How a DEA question is built
Lesson 2 of 3 in The DEA-C01 Exam: Blueprint, Logistics, Strategy.
There are exactly two response types on this exam, and knowing that is worth real time. Multiple choice has one correct response and three distractors. Multiple response has two or more correct responses out of five or more options, and it is graded as an exact set — no partial credit for getting one of two right. Nothing else appears: no drag-and-drop, no ordering a sequence, no matching columns. If a practice product serves you those, it is modelling a different AWS exam.
So every item you meet is one of two things, and both are built the same way. Underneath the prose, a question has three parts:
- The setting — a company, a volume, a current architecture. Usually two or three sentences, and usually mostly decoration.
- The constraint — the clause that actually eliminates options. "Within five seconds." "With the least operational overhead." "Without provisioning servers." "Data must not leave the Region."
- The question sentence — the one that tells you what kind of answer is wanted: which solution meets these requirements?, which action should the engineer take first?, what is the most cost-effective approach?
The engineering of a good item happens in part 2, and the engineering of a trap happens by making part 1 long enough that you forget to look for part 2.
| Format | What it looks like | How it is graded | The tactic |
|---|---|---|---|
Multiple choice | One correct response, three distractors. | Right or wrong. | Read the question sentence first, then hunt the constraint, then read the options. Eliminate on the constraint before you compare the survivors on merit. |
Multiple response | Two or more correct out of five or more options. The stem states how many to pick. | Exact set. One right of two scores the same as none. | Take the count off the stem before reading the options, because it changes how you read them. Then complete the set — a half-answer is a zero, so the last slot is a free guess. |
Two scoring facts, both of which change how you spend minutes rather than how you study.
Unanswered questions are scored as incorrect, and there is no penalty for guessing. A blank is therefore strictly worse than any guess, including a random one. That makes the end-of-exam rule unambiguous: with two minutes left, fill every empty item with something.
A multiple-response item is worth exactly one mark, the same as the easiest recognition question on the paper, while taking two or three times as long to read. That asymmetry is the whole of pacing strategy: an item you cannot resolve deserves a committed guess and a flag, never the time of two other questions.
Working one item — constraint before options
- Read the question sentence first
It is the last sentence, and it tells you which kind of answer wins: the cheapest, the fastest, the least operational effort, the first action. Reading it first means every word of the setting arrives already sorted into relevant and decorative.
- How many answers?
Multiple response always tells you. Fix the number in your head before the options, because two cooperating answers read very differently from one standalone answer.
- Name the deciding constraint
Latency, volume, query shape, cost and operational effort, or governance. Say which family out loud. If you cannot find one, re-read — a well-built item has exactly one, and an item with none is usually testing plain service recognition.
- Eliminate against the constraint, not on taste
Ask of each option: does this satisfy the stated clause? Not "is this a good design". Three of four options are usually good designs; only one survives the clause.
- How many survived?
- Break the tie on the question sentence
Two survivors means you have two correct designs and one of them is more of what the sentence asked for: cheaper, less to operate, faster to build. Re-read that sentence and pick on its axis.
- You missed a constraint — re-read the stem
If nothing was eliminated, there is a clause you skimmed. It is often a number with a unit, a negation, or a phrase like "existing" or "without changing".
- Commit, flag if unsure, move on
A committed answer plus a flag costs you nothing and buys a second look. An empty item costs you the mark outright.
The five distractor families, and how each one is beaten
Incorrect options are written to be chosen by someone with incomplete knowledge, which means they are plausible on purpose. Five recurring shapes cover most of them.
- Right answer to a different question. A perfectly good streaming design, in a stem whose constraint was nightly. Beaten by naming the constraint first.
- The near neighbour. The service next door that does almost the same job: a managed delivery stream where a durable replayable stream was required, a search engine where a warehouse was required. Beaten by knowing the one property that separates the pair.
- Works, but violates a stated clause. It would deliver the result — by provisioning a cluster in a stem that said no servers to manage, or by copying data in a stem that said the data must not leave the Region. Beaten by re-reading the clause after you pick.
- The thing that service does not do. A capability attributed to the wrong product: a lifecycle rule asked to change a value rather than move an object, a monitoring service asked to rewrite data. Beaten by one-line knowledge of each service’s actual job.
- Correct but disproportionate. Build it yourself with code when a managed feature exists, in a stem that asked for the least operational overhead. Beaten by reading the question sentence as a scoring axis, not as flavour.
Multiple response — read the count, then read for cooperation
When a stem says to select two, the two correct answers usually cooperate: one turns a capability on and the other points it at the right target, or one changes the storage layout and the other changes the query that benefits from it. Two options that achieve the same thing by different means are rarely both keys — more often one is right and the other is its near neighbour.
And finish the set. Grading is an exact set, so three of four correct earns exactly what zero of four earns. If you are sure of two of three, pick the least-wrong third and move on: you have converted a guaranteed zero into a real chance.
Where the “first action” questions come from
A sentence like which action should the engineer take first is not asking for the best eventual fix — it is asking what must happen before anything else can. In data engineering that is nearly always one of three things: stop making it worse (pause the ingestion, disable the schedule), preserve what you need to diagnose (the raw landing data, the failed run’s logs), or establish the facts (read the job’s error, look at the metric, check what the catalog says the schema is).
The commonest wrong answer to a "first" question is a correct and necessary later step — rerunning the pipeline, rebuilding the table, fixing the upstream producer. It is not wrong as engineering; it is wrong as an ordering.
Why there are no ordering or matching items here
Only one AWS exam currently uses them, and it is the Security – Specialty. This exam stayed with two response types, which has a practical upside and a subtle downside.
The upside is that nothing is graded as a sequence, so you never lose a whole mark for knowing all four steps and swapping two. The downside is that dependency knowledge still gets tested — it just gets tested as a "first action" multiple-choice item, where the three later steps sit in front of you as distractors that are all genuinely part of the fix.
Interactive checkpoint quiz (2 questions) — open this page in a browser to take it.
Key terms: Multiple response, Distractor, Requirement qualifier, Deciding constraint