Claude Certified Architect – Foundations
What the exam is, what it costs, who can sit it, how it is scored, and how to prepare. Everything below comes from the official Exam Guide v1.0 (effective July 2026), the Anthropic Partner Academy FAQ and Pearson VUE's program page, and it is dated so you can tell what might have changed.
01 Facts02 Eligibility03 Domains04 Scenarios05 Scoring06 Preparation07 Deep dives
01 / Exam facts
| Items | 60 multiple-choice and multiple-response items; each item states how many answers to select |
|---|---|
| Structure | 4 scenarios drawn at random from a bank of 6, each framing a block of questions |
| Time | 120 minutes |
| Passing score | 720 scaled, on a 100–1,000 scale |
| Fee | $125 USD per attempt (was $99 before June 30, 2026); 50% off for tiered partners, free for Global Premier partners through December 31, 2026 |
| Delivery | Pearson VUE: online proctored or test center. Cancel or reschedule up to 24 hours before, or forfeit the fee |
| Validity | 12 months. On-time renewal is a free, non-proctored assessment; a lapsed credential means the full exam again |
| Retakes | Wait 14 days after the first fail, 30 after the second, 90 after the third; at most 4 attempts per rolling 12 months; the fee applies each time |
| Guide | Exam Guide v1.0 (July 2026), downloadable from the Partner Academy certification page. CCAR-F is not part of the December 7, 2026 content refresh that changes the other three Claude exams |
02 / Who can take it
Only people who work at organizations in the Claude Partner Network. Registration requires a company email on a domain listed in the partner's record; personal addresses are rejected, and the FAQ says non-partners "can't register for a certification exam." Anthropic has not announced a public opening. If your company is a partner, your enablement lead can usually confirm the discount tier and whether exam fees are reimbursed.
03 / Domains and weights
| Domain | Weight | Task statements |
|---|---|---|
| 1 · Agentic Architecture & Orchestration | 27% | agentic loops and stop_reason; coordinator–subagent patterns; context passing and parallel spawning; enforcement and handoff patterns; Agent SDK hooks; task decomposition; sessions, resume and forking |
| 2 · Tool Design & MCP Integration | 18% | tool descriptions and boundaries; structured MCP errors; tool distribution and tool_choice; MCP server scoping and resources; built-in tools (Read, Write, Edit, Bash, Grep, Glob) |
| 3 · Claude Code Configuration & Workflows | 20% | CLAUDE.md hierarchy and imports; slash commands and skills; path-scoped rules; plan mode vs direct execution; iterative refinement; CI/CD with claude -p |
| 4 · Prompt Engineering & Structured Output | 20% | explicit criteria; few-shot prompting; structured output via tool use and JSON schemas; validation and retry loops; Message Batches; multi-instance and multi-pass review |
| 5 · Context Management & Reliability | 15% | context preservation; escalation and ambiguity; error propagation; large-codebase exploration; human review and confidence calibration; provenance and conflicting sources |
Out of scope, per the guide: fine-tuning, API auth and billing, computer use, vision, streaming, rate limits and pricing math, cloud-provider configuration, prompt-caching internals, tokenization, and model internals.
04 / The six scenarios
- Customer Support Resolution Agent (Agent SDK;
get_customer,lookup_order,process_refund,escalate_to_human; 80% first-contact resolution target) - Code Generation with Claude Code (slash commands, CLAUDE.md, plan mode vs direct execution)
- Multi-Agent Research System (coordinator with web-search, document-analysis, synthesis and report subagents; cited reports)
- Developer Productivity with Claude (built-in tools plus MCP servers for exploring unfamiliar codebases)
- Claude Code for Continuous Integration (automated reviews, test generation, PR feedback; actionable output, few false positives)
- Structured Data Extraction (JSON-schema validation, edge cases, downstream integration)
You see four of these per sitting. Prepare for all six; candidates report that the CI/CD scenario carried more git and pipeline questions than they expected.
05 / How scoring works
The exam is criterion-referenced: you pass by clearing a fixed standard (720), not by beating other candidates. Scaled scoring equates forms of slightly different difficulty, so there is no published "number of correct answers" that guarantees a pass; on a linear scale 720 corresponds to roughly 69% correct, which is why we recommend 800+ on a fresh practice form as the comfortable target. Your score report shows pass/fail, the scaled score, and percent-correct per domain; the domain percentages are informational.
06 / How to prepare
- Read the Exam Guide in full. The 30 task statements each list "knowledge of" and "skills in" bullets; exam items are written against those bullets, and so are ours.
- Take the 25-question diagnostic to find your weak domains, then drill them by task statement.
- Build the four things the guide's "How to prepare" section lists: an agent with a real agentic loop and hooks, a Claude Code setup with CLAUDE.md hierarchy, rules and a skill, MCP tools with structured errors, and an extraction pipeline with tool-use schemas and a retry loop.
- Learn the answer patterns. Deterministic enforcement over prompt instructions when a rule is critical; the proportionate first step over new infrastructure; the upstream root cause over a downstream symptom; structured context over generic status; least privilege for tools; batch only when latency-tolerant; an independent reviewer over self-review.
- Take full timed forms, then fresh ones. A high score on a form you have seen is memorization. Our multiple-response items and randomized forms exist for that reason.
- On exam day, read every stem twice. Passers say the deciding qualifier (cost, latency, "first step", "without changing X") sits in the middle of a long sentence.
07 / Deep dives
- How much the CCAR-F exam costs, discounts, retake fees and what prep is worth paying for
- The 720 passing score explained: scaled scoring, domain percentages, and what to aim for on mocks
- Exam format: 60 items, four scenarios, multiple-response, 120 minutes, and how the form is built
- Multiple-response ("Select TWO") items: how they are scored and how to practice them
- The CI/CD scenario:
claude -p, structured output, re-runs, and the questions candidates say surprised them - Where the exam guide and today's Claude docs disagree, and which answer the exam wants