Skip to content

Cheat sheet: Architect – Foundations

Everything worth holding in your head the hour before the exam, on one page. Facts come from the official exam guide; the rules of thumb are the maintainer's, distilled from the official rationales.

Claude Certified Architect – Foundations cheat sheet: exam facts, domain weights, and the rules that decide questions

Save or share the card above; the full sheet follows.

The exam
Code CCAR-F
Items 60, framed by 4 scenarios drawn from a published bank of 6
Time 120 minutes, about 2 minutes per item
Pass 720 of 1,000, scaled
Fee 125 USD before partner discount, raised from 99 on June 30, 2026
Valid 12 months, free renewal assessment

Where the marks are

Agentic Architecture & Orchestration   27%  ██████████████████████████
Claude Code Configuration & Workflows  20%  ███████████████████
Prompt Engineering & Structured Output 20%  ███████████████████
Tool Design & MCP Integration          18%  █████████████████
Context Management & Reliability       15%  ██████████████

The six scenarios, four of which you will see

Scenario Rehearse
Support resolution agent (MCP tools, 80% first-contact target) Loop termination, escalation triggers, tool scoping, structured errors
Claude Code for a dev team CLAUDE.md hierarchy, path-scoped rules, plan mode, commands and skills
Multi-agent research with cited reports Coordinator and subagent handoffs, provenance, error propagation
Developer productivity tooling Built-in tool selection, MCP integration, large-codebase context
Claude Code in CI/CD Headless invocation, JSON output, review criteria, false positives
Structured data extraction JSON schema design, validation retry, batch processing, confidence

If you remember nothing else

  1. Agent loops need engineered termination. Iteration caps, repeated-call detection, and an escalation path. Token budget is not a stopping condition.
  2. Escalate on policy gaps and lack of progress. Not on turn count, order age, or customer tone.
  3. Structured errors with a category and a retryable flag. An agent can only recover from a failure it can interpret. Use the isError flag.
  4. Tool descriptions are the selection mechanism. Differentiate similar tools precisely, or consolidate them.
  5. Enforce hard limits in the tool layer. Policy caps belong in code or hooks, never in a request to the model.
  6. CLAUDE.md hierarchy: user, project, directory. Conditional conventions go in .claude/rules/ with path scoping so context is spent only where relevant.
  7. Plan mode is a review gate. Use it when a human should approve the approach, not universally.
  8. CI means headless. -p or --print, --output-format json, --json-schema. Never scrape a transcript.
  9. False positives are cured by explicit criteria. State what to flag, what to ignore, the severity bar, and give examples.
  10. Subagents return distilled findings, not raw documents. Bulk stays at the edge; that is the point of delegation.
  11. Provenance must survive handoffs. Carry claim and source together or citations drift from claims.
  12. Errors propagate with context so the coordinator can retry, exclude with a noted gap, or abort.
  13. Nullable for legitimate absence. Required-everything forces fabrication; optional-everything hides real misses.
  14. Validate, then retry with the validation error. Feed the model what was wrong.
  15. Measure on a labeled sample of everything. Accuracy computed only on records that passed validation is a biased number.

Appendix items worth knowing cold

stop_reason values, PostToolUse hooks, allowedTools, the Task tool for subagents, .mcp.json with environment expansion, isError, .claude/skills/SKILL.md frontmatter (context: fork, allowed-tools, argument-hint), /memory, /compact, --resume, fork_session, the Explore subagent, tool_choice, Message Batches (50% saving, 24-hour window, custom_id, no multi-turn tool calling), lost-in-the-middle effects, scratchpad files, confidence calibration.

Working the clock

60 items across 4 scenarios in 120 minutes. The scenarios each carry a block of questions, so pace by block rather than by item.

  • Read each scenario once, carefully, before its questions. Two minutes spent understanding the system saves more than it costs, because every question in the block leans on it.
  • First pass, about 70 minutes. Work each block in order. Flag anything needing more than 90 seconds.
  • Second pass, about 30 minutes. Return to the flags with the whole scenario now familiar.
  • Last 20 minutes. Re-read multiple-response items and confirm selection counts.

Reading a question the way it is written

Because you know the six scenarios in advance, you can walk in already knowing what each one is about to test. Use that: when the block opens, recall the primary domains listed for that scenario and expect the questions to sit there.

Within a question, find the constraint (reliability, cost, context limits, safety, maintainability) and eliminate everything that does not address it. The winning option almost always removes a structural weakness; the distractors add capacity, add supervision, or ask the model to be careful.

The morning of

  • Run the system test again on the exam machine, even if it passed last week
  • Close everything on the shutdown list before check-in, including the Claude desktop app
  • ID ready, room clear, desk clear, second monitor unplugged
  • Check-in takes longer than you expect; start early
  • 135 minutes of seat time; eat first

If you blank

Move on within the block and come back. Scenario questions are mutually reinforcing: a later question in the same block often clarifies what an earlier one was asking.

Exam day logistics

  • ID name must match your registration exactly
  • Closed book. No notes, no documentation, no translation tools
  • Answer everything; unanswered scores zero
  • Read the scenario once carefully, then answer its questions against that context
  • If you attempted this exam before June 30, 2026 and did not pass, that attempt was cleared: no waiting period

Study guide · Notes · Practice questions · Mock exam · Repository index