Claude Certified Architect – Foundations (CCAR-F)

Complete Study Book

Built directly from the official Claude Certified Architect – Foundations Exam Guide, Version 1.0 (effective July 2026). Every chapter maps to a numbered task statement in the guide's §6 objectives. Nothing here is invented scope, and nothing on the guide's out-of-scope list is taught as testable.

1. Exam mechanics — memorize this table

Core technologies tested: Claude Code · Claude Agent SDK · Claude API · Model Context Protocol (MCP).

2. The real blueprint

Domain 1 is the largest single block. If you are short on time, Domain 1 + Domain 3 = 47% of the exam and deserve the over-investment.

3. The six scenarios

Four of these six appear on your sitting, drawn at random. Every item is anchored in one of them, so knowing the scenarios cold means you arrive at each question already holding the architectural context.

Chapter 13 is a dedicated playbook: one section per scenario, with the question archetypes each generates and the correct architectural reflex for each.

4. What is NOT on the exam

The guide lists these as explicitly excluded. Do not spend an hour here — and stay alert, because plausible-sounding out-of-scope options make excellent distractors.

• Fine-tuning or training custom models
• Claude API authentication, billing, or account management
• Language/framework implementation detail beyond tool and schema configuration
• Deploying or hosting MCP servers (infrastructure, networking, containers)
• Claude's internal architecture, training process, model weights
• Constitutional AI, RLHF, safety-training methodology
• Embedding models or vector-database implementation details
• Computer use (browser/desktop automation)
• Vision / image analysis
• Streaming API implementation or server-sent events
• Rate limiting, quotas, or API pricing calculations
• OAuth, API-key rotation, authentication protocols
• Specific cloud-provider configurations (AWS, GCP, Azure)
• Performance benchmarking or model-comparison metrics
• Prompt-caching implementation details (beyond knowing it exists)
• Token-counting algorithms or tokenization specifics

Notably: there is no RAG domain, no vector-store design, no model-selection-by-price item, and no AI Fluency / 4Ds content. If an option offers "switch to a larger context window" or "use a cheaper model," it is almost always a distractor.

5. File map

ccar-f-study-book/
├── README.md                        ← you are here
├── 00-study-plan.md                 ← day-by-day 3-week plan (+2wk and 4wk variants)
├── chapters/
│   ├── 01-exam-strategy.md              How items are built; how to read a scenario
│   ├── 02-agentic-loop.md               D1: loop lifecycle, stop_reason, anti-patterns
│   ├── 03-multi-agent-orchestration.md  D1: coordinator/subagent, Task tool, sessions
│   ├── 04-enforcement-and-hooks.md      D1: gates, hooks, decomposition strategy
│   ├── 05-tool-design.md                D2: descriptions, distribution, tool_choice
│   ├── 06-mcp-integration.md            D2: .mcp.json, isError, resources, built-ins
│   ├── 07-claude-code-config.md         D3: CLAUDE.md, rules, commands, skills
│   ├── 08-claude-code-workflows.md      D3: plan mode, refinement, CI/CD headless
│   ├── 09-prompt-precision.md           D4: explicit criteria, false positives, few-shot
│   ├── 10-structured-output.md          D4: tool_use schemas, validation, batches
│   ├── 11-context-management.md         D5: summarization, position effects, scratchpads
│   ├── 12-reliability-and-provenance.md D5: escalation, error propagation, sampling
│   ├── 13-scenario-playbook.md          All six scenarios + question archetypes
│   └── 14-hands-on-labs.md              The guide's four preparation exercises, expanded
├── cheatsheets/
│   ├── 01-blueprint-and-mechanics.md
│   ├── 02-claude-code-config-decisions.md
│   ├── 03-tools-and-mcp.md
│   ├── 04-agentic-patterns.md
│   └── 05-antipatterns-and-traps.md
├── flashcards/
│   ├── deck-1-agentic-architecture.md
│   ├── deck-2-tools-and-mcp.md
│   ├── deck-3-claude-code.md
│   ├── deck-4-prompting-structured-output.md
│   └── deck-5-context-reliability.md
├── question-banks/
│   ├── qbank-d1-agentic-architecture.md
│   ├── qbank-d2-tools-mcp.md
│   ├── qbank-d3-claude-code.md
│   ├── qbank-d4-prompting-output.md
│   └── qbank-d5-context-reliability.md
└── mock-exams/
    ├── mock-exam-1.md                     60 items / 120 min, blueprint-weighted
    │                                      Scenarios 1, 2, 3, 6 — use this one to predict
    ├── mock-exam-1-answer-key.md          Full rationales + marking sheet + domain tally
    ├── mock-exam-2.md                     60 items, harder, NOT blueprint-weighted
    │                                      Scenarios 4, 5, 1, 3 — stresses D1 + D2
    ├── mock-exam-2-answer-key.md
    └── scoring-grid.md                    Raw → readiness bands, per-domain minimums,
                                           miss-log method, pre-booking checklist

6. Other resources — and where they disagree

One community study guide is worth knowing about: paullarionov/claude-certified-architect. It's a genuinely useful second pass over the same material — 13 theory chapters, per-task-statement domain notes, the 12 official sample questions with explanations, and a practice test. Reading a second explanation of the agentic loop or tool_choice is a good use of an hour.

It independently confirms the blueprint used in this book: 27 / 18 / 20 / 20 / 15, passing score 720 on a 100–1,000 scale, and the same five domains with task statements 1.1–1.7, 2.1–2.5, 3.1–3.6, 4.1–4.6, 5.1–5.6. That agreement is reassuring — two independent readings of the official guide reached the same structure.

⚠️ But it differs from the official exam guide in two places that matter, and where they conflict, the official guide wins.

Both differences would cost you on exam day if you trusted the community version:

• Scenarios 7 and 8 do not exist. Don't study them — that's time spent on content the exam cannot ask about. The official §5 enumeration is closed at six.
• Multiple-response items are real, and they're all-or-nothing. If you practise only single-answer items you'll be unprepared for "select two" and "select three" stems — this book's question banks and both mocks include them deliberately.

It also omits several mechanics the official guide states: the 60-item count, the 120-minute limit, the $125 fee, and the 12-month validity. Section 1 above has those.

🧠 The general rule for any third-party material on a new certification: use it for explanation, never for scope. Blueprints, item counts, and formats come from the official guide only.

7. Marker legend

8. How to use this book

The exam is not a recall test. Almost every item asks "given this situation, what is the most effective first step?" — so passing depends on internalizing a small set of architectural reflexes, not memorizing API surface.

The loop that works:

1. Read a chapter fast (20–30 min), skimming code blocks.
2. Close it and reproduce the "Decision reflexes" box from memory. That box is the examinable content.
3. Run the matching flashcard deck cold.
4. Do the domain qbank. Don't check answers until the whole set is done.
5. Log every miss with three columns: what I picked · what was right · the rule I violated. Only the third column matters. By exam day you should have fewer than 15 distinct rules there, each instantly recognizable.

Weekly: one full mock under real conditions — 60 items, 120 minutes, no notes, no pausing.

Highest-leverage habit: for every miss, name which of the ten Core Reflexes (Chapter 01 §4) you failed to apply. Nearly all misses collapse into those ten.

Readiness signals

You're ready when all three hold:

• Mock ≥ 80% overall with no domain below 70%
• You can state all ten Core Reflexes from memory
• Given any of the six scenarios, you can name its three most likely question archetypes and the right answer shape for each