Free lesson · Claude Certified Architect - Foundations (CCA-F)

Exam Format, Scaled Scoring, Per-Domain Reporting, and What 720 Means

Easy12 min readccar-fexam-formatscaled-scorepacingscoring

The Most Misunderstood Number in the Program

A candidate scores 720 and passes. Another scores 715 and fails. The second candidate answers 44 of 60 questions correctly. The first answers 43.

That is not an error, and it is not unfair. It is what a scaled score is. Almost every misconception about CCAR-F scoring traces back to reading 720 as a percentage, and almost every practice tool that promises to predict your score is exploiting that misconception.

This topic covers the mechanical facts of the exam and then spends most of its length on scoring, because the scoring model determines how you should interpret every practice result between now and exam day.

The Format Facts

Two of these deserve immediate attention.

Item types are mixed, and the stem tells you which. Some items have one correct answer among four options. Others are multiple-response and require selecting two or three. The item explicitly states how many responses to select. This is a gift, and candidates routinely waste it - a multiple-response item that asks for two answers has told you that exactly two options are defensible and two are not, which is powerful elimination information before you have evaluated anything on its merits.

All 60 questions are scenario-bound. There are no standalone trivia items sprinkled between the scenarios. The next topic covers why this is structurally guaranteed rather than merely typical.

Tip: Read the response count first, before the options and even before the full stem. On a multiple-response item asking for two of five, your job changes from "find the right answer" to "find the two that survive." Those are different cognitive tasks and the second is faster.

Pacing: The Two-Minute Budget

Sixty questions in 120 minutes is exactly two minutes per question. That average is deceptive because scenario-based items are not uniform in cost.

The first question attached to a new scenario is expensive. You are absorbing a narrative - a business context, a system description, a set of constraints - before you can answer anything. Questions four through fifteen inside the same scenario are cheap, because the narrative is already loaded and you are only processing the new stem.

graph LR
    A[Scenario 1<br/>~30 min] --> B[Scenario 2<br/>~30 min]
    B --> C[Scenario 3<br/>~30 min]
    C --> D[Scenario 4<br/>~30 min]
    A --> A1["Q1: absorb narrative<br/>4-5 min"]
    A --> A2["Q2-15: apply context<br/>~1.7 min each"]

This diagram shows the exam's time budget as four sequential scenario blocks of roughly thirty minutes each, totalling one hundred twenty minutes. Branching from the first block are two cost profiles: the opening question of a scenario is expensive at four to five minutes because it requires absorbing the narrative, while the remaining fourteen questions in that block cost roughly one point seven minutes each because the context is already loaded.

The practical consequence is that you should budget roughly thirty minutes per scenario block rather than two minutes per question. If you are twelve questions into a block with eight minutes left in its budget, you are on pace. If you are four questions in with eight minutes left, you over-invested in the narrative and need to move.

Warning: The single most common pacing failure is re-reading the scenario narrative for every question inside its block. Read it thoroughly once, extract the constraints into a written ledger, and then work from the ledger. Re-reading a four-paragraph narrative fourteen times consumes twenty minutes you do not have.

What a Scaled Score Actually Is

A raw score is how many questions you answered correctly. A scaled score is a transformation of that raw score onto a fixed reporting range - here, 100 to 1,000 - that accounts for the difficulty of the specific items you saw.

The purpose is fairness across exam forms. If the certification uses multiple item pools and one form happens to draw harder items, candidates who sat that form should not be penalised for randomness in form assembly. Scaling equates across forms so that 720 represents the same demonstrated competence regardless of which items you received.

graph TD
    A[Your responses<br/>60 items] --> B[Raw score<br/>Items correct]
    B --> C[Item-response weighting<br/>Difficulty of items you saw]
    C --> D[Scaled score<br/>100 to 1000]
    D --> E{At or above 720?}
    E -->|Yes| F[Pass]
    E -->|No| G[Fail]
    D --> H[Per-domain breakdown<br/>Strong or weak by domain]

This diagram traces the scoring pipeline from left to right. Your sixty responses produce a raw score of items correct. That raw score passes through an item-response weighting step that accounts for the difficulty of the specific items you received. The result is a scaled score between one hundred and one thousand. A decision point checks whether the scaled score reaches seven hundred twenty, branching to pass or fail. Separately, the scaled score also produces a per-domain breakdown showing where you were strong or weak.

The weighting model Anthropic uses is not published. That single fact has a large downstream consequence.

Important: Because the raw-to-scaled conversion formula is not public, no practice platform can tell you what scaled score you would earn. Any tool that displays a predicted number between 100 and 1,000 is generating it from an invented formula. Treat that as a signal about the tool's honesty, not as information about your readiness.

Why 720 Is Not 72 Percent

Three distinct reasons, and each one changes how you should read practice results.

The scale does not start at zero. The reporting range runs from 100 to 1,000, not 0 to 1,000. A score of 100 is the floor of the reporting scale, not an indication that you answered nothing correctly. Mapping 720 onto a 0-to-100 percentage is arithmetically meaningless.

Items are not worth equal amounts. Under item-response weighting, a correct answer on a discriminating item - one that reliably separates competent candidates from less competent ones - carries more information than a correct answer on an item almost everyone gets right. Two candidates with identical raw scores can land on different scaled scores because of which items they got right.

The scale is designed to be stable across forms. The whole point of scaling is that 720 means the same thing on every form. A fixed percentage of items cannot deliver that, because the item mix varies.

The practical translation: do not target a percentage. Target consistent competence, measured the only way you actually can measure it.

Per-Domain Reporting and How to Use It

The score report includes a per-domain breakdown showing where you were strong and where you were weak. This matters most for candidates who fail, because it converts a failure into a targeted study plan rather than a vague instruction to try harder.

It also has a strategic implication that is easy to miss. Because the report breaks out five domains, a candidate can fail while being genuinely strong in three of them. A domain you neglect does not merely cost you its weight in points - it drags the whole result down while your strength elsewhere is invisible in the pass or fail line.

Sixteen items in D1 against nine in D5. Neglecting D5 entirely costs you nine items and cannot be recovered by excelling in D1, because D1 only has sixteen to give.

Interview: A useful way to read the weights: D1 is nearly twice D5, but D5 is not optional. The exam is built so that competence must be broad. Scenario blocks mix domains - a customer support agent scenario will draw on agentic architecture, tool design, and context reliability inside the same narrative. There is no clean lane where you can hide a weak domain.

What Honest Practice Reporting Looks Like

Given that no one can predict your scaled score, what should a practice platform report? Three things, all of which are real signal:

Raw score. How many items you answered correctly. Unglamorous and directly comparable across sessions.

Percentage by domain. Your accuracy within each of the five domains, so you can see the distribution rather than the average.

Per-objective accuracy. Your accuracy against each of the 29 task objectives. This is the highest-resolution signal available and the one that actually drives remediation.

The reason per-objective matters more than per-domain is that a domain contains four to eight objectives, and strong performance on six of eight can conceal a total blank on the other two.

Example: A concrete case. A candidate finishes a mock at 78 percent overall and reads that as comfortable.

Domain	Accuracy	Reading
D1	85%	Strong
D2	82%	Strong
D3	88%	Strong
D4	79%	Adequate
D5	52%	Failing

The 78 percent average is real, and it is hiding a domain at 52 percent. On a real form, D5's nine items would be contributing roughly five correct answers instead of seven or eight. The overall number told this candidate to book the exam. The breakdown tells them to spend a week on context management first.

Retakes, Validity, and Renewal

The credential is valid for twelve months, with a free on-time renewal. That is a shorter validity window than most cloud certifications and reflects how quickly the Claude ecosystem moves.

Plan for the renewal rather than being surprised by it. The material that will have changed most by renewal time is the tooling - Claude Code configuration and MCP mechanics - rather than the architectural reasoning, which is more durable. When you renew, that is where to re-verify first.

Warning: Retake policies, waiting periods between attempts, and renewal requirements are program policy rather than blueprint content, and they change. Confirm current policy on Anthropic's official certification page before you assume anything about a second attempt.

Setting Your Readiness Bar

Combining everything above produces one defensible bar, and it is not a scaled-score target:

Consistent accuracy at or above 80 percent in every one of the five domains, on timed, blueprint-weighted practice.

Every word in that sentence is doing work. Consistent means across multiple sessions, not one good result. Every one of the five domains means no averaging. Timed means under the real pacing budget, because untimed practice hides the decision fatigue that shows up at question forty-five. Blueprint-weighted means the practice mix mirrors the real domain proportions rather than giving equal airtime to a 27 percent domain and a 15 percent domain.

This is a reasonable bar, not a guarantee. No practice result is a guarantee, because the conversion is not public. But it is honest signal, and honest signal is what you can act on.

Exercise: Starter. Write out the five domains and their weights from memory, then convert each to an approximate item count on a 60-item form. Being able to do this instantly makes it much easier to judge whether a weak area is worth remediating.

Intermediate. Time yourself reading a four-paragraph technical scenario and extracting its constraints into a list. If it takes longer than four minutes, practise until it does not - this is the single most compressible cost on exam day.

Advanced. Take any practice set you have already completed and recompute your results by domain rather than overall. Then identify your weakest objective within your weakest domain. If your practice tool does not report at objective resolution, that is a reason to change tools, not a reason to skip the exercise.

Quiz: Quick check: A candidate scores 78 percent on a full-length blueprint-weighted mock. Which conclusion is best supported?

A) They are ready - 78 percent is comfortably above 720 B) They will score approximately 780 on the real exam C) The overall percentage is insufficient information; the per-domain and per-objective breakdown determines readiness D) They need 72 percent to pass, so they have a six-point margin

Answer: C - the overall percentage can conceal a failing domain, and the raw-to-scaled conversion is not public, which rules out B and the percentage-mapping logic in A and D.

Back to Claude Certified Architect - Foundations (CCA-F)