Skip to content
CertificationsHow it worksReadinessPricingSign inStart practising free
How it works

Where the questions come from

The thing that decides whether a preparation product is any good is whether the questions are any good. This is how ours are made, and what we refuse to do.

The rules every question is written to

  1. 01Write ORIGINAL questions from the published exam guide and the provider’s public documentation. Never reproduce, paraphrase or reconstruct a question from a real exam.
  2. 02Exactly one defensible answer (or, for multiple-response, exactly one defensible SET). If two options could be argued, the stem is missing a constraint — add the constraint.
  3. 03Every distractor must be the answer somebody would give for a specific, nameable reason. No filler, no obviously wrong option, no joke options.
  4. 04Every option needs its own rationale, the correct one included.
  5. 05Cite documentation the candidate can go and read. Provider documentation only — never a blog, never a forum, never another preparation site.
  6. 06Test judgement, not trivia. No questions about pricing figures, service limits that change, or console menu positions.
  7. 07British English. Metric units. No trademarked marketing language.

These are not marketing copy. They are the literal instructions in the generation prompt and the checklist the review scores against — one document, read by the model, by the reviewer and by whoever is writing by hand.

Why we will never be a brain dump

A brain dump is a set of real exam questions, recalled and republished by somebody who signed an agreement not to. Using one puts your certification at risk — providers do revoke them — and it teaches you the answer to one form of one exam rather than the reasoning that survives the next revision.

It is also a commercial dead end. A brain-dump site competes on how recently somebody broke an agreement, and is one enforcement letter from nothing. SupportExams competes on whether the questions teach you something, which is a thing that can be built on.

From blueprint to published

  1. 01Import the blueprintA human reads the provider’s own exam guide and records the domains, the weightings exactly as published, the question count, the time and the passing standard — with the URL and the date. Anything the provider does not publish is left blank.
  2. 02Write against the objectivesQuestions are written for one domain and one topic at a time, with the existing bank in front of the author so the same scenario is not written twice.
  3. 03Screen for duplicatesBefore anything is stored, every candidate question is compared against the bank and against the rest of its own batch. Near-duplicates are discarded rather than reviewed.
  4. 04Automated technical reviewA second, stronger model scores the question on seven axes and returns a verdict and flags. It cannot publish anything.
  5. 05Human reviewA person reads it, edits it or rejects it. Questions carrying a blocking flag — a factual error, two correct answers, a suspected brain dump — cannot be published at all until they are fixed.
  6. 06PublishOnly then does it reach a candidate. Editing a published question writes a revision, bumps its version and sends it back for review.

What the review checks

accuracy

Is every technical statement correct, and correct TODAY? Flag anything that was true of an older service generation.

alignment

Does this test something in the published exam guide for this domain, at a level the exam would ask?

clarity

Is the stem unambiguous, and is the question actually asked? Would a competent candidate know what is being asked?

distractors

Is every wrong option plausible and diagnostic — the answer somebody would give for a nameable reason?

explanation quality

Does the explanation teach the principle, rather than restating the answer?

ambiguity risk

Could more than one option be defended? Score 100 for "no", 0 for "yes, easily". THIS IS THE MOST IMPORTANT AXIS.

Accuracy and ambiguity count double in the overall score, because those are the two where being wrong costs a candidate a mark on an exam they paid for. The others cost them a slightly worse explanation.

What an explanation looks like

Being told the answer is B teaches you nothing about why you picked C. Every option carries its own reasoning, and the one you chose is pulled to the top.

SAA-C03Question 12 of 20
Correct

A company runs a reporting application on a single EC2 instance. The reports are generated overnight and the application is idle for the rest of the day. Interruptions are acceptable provided the job finishes before 07:00. Which purchase option gives the lowest cost?

A 1-year All Upfront Reserved Instance
Spot Instances managed by an Auto Scaling groupCorrect answer
On-Demand Instances with a scheduled start
Why

The workload is interruption-tolerant and runs in a window with slack in it, which is exactly the shape Spot is priced for — up to 90% below On-Demand. Reserved Instances and Savings Plans bill for a whole term regardless of use, so paying for 24 hours a day to run a job for three is more expensive than On-Demand, not less.

Cost OptimisationCompute costSpot InstancesMedium

Independence

SupportExams is an independent certification preparation platform. It is not affiliated with, sponsored by, or endorsed by Amazon Web Services, Microsoft, Google, or any other certification provider. All trademarks are the property of their respective owners and are used for identification purposes only.

We name the exams so you can find the one you are sitting. We do not use anyone’s logo, and every certification page links to the provider’s own page for that exam — which is where you book it and which is always the authority over anything here.

AWS certification index ↗Azure certification index ↗Google Cloud certification index ↗

Questions people ask

Are these real exam questions?

No. Every question is original work written from the provider’s published exam guide and public documentation. Reproducing real exam questions breaches the provider’s non-disclosure agreement, teaches nothing transferable, and can cost a candidate the certification they paid for.

Are the questions written by AI?

Some are drafted by a model, and none are published by one. Every generated question goes through an automated technical review and then a person, who edits or rejects it before it can reach a candidate. The seeded bank was written by hand. The creation method is recorded on every question.

How do you keep exam outlines up to date?

Every exam version in the catalogue records the URL of the provider’s own guide and the date a human last checked it against it — and both are printed on the certification page. Where a provider has not published a figure, we leave it blank rather than guessing.

What happens when I report a bad question?

It goes to the top of the internal queue, above every coverage metric. A question two people report as ambiguous almost always is, and no automated review finds those as reliably as somebody who has just lost a mark to one.

Start practisingHow readiness is calculated