Claude Courses: The CCAR-F Architect Certification Guide

If you search for claude courses, you will find a mix of general AI tutorials, Anthropic's own Academy content, and third-party prep platforms. This guide cuts through that noise. We focus specifically on what structured learning looks like for the Claude Certified Architect, Foundations exam (CCAR-F), why the exam rewards applied judgment over recall, and how to build a study plan that maps directly to the five weighted domains.

We are an independent prep platform. AI Skill Certs is not affiliated with, endorsed by, or approved by Anthropic.

What is the CCAR-F exam, and why does course selection matter?

The Claude Certified Architect, Foundations exam (exam code CCAR-F) launched on 12 March 2026 as part of Anthropic's Claude Partner Network, a $100M programme. It costs $125 USD per attempt, runs for 120 minutes, and contains 60 scenario-based items scored on a 100-to-1000 scale. The passing score is 720.

Every item is scenario-based. That single fact determines what kind of course material actually helps. A course that drills definitions will leave you under-prepared. A course that walks you through realistic decision trees, trade-off analysis, and root-cause reasoning will not.

The exam draws four scenarios at random from a bank of six at each sitting, so you cannot predict which scenarios will appear. Breadth across all five domains is not optional.

What are the five CCAR-F domains and how are they weighted?

Understanding domain weights is the first act of intelligent course selection. Spending equal time on every domain is a common mistake; the exam does not weight them equally.

Domain 1 alone accounts for more than a quarter of the exam. Any course plan that does not front-load Agentic Architecture & Orchestration is misallocating study time. Domains 3 and 4 are tied at 20% each, making them the second priority tier. Domain 5, at 15%, is the smallest but covers reliability patterns that appear as the deciding factor in many scenario items.

What does each domain actually test?

Domain 1: Agentic Architecture & Orchestration (27%)

This domain tests your ability to design and debug multi-agent systems. Expect scenarios involving hub-and-spoke architecture, coordinator responsibilities, subagent context isolation, and how to handle errors that propagate across agent boundaries. The exam consistently rewards deterministic solutions over probabilistic ones when stakes are high.

Key concepts to master: how a coordinator selects subagents dynamically, how to prevent agentic loop anti-patterns, and when to fork a session versus resuming one. Courses that only cover the Messages API at a surface level will not prepare you for the depth of reasoning these items require.

Domain 2: Tool Design & MCP Integration (18%)

This domain tests tool interface design, MCP scoping, and error handling. The exam distinguishes between an access failure and a valid empty result, between a tool that is too generic and one that is appropriately constrained. Tool Design & MCP Integration concepts include the MCP isError flag pattern, structured error metadata, and how to write tool descriptions that function as selection mechanisms rather than documentation.

Domain 3: Claude Code Configuration & Workflows (20%)

Claude Code configuration is tested at the level of the three-level configuration hierarchy, version control implications, and how to design workflows that are reproducible and auditable. Courses that treat Claude Code as a chat interface miss the point entirely. The exam tests architectural judgment about when to use hooks, how to scope configuration, and how to design CI/CD-adjacent workflows.

Domain 4: Prompt Engineering & Structured Output (20%)

Prompt Engineering & Structured Output is tested through scenario items that ask you to diagnose why a structured output pipeline is failing, choose between goal-based and step-based prompts, or decide when few-shot examples are the highest-leverage intervention. The exam does not reward memorised prompt templates; it rewards the ability to reason about why a given approach will or will not work in a specific context.

Domain 5: Context Management & Reliability (15%)

Context Management & Reliability covers the stale context problem, summary injection for fresh sessions, and the attention dilution problem that emerges in long contexts. Reliability items often hinge on whether a candidate understands the difference between a proportionate fix and an over-engineered one.

What types of claude courses exist, and which formats work?

Not all course formats serve the CCAR-F equally well. Here is an honest comparison.

The CCAR-F exam's scenario-based format means passive consumption of video content has a low return on study time. The items require you to apply a decision framework under time pressure, not retrieve a fact. Courses that build that decision-making muscle through repeated practice with feedback are structurally better suited to this exam.

How does adaptive prep differ from a standard course?

Our platform uses Bayesian Knowledge Tracing with a 0.90 mastery threshold. In practice, that means the system tracks your probability of having mastered each of the 174 atomic concepts mapped to the five CCAR-F domains and 30 task statements. It surfaces the concepts where your mastery estimate is lowest, rather than cycling through material you already understand.

A standard course delivers content in a fixed sequence. An adaptive system delivers content in the sequence that closes your specific gaps fastest. For a 60-item exam with five weighted domains, that difference in efficiency is material.

Our concept library at /concepts covers all 174 atomic concepts across the five domains. Each concept page is mapped to the task statements it supports, so you can trace from a practice item you got wrong back to the underlying concept and then to related items that test the same skill.

What is Archie, and how does it fit into a study plan?

Archie is our Socratic tutor. It does not give you the answer directly; it guides you through graduated hints until you arrive at the reasoning yourself. That approach mirrors what the exam actually demands: you will not have an answer key in the room. You need to be able to construct the correct reasoning from first principles.

Archie is most useful after you have attempted a practice item and got it wrong. Rather than reading the explanation passively, working through Archie's hints forces you to identify exactly where your reasoning diverged from the correct path. That active error analysis is what converts a wrong answer into a durable skill.

How should a study plan be structured for CCAR-F?

A reasonable eight-week plan allocates time proportional to domain weight and reserves the final two weeks for full practice exams.

Practice exams on our platform mirror the real format: 60 items, scored 100 to 1000, with 720 as the passing bar. The score report breaks out percent-correct by domain, which tells you exactly where to focus targeted review in weeks 7 and 8.

The exam draws four scenarios from a bank of six at each sitting. That means you will see different scenario framings across practice attempts, which is intentional. Familiarity with a single scenario framing is not the same as mastery of the underlying domain.

What concepts within Domain 1 carry the most exam weight?

Domain 1 is the largest single domain at 27%, and it is also the most architecturally complex. Within it, the concepts that appear most frequently in scenario items cluster around a few themes.

Coordinator design: How a coordinator selects subagents dynamically, how it passes structured context, and how it handles errors that subagents return. See Coordinator Dynamic Subagent Selection and Structured Context Passing.

Session management: When to resume a session, when to fork it, and when to start fresh. The Stale Context Problem is a recurring scenario trigger.

Decomposition strategy: The difference between fixed sequential pipelines and dynamic adaptive decomposition, and when each is appropriate. Narrow decomposition failure is a common wrong-answer trap.

Hook architecture: When to enforce rules programmatically via hooks versus via prompt instructions, and how to design a hook pipeline that does not introduce latency or failure modes. See Hooks vs Prompts Decision Framework.

How does CCAR-F compare to the other tracks in the Claude Partner Network?

The Claude Partner Network launched four proctored Pearson VUE exam tracks on 12 March 2026. Understanding where CCAR-F sits helps you decide whether it is the right starting point.

CCAR-F is the architect track. It tests system design judgment, orchestration patterns, and configuration strategy. CCDV-F is the developer track and tests implementation skills including API integration, model selection, and security. The two tracks are complementary but distinct; the domain structures do not overlap directly.

As of 3 June 2026, more than 10,000 individuals hold a Claude Partner Network certification across all tracks. The credential is valid for 12 months from the date it is awarded.

Our platform currently offers adaptive prep, Archie tutoring, and practice exams for both CCAR-F and CCDV-F. CCAR-P and CCAO-F prep are not yet available.

What should you look for when evaluating any claude course?

Before committing time and money to any course, apply these four tests.

1. Scenario coverage: Does the course use scenario-based items, or does it test recall? If the sample questions look like "what is the default context window?", the course is not aligned to the exam format.
2. Domain mapping: Does the course explicitly map content to the five CCAR-F domains and their weights? A course that treats all topics as equally important will leave you over-prepared in low-weight areas and under-prepared in high-weight ones.
3. Feedback quality: When you get something wrong, does the course explain the reasoning behind the correct answer, or just tell you the answer? Reasoning explanations build transferable judgment; answer keys do not.
4. Format alignment: The exam is 120 minutes for 60 items, which is two minutes per item. Any course that does not include timed practice is not preparing you for the actual test-taking experience.

A course that passes all four tests is worth your time regardless of who produces it. A course that fails on scenario coverage is unlikely to move your score meaningfully, regardless of how polished the production is.