The exam guide publishes seven domains, 38 sub-objectives, and the weight of each. This series works through all of them in weight order, with the free practice form's 63 questions distributed across the sections they belong to.
Nothing here reproduces the official course. It is written to the published blueprint by someone who sat the exam and scored 965 out of 1000, which means it can also tell you where the published material and the actual paper diverge.
Where to start
Read in weight order if you have time and in this order if you do not. Domain 3 is the largest single block on the exam and domain 7 is about four questions.
| Guide | Weight | Questions | Sub-objectives |
|---|---|---|---|
| Integration | 19% | ~12 | 8 |
| Solution design and architecture | 17% | ~11 | 6 |
| Evaluation, testing and optimization | 16% | ~10 | 6 |
| Governance, safety and risk | 14% | ~9 | 5 |
| Stakeholder communication and lifecycle | 14% | ~9 | 5 |
| Models, prompting and context | 13% | ~8 | 5 |
| Developer productivity | 7% | ~4 | 3 |
Two domains are worth more attention than their weight suggests. Integration at 19% is where the most attractive wrong answers live, because the distractors are usually real improvements aimed at the wrong half of the system. Stakeholder communication at 14% gets skipped because it reads as soft skills, and it is worth twice what developer productivity is.
How to use the questions
Every section carries at least one question from the free practice form, with all four options kept and each one explained. Answer before reading the explanation. The wrong-answer reasoning is where the studying happens, which is why the options are all there rather than summarized.
When you want a full-length sitting rather than a section at a time, the free form runs all 63 questions untimed and gives you a per-domain breakdown at the end. Use that breakdown to decide which guide to reread.
Before you plan around a booking date
Certification is currently open only to people at Claude Partner Network organizations, and registration requires a partner company email address. Personal addresses are not accepted at sign-up. If you are independent, solve that before building a study schedule around a date. The exam facts page has the fees, retake windows and validity period.
What the real exam is like
Two things surprised me on the day, and neither is in any published description.
The questions are shorter than almost all practice material, including mine. Most were one or two sentences. Roughly every third or fourth question carried a longer scenario of about three sentences, and those were the exception. There was much less context to hold than I had trained myself for, which made the paper feel easier than my own practice exams even though the concepts were the same. If you have been grinding through dense scenario questions, the real thing will feel lighter.
Two formats catch people out. A yes/no matrix, where one question gives you five statements and you judge each independently. And drop-down matching, where you classify each item from a shared option set and the options can be reused, so do not assume a one-to-one mapping. There is a free drill for both shapes, because most platforms cannot render them and you should not meet them for the first time on exam day.
A lot of questions also turn on sequence: before you do X, should you do Y. The wrong answers there are usually things you would genuinely do, just later.
Not everything on the blueprint appears on your paper
My score report broke the exam down by sub-objective, and only 28 of the blueprint's 38 appeared on it at all. The other ten were simply not sampled. With 63 questions spread across 28 sub-objectives, that is roughly two questions each.
The ten absent from my form were: selecting a model on trade-offs, optimizing context windows, observability at scale, RAG pipeline chunking and indexing, retrieval strategies by data shape, progressive against monolithic context, defining evaluation metrics, A/B testing, implementing guardrails, and lifecycle phases.
Do not read that as a list to skip. It is one form, forms are sampled, and yours will be sampled differently. What it does tell you is that no single sitting can cover the blueprint, so a topic you never see is not evidence it is untested, and a topic you do see may be worth two marks rather than a section's worth.
What I got wrong
Two sub-objectives came back at 0%, and everything else at 100%: evaluating accuracy-latency trade-offs, and compliance with regulations. Two very different failures. The first was a habit problem, naming the benefit of a configuration without naming its cost. The second was a knowledge problem, where reasoning from general privacy instincts produces a confident wrong answer and there is no substitute for knowing what the regulation actually requires.
How I actually prepared
Read the official course material once, carefully. Do the four prerequisite courses, which were more useful than I expected. Then practice tests until your weak domains show up in the per-domain breakdown. I did about ten.
Sit it at a test center if you can. No room scan, no environment fiddling.
Two mechanical things worth using: strike through options you have ruled out, on every question that is not a gift, and flag anything you are unsure of so you can return to it. I flagged three questions I genuinely could not settle. I scored 965, so those were probably the ones I lost.
Why this exists
I wrote up my exam experience on Reddit and two people replied with the same problem: the official course material sits behind a partner login, and they could not get to it. One was offering to sponsor strangers through a partner org just so they could sit the exam.
The blueprint is public. The material teaching it is not. That gap is the whole reason these pages exist, and it is why everything here is written to the published blueprint rather than lifted from a course you may not be able to open.