How exam readiness is calculated
Percentage correct answers the wrong question. Ninety questions about IAM and none about cost optimisation scores 92% and is not ready. This is what we do instead, written down so you can argue with it.
Why not just percentage correct
It ignores what the exam is made of
Providers publish domain weightings, and they are not equal. On SAA-C03, Design Secure Architectures is 30% of the exam and Cost Optimisation is 20%. A practice score that treats a gap in the first like a gap in the second is measuring the wrong exam.
It ignores difficulty
Fifty recall questions answered correctly is not the same evidence as twenty multi-constraint scenarios, and a single number that cannot tell them apart will read high for the candidate who has only ever practised the easy ones.
It ignores when
A candidate who spent last week lost and this week turned it around is closer to ready than an average across both says. Recency has to count, or the number cannot fall either.
It ignores how little it knows
Three right answers in a domain is not 100%. It is barely tested, and a product that reports it as 100% is flattering somebody into booking an exam.
The five things that move it
- 01Correctness, weighted by difficultyA multi-constraint scenario is worth nearly three times a recall question. The weights are 1, 1.8 and 2.8 for the three difficulty levels.
- 02RecencyAttempts decay with a half-life of twenty attempts — one good practice session. Evidence older than ninety days is discounted further for staleness, floored at 45%, because cloud exams move and a domain you practised before a blueprint revision is weaker evidence than the count suggests.
- 03EvidenceEach domain's score is shrunk towards a neutral 50% in proportion to how little evidence stands behind it. Ten medium questions give roughly 60% confidence; thirty give roughly 78%. Confidence also rises with difficulty and falls as evidence ages, so thirty hard questions count for more than thirty easy ones. Below 35% the domain is described rather than scored.
- 04Blueprint coverageDomains are combined using the provider’s own published weights. A domain you have never practised contributes the neutral prior rather than being quietly left out of the average — because not having touched 20% of the exam is a fact about your readiness, not a gap in our data.
- 05ConsistencyScores that swing between sessions attract a penalty, capped at six points. A candidate whose sessions vary is slightly less ready than one whose do not; it is a nudge, not a verdict.
Time management is measured but not folded in. Most practice is untimed, so seconds per question is mostly noise. Where it does mean something — a full mock exam — it is reported on the results screen against the real exam’s own budget, rather than blended into a number that would then move for reasons you cannot see.
The four things it will say
| Band | From | What it means |
|---|---|---|
| Ready | 88% | Strong in every domain, on enough questions to mean it. |
| Likely ready | 75% | Consistently above the level this exam asks for, across the blueprint. |
| Nearly ready | 60% | Close. Clear the weakest domain, then take another full mock exam. |
| Keep practising | 0% | There is real ground to cover before this exam is worth booking. |
Two examples
Both are the real component, driven by the real engine, over an illustrative attempt log. The second is the one that matters: it is what the product says when it does not know.
Three weeks in, most of the blueprint covered
- Design Secure Architectures95%
- Design Resilient Architectures88%
- Design High-Performing Architectures82%
- Design Cost-Optimized Architectures68%
At 68% it is your weakest domain, and at 20% of the exam it is also where marks move fastest. 20 questions here will shift your readiness further than 20 anywhere else.
Thirteen questions in, two domains untouched
- Design Secure Architectures80%
- Design Resilient Architectures86%
- Design High-Performing ArchitecturesNot tested
- Design Cost-Optimized ArchitecturesNot tested
Twenty mixed questions across the whole blueprint is enough for a first read on where you stand. Until then there is not enough evidence here to tell you anything worth acting on.
What it is not
- Not a probability of passing. We have no data linking this number to real exam outcomes and will not imply one until we do.
- Not the provider’s scoring model. AWS scales 100–1,000 and Microsoft asks for 700; neither is a percentage of questions answered correctly, and neither is comparable to this.
- Not a measure of ability. It is a summary of what you have demonstrated on this product, on these questions.
- Not an average. The marker on every domain bar is 85% — the level at which a domain stops being the thing holding you back — and readiness is a weighted, shrunk, recency-adjusted combination rather than the mean of those bars.
Questions people ask
Is readiness a probability of passing?
No, and we will not present it as one until there is evidence to support the claim. It is a summary of what you have demonstrated on this product, against the provider’s published blueprint. Nobody has published data linking a third-party practice score to real exam outcomes, and a number that implied otherwise would be the most consequential thing on the site to be wrong about.
Why is my readiness lower than my domain scores?
Because readiness accounts for two things a domain score does not: how much of the blueprint you have actually tested yourself on, and how consistent your sessions are. A candidate scoring 90% in three domains and nothing at all in the fourth has not demonstrated readiness for the exam, only for three quarters of it.
Why does it sometimes refuse to give me a number?
Because below a certain amount of evidence any number would be a guess dressed as a measurement. Under 30% overall confidence the product says what is missing instead — usually that a domain has not been touched.
Can my readiness go down?
Yes, and it should be able to. Recent attempts carry more weight than old ones, so a run of wrong answers moves it. Evidence also goes stale: a domain you last practised four months ago counts for less than one you practised last week, because cloud exams change.
Does getting easy questions right count the same as hard ones?
No. A multi-constraint scenario question carries nearly three times the weight of a recall question, because answering it correctly demonstrates nearly three times as much — and getting it wrong costs accordingly.