CCAO-F cheat sheet: the decision rules for all seven domains on one page

This CCAO-F cheat sheet compresses the 60-item, 120-minute Claude Certified Associate Foundations exam into decision rules by domain. Seven domains are weighted from 21 percent for Output Evaluation and Validation down to 10 percent for Troubleshooting and Optimization; 720 of 1,000 passes; the fee is $99. Each domain below gets the CCAO-F decision rules that settle an item when two options both look reasonable, then the traps, the terms and a final-week checklist.

How to use this CCAO-F cheat sheet

Read the weights first, because they decide where revision time goes. Output Evaluation and Validation alone is about 13 of 60 items, more than twice Troubleshooting and Optimization and nearly double either Product and Model Selection or Configuration and Knowledge Management. Then work through the domain sections in order of weight, not in blueprint order, and use each rule as a test on your own practice misses rather than as text to memorise.

Checked against the official sources on 17 September 2026: the CCAO-F Exam Guide v1.0 (July 2026), the Anthropic Certification Exam Policy dated 25 June 2026, and the partner certification FAQ. Where this page states a fee, a count or a policy, that is where it comes from.

The item column is a whole-number reading of the published percentages, the same allocation the free timed mock at /practice/ccao-f uses. The guide promises approximate weights, not exact counts, so treat 13 as the shape of the paper rather than a guarantee. The guide lists 30 objectives across the seven domains, which on 60 items is two items per objective on average; a gap in any single objective is likely to cost more than one mark. The generic Claude exam cheat sheet covers the twelve judgements that repeat across all four tracks; this page is the CCAO-F-specific layer beneath it, and the Claude certification study guide supplies the repair loop the checklist at the end assumes.

Domain 1: Prompting and Task Execution (14%)

About eight items. The four objectives ask you to create prompts for business and technical tasks, decompose complex requests, iterate to improve output and adapt the approach to the task type. The items are rarely about wording. They are about whether the prompt removes ambiguity, supplies the context the task needs and defines what a good output is, without asking Claude to invent anything quietly. The deeper treatment is in prompting and task execution on CCAO-F.

1. 1.1 Specify the output before the task. An option that names the audience, the format, the length and what counts as done beats one that adds adjectives such as thorough or accurate. Fluent output that fits no downstream use is the failure this rule catches.
2. 1.2 Split the work where a mistake would spread. Decomposition on this paper means a checkpoint between pulling facts out of a document and reasoning over them, not a longer prompt. If the facts are wrong, everything built on them is wrong, so they are checked first.
3. 1.3 Change one thing, then rerun the same cases. When an output disappoints, the option that rewrites the whole prompt is the distractor. The option that adjusts a single instruction and reruns on identical inputs is the one that can tell you what fixed it.
4. 1.4 Match the recipe to the task type. Analysis separates evidence from inference. Research needs a scope and a source standard. Drafting preserves the facts while changing the form. Brainstorming widens before it narrows. Applying the drafting recipe to an analysis is wrong however well written.

Tie-breaker. Which option leaves Claude with the least to guess about the output and the least chance to carry a wrong fact forward?

Domain 2: Output Evaluation and Validation (21%)

About thirteen items, the largest block on the paper. Six objectives: evaluate accuracy and completeness, identify hallucinations and bias, apply fact-checking, decide when human review is required, adapt outputs for the audience, and choose the output format. The official sample item for this domain is instructive: a confident summary citing a specific subsection, bound for a compliance team, and the key verifies the citation against the regulation before anything is sent. Every distractor offers a signal of quality in place of evidence.

1. 2.1 Precision is where fabrication hides. A specific subsection number, a quoted figure, a URL or a named study is the part most likely to be invented and the part most likely to be trusted. Check it against the source that governs it before anything else.
2. 2.2 Verify at the controlling source, not a convenient one. For a regulation, the regulation text. For a figure, the system that records it. For a policy, the policy owner. Asking Claude to rate its own confidence, or asking a second time, is not verification on this exam.
3. 2.3 Review must be qualified and placed before the harm. A colleague glancing at a compliance summary is not human review. The reviewer needs the authority and the domain knowledge to catch the error, and the review has to happen before the output reaches its audience or triggers an action.
4. 2.4 Completeness is coverage of the requirement, not length. A long answer can miss half the brief. Check each requested element is present and distinguish not found from not applicable. An option that praises the detail of an output without checking coverage is a distractor.
5. 2.5 Choose the format by what happens next. Inline for a conversational reply, an artifact for a document or code that will be edited and reused, structured data when a spreadsheet or another system will consume it. Novelty is never the reason to pick a format.

Tie-breaker. If this specific claim turned out to be wrong, what would it cost, and has anyone checked it against the source that actually decides?

Domain 3: Product and Model Selection (12%)

About seven items. Four objectives: select product features, differentiate the model tiers, align model choice with cost, speed and quality, and manage context and memory. The official sample here is a high-volume batch of short customer-reply drafts where speed and cost outrank deep reasoning, and the key is the faster, lower-cost model. The distractors are the most capable model for everything, turning features off to save money, and changing platform. The feature side of this domain is covered in Projects, artifacts and connectors on CCAO-F.

1. 3.1 Chat, Project, research, artifact: pick by persistence and scope. A one-off question is chat. Recurring work with shared instructions and knowledge is a Project. A broad, current, multi-source question is research mode. A substantial work product that will be edited over time is an artifact.
2. 3.2 Model tiers are capability tiers, and the lightest sufficient one wins. Haiku for fast, low-cost, high-volume and straightforward work. Sonnet as the general default. Opus for the hardest reasoning. Always using the most capable model is a distractor, and so is always using the cheapest.
3. 3.3 Cost is the whole workflow, not the model call. Model spend plus review time plus rework plus escalation is the number that matters. An option that saves on the model and doubles the checking is not the cheaper option, and the items are written to test exactly that.
4. 3.4 Context is a working set, not an archive. Restart when the topic changes or the thread has picked up conflicting instructions. Summarise validated state to carry it forward. Persist stable rules and sources in a Project. Pasting the entire history back in is the distractor.

Tie-breaker. What is the lightest feature and model combination that clears the stated quality bar at an acceptable cost and speed?

Domain 4: Workflow Integration and Solution Design (16%)

About ten items. Five objectives: analyse requirements, use Claude for research and planning, support solution design and iteration, integrate Claude into existing workflows, and communicate value and limitations to stakeholders. These items describe a team, a process and a proposal, then ask what an Associate should do next. The guide's own audience statement is the frame: Associates translate business objectives into effective AI interactions and escalate technical or enterprise-scale design to the Claude Certified Developer and Claude Certified Architect credentials. The full domain post is workflow integration on CCAO-F.

1. 4.1 Start from the process and the outcome, not the feature. The key describes the current workflow, names the bottleneck and states a testable requirement: who, when, with what inputs, to what acceptance criterion. Helpful and accurate is not a requirement.
2. 4.2 Classify each step: automate, augment or keep human. Stable inputs, detectable errors and reversible results can be automated. Judgement that an expert can check is augmented. Rights, safety, legal exposure or irreversible consequences stay with a person.
3. 4.3 Place Claude at the bottleneck and leave the record where it lives. Inserting Claude at every handoff is the distractor. The key augments or redesigns the slow step and keeps the system of record, its ownership and its audit trail intact.
4. 4.4 Pilot with a baseline and a stop rule before rollout. Compare against the current process on representative and edge cases, and decide in advance what result scales and what result stops. Measuring adoption by the number of conversations is not measuring value.
5. 4.5 Communicate limits alongside gains. The stakeholder objective rewards a balanced statement: what improves, what still needs a person, what the risks are. An option that promises eliminated errors or zero review is wrong on its face.

Tie-breaker. Given how easily an error at this step would be noticed and how much it would cost, should the step be automated, augmented or left with a person?

Domain 5: Configuration and Knowledge Management (12%)

About seven items. Four objectives: configure Projects with instructions and knowledge, manage uploads and connectors, write system-level instructions, and maintain and update configurations over time. The recurring test is layering: which information belongs in the standing instructions, which in the knowledge base and which in the individual request. Get the layer wrong and every later conversation inherits the mistake.

1. 5.1 Three homes for information, never blended. Project instructions hold behaviour that should apply every time. Project knowledge holds durable, approved sources. The conversation prompt holds one-off detail. A one-off in the instructions pollutes every later conversation.
2. 5.2 Instructions state precedence and the empty case. Good system-level instructions say which source wins when two disagree and what to do when the answer is not in the sources: say so and ask. Vague ideals such as be helpful and accurate are distractors.
3. 5.3 Knowledge has an owner, a date and one current version. Replace a superseded document rather than uploading the new version beside the old one, because Claude cannot know which is current. Label approval status and effective date where the source itself does not.
4. 5.4 Uploads are snapshots, connectors are live and permissioned. An uploaded file is frozen at upload. A Google Drive or Gmail connector reads live under the permissions of the account. Connecting a source does not authorise every use of what it contains, and the minimum connection is the right one.

Tie-breaker. Is this a rule that should hold every time, a reference that should stay current, or a detail for this one request, and which of them wins if they disagree?

Domain 6: Governance, Risk, and Responsible Use (15%)

About nine items. Four objectives: identify appropriate and inappropriate use cases, apply data sensitivity and privacy considerations, follow organisational AI policy, and understand the ethical implications of use. The official sample is a spreadsheet of customer names and account numbers that a project manager wants to analyse under a policy that restricts regulated personal data. The key removes or anonymises the identifiers first. Uploading as is violates the policy, telling Claude not to retain the file is not a control, and skipping the analysis throws away value a compliant route would have delivered. The domain post is governance and responsible use on CCAO-F.

1. 6.1 Feasibility is not permission. That Claude can read the file is never the question. Who authorised this data, for this purpose, at this level of impact, is the question. If the answer is unclear, the key escalates rather than proceeds.
2. 6.2 Minimise before you upload. Strip or pseudonymise names, account numbers and other identifiers when the task does not need them. Pseudonymisation reduces obligations without removing them if re-identification remains possible.
3. 6.3 Access for one purpose is not access for another. A connector approved for scheduling is not approved for performance analysis. Apply least privilege to accounts, Projects and connectors, and treat a new purpose as a new authorisation.
4. 6.4 Regulated outputs need a qualified reviewer, and a disclaimer is not a control. Legal, medical, financial, employment and eligibility decisions need a person with the standing to be accountable. Telling Claude not to retain the data is an instruction, not a control, and it satisfies no policy.
5. 6.5 Policy beats the deadline, and abandoning the task is over-correction. The key follows organisational policy over convenience and documents any exception formally. But when a compliant route exists, such as anonymising first, skipping the task entirely is also wrong. The exam rewards enabling the work safely.

Tie-breaker. Does policy or law authorise this data, for this purpose, at this level of impact? If it is unclear, the answer is to escalate, not to proceed.

Domain 7: Troubleshooting and Optimization (10%)

About six items. Three objectives: find and fix the cause when a prompt underperforms or an output falls short, adjust the approach on feedback, and optimise workflows for efficiency. The items present a symptom and four fixes, and the distractors are the fixes that act before they diagnose. It is the smallest domain, but it is also the one where a single habit, locating the failure layer before editing, converts directly into marks.

1. 7.1 Locate the layer before you edit anything. Objective, instruction, input, knowledge source, context, model or feature, output format, review. A stale or wrong citation is a knowledge and retrieval problem, and rewording the prompt will not fix it.
2. 7.2 Reproduce, change one variable, retest on old and new cases. Build the smallest case that shows the failure, fix one thing, and rerun the cases that used to work as well as the one that failed. Skipping the regression step is how a fix ships a second bug.
3. 7.3 Optimise the whole workflow, not the token count. Time to a validated result, review burden and rework are the costs that matter. Route simple work to a lighter model, keep stable instructions in a Project instead of repeating them, and target review where errors are hardest to detect.

Tie-breaker. Which layer actually produced this failure? Answer that before you change anything.

The CCAO-F traps: wrong-answer patterns and why they tempt

Distractors on this exam are not silly. Each one describes something a sensible colleague might do, which is why it is on the paper. The table names twelve that recur, the reason each feels right, and the reason it loses. Half of them come straight from the three official sample rationales; the rest follow the same construction across the other domains.

CCAO-F quick reference: the terminology

Fifteen terms the items use without defining. None of them is exotic, but each carries a specific meaning on the paper that differs slightly from everyday use, and several distractors depend on the everyday reading.

Final-week checklist and CCAO-F exam tips

Six steps, in order. They assume you have already done the studying and are now converting it into marks, which is a different activity. The two-week CCAO-F study plan is the version of this that starts a week earlier, and the CCAO-F certification page carries the published objective list, domain by domain.

Two policy facts worth carrying in: a failed attempt can be retaken after 14 days, then 30 after a second failure and 90 after a third, with four attempts per rolling twelve months and the full fee each time (your partner-tier discount still applies), and the credential is valid for 12 months with a free, non-proctored renewal if you complete it on time. Registration requires a work email on a recognised Claude Partner Network company domain, and Select, Preferred and Global Premier partners pay half the fee, with Global Premier paying nothing through 31 December 2026. Confirm all of it against the guide and the Pearson VUE Anthropic page before booking. Our guide to Claude certification practice exams sets out what official material exists to practise on, which is less than most candidates expect.

Key takeaways

• Output Evaluation and Validation is a fifth of the paper. Thirteen of 60 items, and almost all of them turn on verifying at the source or routing to a qualified reviewer.
• Every domain has a tie-breaker question. Learn the seven, and most two-option stand-offs resolve without further knowledge.
• The distractors are made of quality signals. Confidence, detail, formality and a human glance are what wrong answers are built from.
• Feasibility is never permission. Governance items ask who authorised the data for the purpose, and the compliant route usually still delivers the task.
• Diagnose the layer before editing the prompt. Troubleshooting is the smallest domain and the one where a single habit pays directly.
• No reference material in the room. This page is for the final week; the exam requires a clear workspace and photo ID that matches the registration.