Here is the whole thing, heaviest first.
| D2 | Output Evaluation and Validation | 21% |
| D4 | Workflow Integration and Solution Design | 16% |
| D6 | Governance, Risk, and Responsible Use | 15% |
| D1 | Prompting and Task Execution | 14% |
| D3 | Product and Model Selection | 12% |
| D5 | Configuration and Knowledge Management | 12% |
| D7 | Troubleshooting and Optimization | 10% |
Output Evaluation and Validation · 21%
The largest domain, and the one that defines the credential. Judging whether an output is accurate and complete, spotting hallucinations and bias, fact-checking, deciding when a human has to review, editing for an audience, and choosing an output format. The recurring judgment: fluent and confident is not the same as verified, and the credited answer checks the specific claim against the source rather than asking the model how sure it is.
Workflow Integration and Solution Design · 16%
Analysing requirements, using Claude for research and planning, integrating it into an existing workflow, and communicating both its value and its limits to stakeholders. Part of this domain is knowing the edge of your own scope: genuinely technical integration work escalates to an Architect or Developer rather than being improvised.
Governance, Risk, and Responsible Use · 15%
Appropriate and inappropriate use cases, data sensitivity and privacy, organizational AI policy, and ethics. The trap in this domain is assuming the most cautious answer always wins. It does not. Refusing permitted work, escalating what you should simply handle, or stripping information the task needs are all wrong answers. The credited response is proportionate.
Prompting and Task Execution · 14%
Writing effective prompts, decomposing complex requests, iterating when output falls short, and adapting strategy to the task type. The most testable idea: when output fails on form rather than facts, showing beats telling — a couple of worked examples outperform another paragraph of instructions.
Product and Model Selection · 12%
Choosing between chat, Projects, Artifacts and research mode, distinguishing the model tiers, and matching a choice to cost, speed and quality. Also managing context limits as a user: when to restart, summarize, or persist something into a Project. The cheapest tier that clears the bar wins.
Configuration and Knowledge Management · 12%
Configuring Projects with instructions and knowledge sources, managing uploads and connectors, and keeping all of it current. Knowledge that has gone stale is worse than knowledge that is absent, because people trust it.
Troubleshooting and Optimization · 10%
Diagnosing why a prompt or output underperforms, adjusting on results, and optimizing a workflow. Diagnose before you change: the credited answer names why the output is weak rather than reaching for a bigger model or a longer prompt. Regenerating and hoping is the most common wrong instinct.
What the weights mean for a study plan
Output Evaluation, Workflow Integration and Governance are 52% of the exam between them. All three are judgment domains rather than knowledge domains, which is the single most useful thing to know before you start: this exam rewards deciding well far more than it rewards recall.
Practice forms should carry the same proportions. Every form in our bank matches the published weights to within half a question.