A good Claude Certified Architect practice exam

The test of a practice exam is whether one of its questions could be slipped into the real one without anyone noticing. Practice material built from definitions to recall, true-or-false statements and a right answer that is always the longest option fails that test on the first page. The Claude Certified Architect — Foundations exam asks none of that. It describes a production system built on Claude that is misbehaving and asks what you would do, 60 times in 120 minutes, and a practice exam has to do the same or it is practice for a different exam.

Match the format exactly

A shorter practice exam is fine. A smaller set at the same pace, in the same shape and the same mix is practice. Every row above except the question count and the time limit is a property of the questions and the marking, so a shorter set keeps all of them.

Ask about a system that is misbehaving

Every real question is one scenario with a symptom and four remedies, and the stem has four parts that each do a job. A practice exam written by someone who knows this reads like the exam; one written from a glossary reads like a glossary.

• The scenarioIts jobSets the domain. Does not decide the answer.What a good author doesA concrete production setting: a support agent, a CI job, an extraction pipeline. Never the detail that decides the answer.
• The symptomIts jobNames the failure. Points to the task statement being tested.What a good author doesStates what was observed, never why. The cause is what the question tests, so it cannot be in the stem.
• The constraintIts jobKills whole families of answers at a stroke.What a good author doesOptional, and decisive when present: "must never", "cannot be bypassed", "with the context intact". It is what makes a family of wrong options unambiguously wrong.
• The askIts jobThe final sentence. Decides which kind of answer wins.What a good author doesOne of five phrasings, and never ambiguous between two of them.

Read the ask first, then the constraint, then the symptom and the scenario. A practice exam that puts nothing in the constraint and always ends "what should you do?" cannot be read that way, so it never trains the reading. How to answer CCAR-F questions is that method from the candidate's side.

Use the five asks, and never blur two

The last sentence of a real question is one of five, and each wants a different kind of answer from the same scenario. A practice exam that only ever asks for "the best fix" trains none of this.

• The root causeWhat it wantsThe earliest point where information was lost or never assignedWhat it throws awayEverything downstream, however true
• The best first stepWhat it wantsThe lowest-effort change that addresses the causeWhat it throws awayValid fixes that cost more than the first step needs
• The most effective solutionWhat it wantsThe highest-leverage change, effort no objectWhat it throws awayPartial fixes that leave the cause in place
• The correct approachWhat it wantsThe mechanism whose guarantee matches the stakesWhat it throws awayAnything probabilistic when the stem says always
• One choice for each of three casesWhat it wantsOne axis, applied consistently to all threeWhat it throws awayOptions that treat the three the same

Build the wrong options the way the exam does

The real exam's wrong options come from a small set of families, and each family is wrong for a reason a candidate can learn to see. A practice exam whose distractors are merely untrue teaches recognition of nonsense, which the exam never offers. The wrong-answer patterns walks through all six; in one line each:

1. 1The capacity fix. A larger context window, a higher-tier model, more retries, a bigger iteration cap. The problem is quality.
2. 2Wrong side of the line. Something that detects after the fact where the stem demands prevention. A strong distractor in enforcement questions.
3. 3The over-engineered fix. A classifier, a custom server, a retraining loop, where rewriting a description would do.
4. 4The premature fix. Correct, but not first, so it answers a different ask.
5. 5The symptom tuner. Adjusts the number the failure was reported in and leaves the cause alone.
6. 6The invented mechanism. A setting or guarantee that does not exist. It sounds plausible, and a candidate can check it.

Several constraints keep the options honest, and a careless practice exam breaks each of them. The three wrong options should come from at least two different families, so a candidate cannot clear a question by spotting one pattern. Exactly one of them is the strongest distractor, and the author has to be able to say why it fails. The key must not be reliably the longest option, and the keys must be balanced across the four letters over any set, or the set teaches a tell. And the key has to stay correct if the scenario's numbers changed: a distractor may be tuned to a figure in the stem, a key never may, because the exam tests principles.

Deal across the domains at the exam's weights

The exam guide splits the content into five domains with stated weights, and a practice exam should deal to the same proportions, because the proportions are what make a raw score comparable with the real thing.

• 1. Agentic Architecture & OrchestrationWeight: 27%Questionsabout 16
• 2. Tool Design & MCP IntegrationWeight: 18%Questionsabout 11
• 3. Claude Code Configuration & WorkflowsWeight: 20%Questionsabout 12
• 4. Prompt Engineering & Structured OutputWeight: 20%Questionsabout 12
• 5. Context Management & ReliabilityWeight: 15%Questionsabout 9

The deal also has to mix. A practice exam that runs one domain to exhaustion before the next trains recall inside a topic; the real exam, and the real job, demand telling which principle a scenario calls for when the scenarios arrive in any order. A practice exam can go further and deal in small groups built the way the real exam builds them: one scenario asked twice, for the root cause and then the first step; one setting varied by a single decisive detail so that two different remedies become correct; one principle dressed in three different domains' clothes. Each group is a test of discrimination, which is the skill the exam tests.

Explain every option, not only the key

A practice exam that marks you right or wrong and moves on has done half its job. The learning is in the wrong options. Why the strongest distractor fails is what the question exists to teach, and the other two are worth a line each. And a miss should be filed under the principle behind it, so that the principle comes back, in its cards and in other questions that test it, and the one question you remember missing comes back with it.

Keep measurement away from practice

A number you are asked to trust has to come from questions you have never seen. Answer a question twice and the second answer is partly memory of the first, and a score built on seen questions says how familiar the set is. So a good practice exam comes in two kinds and keeps them apart: practice at the exam's length on questions you have met, as often as you like, with a raw score and no scaled one; and measurement on fresh questions, timed, dealt at the exam's weights, scaled to the 1000-point scale, taken rarely, with questions held back for it that nothing else is allowed to spend.

This is the specification Spaced's questions are written to

Everything above is the item specification our own question bank is authored against, and the 340 questions behind the quizzes are held to it: the four-part stem, one of the five asks, three wrong options on a single-answer question and two on a select-two, drawn from at least two families, with the strongest one named and its failure explained, keys balanced across the letters, a rationale on the key and on the strongest distractor, and the key written to survive the scenario's numbers changing. The nightly quiz deals them in scenario settings, mixed across the five domains in a different order each night, and every miss files under its principle: the question comes back sooner, and so do that principle's cards. Mock exams and section tests are practice: they run on questions you have met, at the exam's length and dealt at the exam's weights or one domain at a time, with no scaled score; rehearsals are the measurement: they run only on unseen questions held back for them, timed and weighted like the exam, and are the number the plan asks you to trust.

Start with the baseline quiz

Sign up today and the first thing Spaced asks of you is a baseline quiz: a short set of questions in exactly this shape, marked nowhere until the end, that finds the gaps you already have and builds your plan from them. Nothing is taught before it, so the score is honest, and your first evening of learning starts the same night. Domain 1 is free in full. Take the baseline quiz and start learning.

The exam's format here follows the CCAR-F Exam Guide v1.0 (July 2026); the pass mark is the one the Partner Academy FAQ states for all four certifications as of 2026.