I said the certificate was a receipt. Then I collected four of them in two months
Mihai Perdum
Author
11 min readSeptember 8, 2026
Key takeaways
Four ACPs in two months — ACP-120, ACP-420, ACP-520 and ACP-220 — and Atlassian Certified Expert at the end of it. I did not enjoy most of it.
The dump sites are the only broad source of exam-shaped questions, and their answers are frequently wrong. Harvesting them is fine. Believing them is not.
The method that worked: treat every harvested answer as a CLAIM, then confirm or overturn it against Atlassian's own documentation, keeping the citation.
Verifying an answer you already thought was right is the highest-value revision there is, because it is the only way to find the ones you were confidently wrong about.
I still think the certificate is a receipt. Four receipts did not change that. What changed is that I now know exactly which corners of Jira I had been avoiding.
The free practice exams are live for all five ACPs — original questions, a verbatim documentation citation behind every answer, per-option explanations, and full-length papers with the real clock.
A few days ago I wrote that a certification is a receipt, and that the thing with actual value is the syllabus — being walked across ground you would never have chosen to walk across yourself.
I would like to report that I have now collected four receipts.
Four ACPs, two months, and an Atlassian Certified Expert at the end of it: Jira Administration for Cloud (ACP-120), Managing Jira Service Projects for Cloud (ACP-420), Cloud Organization Administration (ACP-520), and — today — Confluence Administration for Cloud (ACP-220), which is the one that closed it out.
It was a gruelling march and I would not describe most of it as enjoyable. But it is done, and the interesting part is not the badge. It is that the method I used to get there was a gamble, and it is the opposite of what most people do.
The problem with how you are supposed to study
The official preparation material is fine and it is not enough. It tells you what the product does. The exam asks what the product does in a specific situation you have not been in, with four answers that are all things a reasonable person might say, three of which are wrong for a reason you have to actually know.
So everyone ends up at the dump sites. Scraped question banks, uncertain provenance, no attribution, and answers supplied by whoever typed them in. That is the only place on the internet with a large volume of exam-shaped questions, and pretending otherwise is not useful to anyone revising this weekend.
The problem is not that using them feels grubby. The problem is that they are frequently wrong, and wrong in the worst possible way: confidently, with a stated letter, and no indication of which entries you should doubt. Revise against that and you are memorising somebody else's mistakes with perfect fidelity.
The gamble
So the gamble was this: gather all of it. Every dump site I could find, everything they had, deliberately and without squeamishness.
And then believe none of it.
Every harvested question was treated as two separate things: a question, which is genuinely useful, and an answer, which is an unverified claim by an anonymous stranger. The question shape is the valuable part — the phrasing, the distractors, the specific corners the exam actually probes. The letter next to it is worth nothing until something authoritative says so.
So every answer got checked against Atlassian's own documentation. Confirm it, or overturn it, and keep the citation either way.
That is slower than reading a dump. It is enormously slower. It also turns a bank of uncertain claims into something you can actually revise against, and it does one more thing that I did not anticipate.
What verification actually taught me
Here is what I expected: the checking would catch a handful of wrong answers, I would correct them, and the rest of the value would come from volume.
Here is what happened: the questions I got wrong while verifying were not the ones I was unsure about.
When you are unsure, you look it up anyway. Uncertainty is self-correcting — it has a built-in trigger. The dangerous ones are the questions where the dump says C, and you read C, and you think yes, obviously C, and move on. Confidence is what stops you checking. So the only way to find those is a process that checks everything, including the answers you already agree with.
That is where the real gaps were. Not in the exotic corners. In ordinary behaviour I had been confidently wrong about for years, in ways that had never happened to bite me on a client instance — which is not the same as being right, it is being lucky and calling it experience.
It is the same rule I apply to my own work and had somehow never applied to my own knowledge: a green number is not a passing test. Feeling certain is a proxy metric. It correlates with being right often enough to be useless as a signal, and the entire value of the exercise was in the cases where it decoupled. It is the same reason a JQL count of zero can mean a field is missing rather than empty — a confident answer and a verified one look identical from the inside, and only one of them has been checked.
Four exams' worth of that is a strange experience. It is not learning new things so much as being repeatedly shown that a thing you already "knew" has a footnote you never read.
What the dump sites actually get wrong
It is worth being specific, because "the answers are sometimes wrong" is not actionable and the failure has a shape.
They are old. A great deal of what circulates was written against Server or Data Center. Those answers are not obviously wrong — that is the whole difficulty. They describe a real Jira, correctly, in detail, with confidence. Just not the one the Cloud exam is asking about. A plausible answer from the wrong product is far more dangerous than an implausible one, because nothing in you objects to it.
They are unattributed. No source, no date, no way to tell a question transcribed by somebody who passed from one invented by somebody who did not. Two entries sitting next to each other, identical in presentation, one accurate and one fabricated.
And they disagree with each other. The same question appears across several sites with different stated answers, which is at least honest about the situation. The temptation is to take the majority. Do not — I found cases where the majority was wrong, because the sites copy from each other and a popular error propagates faster than a correct answer nobody bothered to republish. Frequency is not evidence. It is the same claim repeated.
None of that is an argument for avoiding them. It is an argument for the layer on top: the question is the asset, the answer is a claim.
Why four, and why two months
The honest answer to "why four" is that ACE requires it, and once I had decided the syllabus was the valuable part, doing one and stopping made no sense. The syllabuses do not overlap as much as you would hope. Each one walks a different surface, which is precisely why doing several of them found more than doing one.
Two months was not a plan, it was what it took. I do not recommend the pace. The compression is real and it does not leave much room for the thing that makes the work worthwhile, which is time to be annoyed about something you got wrong and go and understand why the correct answer is correct.
If you have the option, spread it out. If you do not — and I did not, because there was a window and I took it — the thing that makes a compressed run survivable is the verification loop, because it is the only part that does not degrade when you are tired. You cannot fake it and you cannot skim it. It either has a citation or it does not.
What I still think about certificates
Nothing has changed. The certificate is still a receipt.
Nobody should hire me because I hold four of them, and I would not hire on that basis either. If somebody tells you they are an Atlassian Certified Expert, you have learned that they sat several exams and passed them. You have not learned whether they can run your migration.
What I will say, and what is genuinely different from a week ago, is narrower and more useful: I now know which corners of Jira I had been avoiding. Not because the exam taught them to me, but because the syllabus refused to let me skip them and the verification loop refused to let me be confidently wrong about them.
That is worth two months. The badge is not worth two months. Those are different claims and the marketing around certification collapses them constantly.
They are built the way the two months taught me to build them.
Every question is written from scratch against Atlassian's published documentation. Nothing is harvested, nothing is a reproduction of real exam content — that breaches Atlassian's agreement and, as established above, would only teach you somebody else's mistakes.
Every answer carries the verbatim documentation sentence that decides it, plus a per-option explanation of why each of the other choices is wrong. This is the part that matters and it is the part every dump site omits. A tool that gives you a letter trains you to recognise questions. A tool that shows you why the other three fail trains you to understand the product, which is the only version that survives contact with a real instance on a Tuesday afternoon.
And a full paper is exactly as long as the real one. Each certification publishes its own length and each paper matches it — 45 questions for ACP-520, 60 for ACP-120, 61 for ACP-220, 63 for ACP-620, 75 for ACP-420 — with the real clock running. Practising 20 questions tells you nothing about the last twenty of a seventy-five-question paper, which is precisely where I found out what my concentration actually does.
There is no account and no password. You give an email address, we send an access key, and the key unlocks the exams in your browser. That is the only thing asked in exchange for them being free, and you can unsubscribe whenever you like.
If you are grinding through this yourself right now: harvest whatever you like, and then go and check every single answer, including the ones you are sure about. Especially those. That is the whole method, it is not clever, and it is the only reason the two months worked.
