.

Weeks 4–5 — Remaining three domains

Weeks four and five cover the three remaining domains. The combined weight is 55%, but these domains tend to overlap more with day-job intuition, so you will move faster than in weeks two and three. Treat week four as Claude Code plus the start of Prompt Engineering, and week five as the finish of Prompt Engineering plus Context Management.

Claude Code at 20% comes first. Read /domains/claude-code and focus on the architectural questions: when Claude Code is the right tool, how it differs from a generic agent loop, file editing and verification patterns, and how to scope tasks for reliable completion. Drill 25 questions, then write a short note on the three most common failure modes you got wrong. Target 70% accuracy by the end of week four.

Prompt Engineering at 20% spans the boundary of week four and five. Read /domains/prompt-engineering and pay attention to system prompt structure, few-shot patterns, output formatting, and how prompting interacts with tool use. Drill 30 questions across the boundary.

Context Management at 15% closes week five. Read /domains/context-management and focus on context window economics, summarization strategies, retrieval boundaries, and when to start a new conversation versus extend the current one. Drill 20 questions. By end of week five every domain should be at or above 70%, with no domain below 65%.

Week 6 — Full simulations and weak-spot drilling

Week six is rehearsal. No new domain reading, no new concepts. You are converting knowledge into exam performance, which is a distinct skill.

Take your first full 60-question 120-minute simulation on the morning of day one. Use the format described on /exam-format: scaled 100 to 1000, 720 to pass, no pausing. Score it by domain. Whatever your two weakest domains are, spend the next two evenings drilling 40 to 50 targeted questions in those domains, re-deriving every miss from first principles. Do not retake the same sim.

Midweek, take a second full simulation under the same conditions. Compare the domain-level scores against sim one. Any domain that dropped is a sign of shallow knowledge rather than a bad day, and deserves another focused drilling session. Any domain that stayed flat at 70 to 75% is ready.

Take a third full simulation on the weekend, ideally at the same time of day you will sit the real exam. Three sims is the floor. A fourth in the final two days is helpful if your scores are still climbing. Stop new drilling 24 hours before the exam. Use that final day for light review of your running spreadsheet of missed questions, a full night of sleep, and confirming the logistics of how you will actually sit the exam.

Weekly checklist (every week)

·Take at least one timed 20-question set with the clock visible, not paused, so pacing pressure becomes routine rather than novel on exam day.
·Re-derive one wrong answer from first principles each week, writing a short note explaining why the correct option wins on architectural grounds.
·Review one domain deep-dive on /domains/[slug] and reconcile any gap between the deep-dive language and how the practice questions phrase the same concept.
·Log every missed question into a single running spreadsheet with columns for domain, sub-topic, root cause, and a one-line lesson.
·Read one short primary source per week from Anthropic's public documentation on the domain you are currently studying.
·Spend thirty minutes on free-form recall, writing out the architecture of a small agentic system from memory without looking anything up.
·End the week with a fifteen-minute review of the running spreadsheet, marking any lesson that has appeared twice as a priority drill for next week.

Frequently asked questions

How many hours per week do I actually need for the CCA-F?: Most working professionals with solid LLM exposure land in the 6 to 10 hour range per week across six weeks, which is roughly 36 to 60 total hours. If your daily work already involves Claude Code, agents, or MCP, you can lean toward the lower end. If LLM work is something you read about but rarely build, plan for the upper end and extend the plan by a week if your week-three diagnostic shows accuracy below 60%.
Can I compress the CCA-F study plan into three or four weeks?: Yes, but only by raising weekly hours, not by skipping phases. A compressed schedule still needs a cold diagnostic, domain depth in blueprint-weighted order, and at least two full timed simulations. Compressing usually means 14 to 16 hours per week instead of 8, and you should expect lower retention. If you have a hard deadline, keep the structure and double the hours rather than dropping weeks four or six.
What should I do in week one if I have no LLM experience at all?: Add a pre-week zero focused on Anthropic's public documentation and the /study-guide overview. Read the model behavior basics, work through a few prompt engineering examples, and try a small Claude Code task end to end. Without this foundation, the week-one diagnostic on /free-practice will be noisy and demoralizing rather than diagnostic. Once you can follow a simple agent loop in your head, start the six-week plan.
How is this study plan weighted against the real CCA-F blueprint?: Weeks two through five mirror the published domain weights. Agentic Architecture gets the most concentrated time at 27%, Tool Design and MCP comes next at 18%, then Claude Code and Prompt Engineering each at 20%, and Context Management at 15%. Allocating study hours roughly proportional to domain weight is the single highest-leverage scheduling decision, since the exam scoring reflects the same distribution.
When should I take my first full 60-question timed simulation?: End of week three at the earliest, week four for most people. Earlier sims tend to measure unfamiliarity with the format rather than knowledge gaps, which wastes a scarce resource. Save full sims for after you have studied at least two domains in depth. Week six should include three or more full sims under exam conditions, with focused drilling on weak areas in between rather than back-to-back simulations.
What score on practice sims roughly predicts a passing 720 on the real exam?: Treat this as guidance rather than a guarantee. Most candidates who consistently hit 75 to 80% accuracy on full timed 60-question simulations across multiple attempts, with no single domain below 65%, are in a reasonable position. A single high score is less predictive than three sims clustered above 75%. If your last sim is much higher than the previous two, repeat it before trusting the number.

Drop this plan into Notion and start week one

Start with the 15 free questions, then pace yourself through the six weeks with the full 1,042-question bank and timed simulator. Lifetime access is $24.99 with a 7-day refund.

Get Lifetime Access — $24.99

Or try 15 free questions first, or see 10 free sample questions.