CCAR-F · blueprint
Claude Certified Architect — Foundations publishes 52 objectives across 5 domains. This is all of them, with the weight each carries and roughly how many of the 60 questions that weight buys.
The weights are the exam's own, not ours. They matter because study time spent proportionally to interest rather than to weight is the most common way a well-prepared candidate still fails: a full mark on the lightest objective here is worth less than a single question in the heaviest.
The question counts are the weight applied to a 60-item sitting. Treat them as the shape of the exam rather than a guarantee: a given sitting varies, and an objective worth 2.4 questions will sometimes ask two and sometimes three.
27.0% of the exam · about 16 of 60 questions · 14 objectives
| Objective | Weight | Questions |
|---|---|---|
| Decompose complex tasks into dynamically generated subtasks | 1.9% | 1.2 |
| Evaluate subagent delegation strategies — goal-oriented vs procedural | 1.9% | 1.2 |
| Apply escalation criteria — honor a human request versus resolve autonomously | 1.9% | 1.2 |
| Design structured handoff packages preserving context, findings and authorization state | 1.9% | 1.2 |
| Explain the agentic loop and stop_reason-driven termination | 1.9% | 1.2 |
| Evaluate multi-agent orchestration patterns — coordinator-worker, parallel, sequential | 1.9% | 1.2 |
| Configure agentic loops to emit parallel tool calls in a single turn | 1.9% | 1.2 |
| Apply session resumption techniques | 1.9% | 1.2 |
| Select the appropriate agentic review architecture — plan mode, direct, multi-phase | 1.9% | 1.2 |
| Design orchestration-layer safeguards — every session ends in resolution or escalation | 1.9% | 1.2 |
| Design state persistence for multi-agent pipelines | 1.9% | 1.2 |
| Diagnose misconfigured subagent spawning | 1.9% | 1.2 |
| Construct self-sufficient subagent prompts | 1.9% | 1.2 |
| Configure subagent invocations with tool restrictions and context scoping | 1.9% | 1.2 |
20.0% of the exam · about 12 of 60 questions · 13 objectives
| Objective | Weight | Questions |
|---|---|---|
| Select the correct Claude Code configuration mechanism — CLAUDE.md, rules, Skills, hooks, permissions | 1.5% | 0.9 |
| Design Claude Code review configurations | 1.5% | 0.9 |
| Configure Claude Code CLI for CI/CD pipelines | 1.5% | 0.9 |
| Distinguish what must be enforced by permissions or hooks from what belongs in CLAUDE.md | 1.5% | 0.9 |
| Apply the context: fork frontmatter option | 1.5% | 0.9 |
| Apply PreToolUse and PostToolUse hooks to enforce policy not left to model discretion | 1.5% | 0.9 |
| Implement PostToolUse hooks that enforce formatting, linting and tests after every edit | 1.5% | 0.9 |
| Structure iterative refinement workflows | 1.5% | 0.9 |
| Determine plan mode versus direct execution from scope, reversibility and review need | 1.5% | 0.9 |
| Select how to supply project context — @ references, CLAUDE.md, or inline description | 1.5% | 0.9 |
| Create and deploy custom slash commands in the correct project or user directory | 1.5% | 0.9 |
| Improve automated test generation quality | 1.5% | 0.9 |
| Select the appropriate built-in tool — Grep, Glob, Read, Bash | 1.5% | 0.9 |
20.0% of the exam · about 12 of 60 questions · 11 objectives
| Objective | Weight | Questions |
|---|---|---|
| Apply extraction accuracy patterns | 1.8% | 1.1 |
| Design extraction schemas — optional, nullable, enums | 1.8% | 1.1 |
| Reduce review false positives with project conventions and exclusion criteria | 1.8% | 1.1 |
| Design feedback loops that capture error metadata and improve prompts and schemas | 1.8% | 1.1 |
| Design human review routing by confidence and field-level ambiguity | 1.8% | 1.1 |
| Select the most reliable structured output method — tool use, prompt formatting, prefill | 1.8% | 1.1 |
| Design prompt criteria with explicit inclusion and exclusion boundaries | 1.8% | 1.1 |
| Design specialized review passes that separate competing concerns | 1.8% | 1.1 |
| Design subagent output schemas for downstream synthesis | 1.8% | 1.1 |
| Design synthesis behavior that preserves source-level uncertainty | 1.8% | 1.1 |
| Resolve structured output truncation by splitting and merging, not raising max_tokens | 1.8% | 1.1 |
18.0% of the exam · about 11 of 60 questions · 9 objectives
| Objective | Weight | Questions |
|---|---|---|
| Write MCP tool descriptions that prevent misrouting | 2.0% | 1.2 |
| Implement MCP tool errors that surface recoverability and suggested next actions | 2.0% | 1.2 |
| Integrate MCP servers — scope, authentication, discovery | 2.0% | 1.2 |
| Distinguish between MCP resources and tools | 2.0% | 1.2 |
| Configure MCP servers and project settings at project versus user scope | 2.0% | 1.2 |
| Configure tool_choice and sequence multi-tool workflows by dependency | 2.0% | 1.2 |
| Expand tool descriptions with examples and disambiguation for similar tools | 2.0% | 1.2 |
| Configure tool distribution so each subagent holds only its role's tools | 2.0% | 1.2 |
| Implement tool use with JSON schemas to enforce structured output compliance | 2.0% | 1.2 |
15.0% of the exam · about 9 of 60 questions · 5 objectives
| Objective | Weight | Questions |
|---|---|---|
| Select the API processing mode — synchronous Messages versus Message Batches | 3.0% | 1.8 |
| Apply context management strategies — subagent isolation, scratchpads, targeted reads | 3.0% | 1.8 |
| Apply context window optimization — summarization, sliding windows, selective retention | 3.0% | 1.8 |
| Apply systematic codebase exploration strategies | 3.0% | 1.8 |
| Explain why conversation history must be resent on every request | 3.0% | 1.8 |
Select the API processing mode — synchronous Messages versus Message Batches is the single heaviest objective at 3.0% — about 1.8 questions. Being weak there costs more than being weak anywhere else on this page, and it is worth knowing before you plan a revision week.
Knowing the weights is not the same as knowing where you stand against them. Our CCAR-F bank draws items in these proportions and re-estimates each objective after every sitting, so a scaled estimate accounts for what the exam actually asks rather than for what you happened to practice.