DEA-C01 Exam Guide and the Data Engineer Role

Every AWS certification begins with one document: the official exam guide. For DEA-C01 that single document does two jobs at once, it tells you exactly how the exam is built (code, minutes, questions, cut score, weights) and it defines the data engineering job you must already do: ingest, transform, store, operate and secure data pipelines on AWS. Candidates who skip the guide and jump straight to flashcards fall into the same three traps: they believe a third-party page that says the exam is 170 minutes (it is 130), they convert the scaled cut score of 720 into a percentage (it is not 72%), and they cannot separate what is in scope from what merely sounds like real data work, such as training a machine-learning model, which the guide explicitly excludes even though integrating a large language model into a pipeline is now a named skill.

In this lesson you will:

• read the exam guide the way an exam writer does: code, timing, cost, scoring, languages, and the 170-minute / 85-question / 300-USD myths that third-party sites still print;
• turn domain weights into a concrete study-hour budget;
• map all 17 task statements and 120 skill bullets across the four content domains;
• master the two question types and the compensatory scoring model;
• separate the in-scope service list (14 categories) from the out-of-scope list that feeds distractors;
• define the data engineer role on AWS: what the four job-task clusters mean day to day, and the three tasks the guide says are not yours;
• build the end-to-end pipeline architecture, ingest, store, transform, govern, serve, with two reference diagrams and a store decision tree;
• work through nine worked examples: time budget, derived item counts, study hours, retake cost, the 720 arithmetic, the Athena cost ladder, FINRA throughput;
• study four real AWS customer case studies, Nasdaq, Integral Ad Science, FINRA and GE Aerospace, with services, numbers and sources;
• read the October 2026 update box: guide v1.1's eight new skills, six added services, and the 2025-2026 service wave;
• practise with 14 exam-style questions plus four interactive checks (matching, fill-in-the-blank, ordering and an interactive chart).

---

1. What the DEA-C01 exam measures

1.1 The target candidate

DEA-C01 is an Associate credential, and the guide states its purpose in one sentence: the exam "validates a candidate's ability to implement data pipelines and to monitor, troubleshoot, and optimize cost and performance issues in accordance with best practices." This is not an architect's exam and not an analyst's exam; it is the exam of the person who builds and runs the pipelines.

1.2 The out-of-scope job tasks: verbatim

The guide draws a hard line, and it is worth memorising word for word:

"The following list contains job tasks that the target candidate is not expected to be able to perform. This list is non-exhaustive. These tasks are out of scope for the exam: Perform ML training and inferences. / Demonstrate knowledge of programming language-specific syntax. / Draw business conclusions based on data."

This list is not decoration. Items from it appear as distractors, because a distractor that sounds like "real data work" is the most tempting wrong answer of all.

1.3 Recommended knowledge the guide assumes

The guide lists prerequisites in two columns. Nothing here is a separate domain; it is the floor of assumed knowledge:

• 📚 Did you know? "General concepts of vectors" and "vector indexes" are on this exam because guide v1.1 (12 December 2025) added them: Skill 2.1.8 "Describe vector index types (for example, HNSW, IVF)" and Skill 2.4.6 "Describe vectorization concepts (for example, Amazon Bedrock knowledge base)". A 2023 study plan would not have contained either bullet; the current guide does, and v1.1 revisions were published at least one month before they could affect a live exam (revisions page, as of Oct 2026).

1.4 The four job-task clusters: the exam's spine

The guide opens with four bullets that map 1:1 onto the four domains. Read them as the loop of the job:

The ordering is deliberate: ingest → store → operate → secure. Domain 1 is heaviest because ingestion and transformation are where data engineering either works or fails; Domain 4 is smallest but is the only domain that can fail a regulated company outright.

---

2. Exam logistics: the numbers that change your strategy

2.1 The logistics table

Every row below comes from the exam guide, the certification page's exam overview, or the AWS Certification FAQs, all accessed October 2026.

2.2 The three myths every prep site still prints

The digest of official sources flags three conflicts that recur on third-party pages (checked Oct 2026). In each case the official number wins:

2.3 Worked example E1: the per-question time budget

The single most useful number in the whole guide is arithmetic that nobody publishes:

$$t = \frac{130 \times 60}{65} = \frac{7{,}800}{65} = 120 \text{ seconds per question, exactly}$$

Reserve 10 minutes for review and the real budget becomes $120 \times 60 / 65 \approx$ 110.8 seconds per item. Budgeting rule of thumb: rapid-fire items at 60 to 90 seconds early, spend 3 to 4 minutes on the slower scenario and multiple-response items, and never let a single item eat more than about five minutes.

2.4 Worked example E2: scored versus unscored

$$65 = 50 + 15 \qquad \frac{15}{65} = 23.1\% \text{ of the exam does not affect your score}$$

Nearly a quarter of the paper is invisible pilot material, and you cannot tell which 15 they are. The only sane strategy is to treat every single item as scored.

2.5 Worked example E3: the cost of a retake

Three attempts cost 450 USD as of Oct 2026 plus at least 28 calendar days of waiting between them. The mirror-image benefit: once you pass, the certification page grants 50% off your next AWS exam: a 150 USD Associate registration becomes 75 USD (as of Oct 2026), and a 300 USD Professional or Specialty registration becomes 150 USD.

2.6 What happens after you sit the exam

Results arrive within five business days. The FAQs say "most AWS Certification exams" report without an on-screen pass/fail, so plan for delayed results: the deterministic artefacts are the score report and the Credly badge if you passed. Between now and exam day, the only mechanics that matter are: answer every item (blank = wrong), finish all 65 (you have 120 seconds each), and budget 150 USD per attempt (as of Oct 2026).

• 📚 Did you know? DEA-C01 had exactly one beta window: registration ran from 2 November 2023, delivery from 27 November 2023 to 12 January 2024, at the beta price of 75 USD. The exam went GA in April 2024, the same month DAS-C01 (Data Analytics, Specialty) retired on 9 April 2024, DEA-C01 is its spiritual successor at the Associate level (AWS training and certification blog, as of Oct 2026). There has been no DEA beta in 2026; if a prep page mixes another exam's 2026 beta logistics into a DEA-C01 summary, it is wrong.

---

3. The four domains and their weights

3.1 Official weights

The guide states four content domains, and the percentages are shares of scored content only:

3.2 Worked example E4: derived item counts (arithmetic, not AWS's)

Multiply the 50 scored items by each weight:

3.3 Worked example E5: turning weights into study hours

Suppose you have 40 hours before exam day. The weights convert directly into hours:

Domains 1 + 2 are 60% of the score. A candidate who spends most of their time on security because "that's always the biggest AWS topic" has misallocated the paper: on this exam the pipeline and the store come first.

3.4 Guide v1.0 to v1.1: what changed and when it counts

The exam was refreshed in December 2025: version v1.1 (12 December 2025) consolidated the separate "knowledge of" and "skills in" lists into one skills list per task, added eight new skills, removed none, and changed both service lists. AWS's revision mechanic (training and certification blog, 17 December 2025) is explicit:

The eight new v1.1 skills:

• all directly examinable now: 1.2.10 integrate LLMs for data processing
• 2.1.7 manage open table formats (for example Apache Iceberg)
• 2.1.8 describe vector index types (HNSW, IVF)
• 2.2.6 create and manage business data catalogs (Amazon SageMaker Catalog)
• 2.4.6 describe vectorization concepts (Amazon Bedrock knowledge base)
• 4.1.7 use domain, domain units and projects for SageMaker Unified Studio
• 4.5.6 manage data access through SageMaker Catalog projects
• 4.5.7 describe governance data framework and data sharing patterns

---

4. The task-statement map: 4 domains, 17 tasks, 120 skills

The guide does not publish vague topics; it publishes task statements, each carrying a numbered list of skills. Everything examinable hangs off one of these 17 tasks and its 120 skill bullets.

4.1 The 17 tasks in one line each

4.2 Skill-density: where effort actually goes

Skill bullets per weight point reveal the exam's texture:

D4 carries 1.61 skill bullets per weight-point, the highest density on the exam. Security questions are narrow, specific and unforgiving; Domain 1 questions are wide and scenario-based. Study time follows weights; revision cadence follows density, revisit Domain 4 more often in short bursts.

Hover any bar to read the exact ratio: D2 sits at 1.00 (the baseline), D1 just above it at 1.09, D3 at 1.27 and D4 at 1.61. The chart makes the study rule visual: the tallest bar is the shortest domain, so Domain 4 earns the most frequent, shortest revision sessions per hour of exam weight.

---

5. Question types and scoring mechanics

5.1 Exactly two question types

There are no ordering, matching, drag-and-drop or case-study items in the DEA-C01 guide, and no successor code has been published as of Oct 2026, so if a practice test shows you those formats, it is not following this exam's guide.

5.2 The scoring pipeline

Compensatory means one thing: only the overall result must clear 720. You do not need to reach 720 inside each domain: a weak Domain 4 can be carried by a strong Domain 1, and dominating one domain cannot rescue an overall score below the cut.

5.3 Worked example E6: 720 is not 72%

720 out of 1,000 looks like 72%, and the naive conversion $0.72 \times 50 = 36$ correct answers is arithmetically clean, and wrong as an exam fact. The scale starts at 100, so the usable range is $1{,}000 - 100 = 900$ points:

$$\frac{720 - 100}{900} = \frac{620}{900} \approx 68.9\%$$

AWS scales scores across multiple exam forms that may differ in difficulty, so the raw number of correct answers required for 720 is not fixed and is not published. The only deterministic rule is behavioural: unanswered items are scored incorrect, so answer all 65.

5.4 Worked example E7: compensatory arithmetic (planning illustration)

As a derived planning aid (not an AWS-published conversion: exact per-domain counts on a live form are never published), take the E4 grid of 17/13/11/9 scored items. A candidate who performs strongly on Domains 1 to 3 and poorly on Domain 4 has still cleared three of four weight bands; the compensatory model then only asks whether the overall scaled result reaches 720. The mirror case is the danger: a candidate who aces Domain 4 (18%) but neglects Domains 1 to 2 (60%) is fighting the weights themselves. Conclusion the exam rewards: no domain may be abandoned, but the correct order of coverage is 1 → 2 → 3 → 4 by weight, with Domain 4 revisited often because of its 1.61 skills-per-point density.

---

6. In-scope versus out-of-scope services

6.1 Two lists, one rule

The guide publishes an In-Scope AWS Services list spanning 14 categories (78 line items, about 75 distinct because SageMaker AI and S3 Tables are each printed twice: as of Oct 2026), and a separate Out-of-Scope AWS Services list. Both carry AWS's standard disclaimer that the list is non-exhaustive and subject to change. The in-scope list is what you must recognise; the out-of-scope list tells you which names are being used as baits.

Bold names were added at guide v1.1 (12 December 2025): Amazon Aurora, Amazon Q, Amazon Bedrock, Amazon Kendra, AWS Data Exchange and Amazon S3 Tables.

6.2 The out-of-scope list: services that exist but are not examinable

• Amazon FinSpace
• Alexa for Business
• Amazon Chime
• Amazon Connect
• AWS IQ
• Amazon WorkMail
• AWS App Runner
• AWS Elastic Beanstalk
• Amazon Lightsail
• AWS Outposts
• AWS Serverless Application Repository
• Red Hat OpenShift Service on AWS (ROSA)
• AWS Fault Injection Simulator (AWS FIS)
• AWS X-Ray
• AWS Amplify
• AWS AppSync
• AWS Device Farm
• Amazon Location Service
• Amazon Pinpoint
• Amazon SES
• FreeRTOS
• AWS IoT 1-Click
• AWS IoT Device Defender
• AWS IoT Device Management
• AWS IoT Events
• AWS IoT FleetWise
• AWS IoT RoboRunner
• AWS IoT SiteWise
• AWS IoT TwinMaker
• Amazon DevOps Guru
• AWS Activate
• AWS Managed Services (AMS)
• Amazon Elastic Transcoder
• AWS Elemental Appliances and Software
• AWS Elemental MediaConnect
• AWS Elemental MediaConvert
• AWS Elemental MediaLive
• AWS Elemental MediaPackage
• AWS Elemental MediaStore
• AWS Elemental MediaTailor
• Amazon Interactive Video Service (Amazon IVS)
• Amazon Nimble Studio
• AWS Mainframe Modernization
• AWS Migration Hub
• EC2 Image Builder

Four names were removed from the out-of-scope list at v1.1 (Amazon Honeycode, Amazon WorkDocs, Amazon Timestream, Amazon CodeWhisperer), and nothing was added to it: removal from the out-of-scope list does not make a service in-scope; services on neither list are simply not guaranteed.

6.3 Naming traps in the in-scope list

• "Amazon Quick" is the literal string on the in-scope Analytics list (checked Oct 2026), while Task 3.2.1 names "Amazon QuickSight" and the product line is now Amazon Quick Suite. Treat the Quick family as in-scope; do not claim the guide prints "Amazon QuickSight"; it does not.
• Amazon Kinesis Data Firehose is what the in-scope list still prints, while the product was renamed Amazon Data Firehose on 9 February 2024 with no API, endpoint, CLI, IAM-policy or CloudWatch-metric changes. Either name can appear on the exam; never "correct" an option that says Kinesis Data Firehose.
• AWS SCT (Schema Conversion Tool) was removed from the in-scope list at v1.1, yet Skill 2.4.3 still reads "Perform schema conversion (for example, by using AWS SCT and AWS DMS Schema Conversion)". Prefer AWS DMS Schema Conversion as the current tool; recognise SCT as the legacy name the skill text preserves.
• Amazon Q (Developer Tools) is not the same entry as Amazon Quick (Analytics). Two different strings, two different categories.
• Amazon SageMaker AI appears twice (Analytics and Machine Learning), and Amazon S3 Tables is printed twice in Storage: both are document duplication on the official page, not extra services.

6.4 Which artefact wins

• 📚 Did you know? The official in-scope page literally contains the single entry "Amazon Quick" in the Analytics category (verified Oct 2026): "Sight" appears nowhere on that page, even though Task 3.2.1 and the product pages use QuickSight / Quick Suite. AWS has not clarified whether it denotes the Quick Suite family or a truncation, so the safe exam behaviour is: treat any Quick-family name as in-scope, and never assert the guide says "Amazon QuickSight".

---

7. The data engineer role on AWS

7.1 What the four clusters look like on a Tuesday

The guide's job-task clusters are not abstract: each maps to a concrete working loop:

7.2 What the role is not

Restating the out-of-scope list as positive boundaries: the DEA-C01 data engineer does not train or run inference on models (Skill 1.2.10 stops at integration), does not demonstrate language syntax (1.4.3 tests concepts), and does not draw business conclusions (Task 3.2 stops at aggregation, cleaning and verification). The role's product is data that is correct, timely, complete, governed and cheap enough to keep: every question can be scored against that sentence.

7.3 The comparison habit the guide tests

Three times across the skill lists the guide uses the same verb: compare. You must be able to compare services for cost, performance and functional differences (recommended knowledge), provisioned versus serverless (Skill 3.2.5), and managed versus unmanaged (Skill 4.1.6). That is why this course teaches every service with a "choose it when / avoid it when" pair, and why the exam's favourite distractor is a real AWS service that is merely worse for the stated requirement.

---

8. End-to-end pipeline architecture: ingest, store, transform, govern, serve

8.1 Reference architecture 1: the streaming path

Domain 1's streaming skills (1.1.1, 1.1.7, 1.1.10, 1.3.4) compose into one canonical shape. Note the stage order the guide implies: ingest → store → transform → govern → serve, with monitoring wrapping everything.

8.2 Reference architecture 2: the batch lake house

Domain 1's batch skills (1.1.2, 1.1.5, 1.1.6, 2.2.3) plus Domain 2's lifecycle skills compose into the lake-house shape: the pattern behind most of the case studies in this course:

8.3 Reference architecture 3: the store decision tree

Task 2.1 asks you to choose the store from the access pattern, not from habit. This tree is the single most reusable diagram in Domain 2:

8.4 Worked example E8: the format decision that dominates the Athena bill

Skill 1.2.6 ("transform data between formats, for example from .csv to Apache Parquet") and Skill 1.2.4 ("optimize costs while processing data") meet in one calculation. Amazon Athena bills 5 USD per TB scanned with a 10 MB minimum per query (as of Oct 2026; AWS Athena pricing page). Scan a 3 TB dataset:

The arithmetic ($3 \times 5 = 15$, $1 \times 5 = 5$, $0.25 \times 5 = 1.25$) is trivial; the examinable insight is the ordering: columnar conversion first, partition pruning second, then worry about cluster size. A candidate who "optimises" by provisioning a larger EMR cluster before converting CSV to Parquet has optimised the wrong layer, and Domain 1 tests exactly that judgement.

8.5 Worked example E9: throughput arithmetic for a streaming ingest

FINRA, a real AWS customer, states it "processes approximately 6 terabytes of data and 37 billion records on an average day" with busy days generating "75 billion+ records" (AWS Public Sector blog, as of Oct 2026). Convert to per-second rates (derived: AWS publishes the daily totals, not the rates):

$$\frac{37 \times 10^9}{86{,}400} \approx 428{,}000 \text{ records/second (average day)}$$ $$\frac{75 \times 10^9}{86{,}400} \approx 868{,}000 \text{ records/second (busy day)}$$

Two exam-relevant consequences: (1) at those rates, shard or partition sizing and fan-out (Skills 1.1.10, 1.1.9) are first-order design decisions, not tuning details; (2) when an exam scenario quotes a daily record count, converting it to records-per-second is a legitimate 20-second move that instantly tells you whether the story is a Kinesis story, a Firehose story, or a batch-window story.

2026 Updates (as of October 2026)

• 📚 Did you know? AWS's own revision page shows only v1.0 and v1.1 for DEA-C01 as of October 2026: no successor series code has been published, and no dedicated "DEA-C01 2026 update" certification-blog post exists. The safe phrasing for anything version-sensitive is "in the current v1.1 exam guide"; the exact date v1.1 content first appeared on a live form per language is not published, because delivery "takes effect on a rolling basis" (certification page, as of Oct 2026).

• 📚 Did you know? The version you are being tested on is a moving target by design: AWS Glue 5.1 (26 November 2025, Spark 3.5.6) is the default version for new Glue jobs, while Glue 6.0 (21 August 2026) is simply the latest, so a 2024 note that says "pin GlueVersion=4.0" is already stale, and a claim that "you must run the latest version" is wrong too. Defaults matter on this exam because Skill 1.2.4 (optimise cost while processing) and Task 3.1 (automate data processing) both assume you know what a service does without you specifying anything (AWS Glue release notes + version support policy, as of Oct 2026).

---

Real-World Case Studies

AWS publishes what these abstractions look like in production. Every figure below is customer- or AWS-claimed and unaudited, with the source named so you can check it: the examinable point is the pattern (which skill was applied, which services did the work, which number moved), not the marketing. All four stories are drawn from AWS-owned properties only, accessed October 2026.

Case A, Nasdaq: decoupled storage and compute at exchange scale (Domains 1, 2, 4)

"We were able to easily support the jump from 30 billion records to 70 billion records a day because of the flexibility and scalability of Amazon S3 and Amazon Redshift.": Robert Hunt, VP Software Engineering, Nasdaq

Read Nasdaq as the reference lake-house pattern from Section 8.2: a cheap, durable S3 write path decoupled from an expensive, optimised Redshift read path, with Glacier for the coldest tier and Object Lock for the compliance tier. The examinable idea is the decoupling itself: storage and compute scale independently, which is what makes "90% of load 5 hours sooner" possible without doubling warehouse cost.

Case B, Integral Ad Science: governance that collapses complexity (Domain 4)

"With Lake Formation tag-based access controls, IAS reduced hundreds of permission rules down to precisely two rules."

The examinable idea is policy as tags, not as rule lists: classify data once, attach tags, grant on tags, inherit downward. That is Skill 4.2.5 ("apply authorization methods that address business needs, role-based, tag-based, and attribute-based") rendered as a production architecture, and it is why Lake Formation appears in both Domain 2 and Domain 4 skill lists.

Case C, FINRA: regulated, high-volume ingestion where design is the job (Domains 1, 2, 3)

"FINRA processes approximately 6 terabytes of data and 37 billion records on an average day … On busy days, the stock markets can generate 75 billion+ records.": John Brady, VP Cyber Security/CISO, FINRA

FINRA is the scale case: it is the live worked example in Section 8.5, where 37 billion records per day converts to roughly 428,000 records/second (derived) and 75 billion to roughly 868,000/second. The examinable idea is that at that rate ingestion design (shard/partition sizing, fan-out, replay) is the whole job, and that separating a cheap S3 storage tier from EMR/Redshift query engines is what turns "petabytes of surveillance data" from a cost problem into a query problem: the same decoupling Nasdaq uses, at a different domain emphasis.

Case D, GE Aerospace: migrating an operational data store without a rewrite (Domains 2, 3)

"We were seeing queries that used to run for an hour and a half running in 7 minutes using Amazon Redshift.": Bejoy John, Senior Director of Data Analytics, GE Aerospace

Read GE Aerospace as the migration + right-sizing pattern (Tasks 2.1 and 3.2): prove it in a PoC, move the store with its reports rather than around them, and let a managed, separately-scalable warehouse absorb growth instead of capacity guesses. Two exam-discipline notes travel with this case: the percentages are GE's claims against GE's own baseline, never generalise "70% faster" to another workload, and the case is a store-selection story, not permission to skip governance; compliance constraints were a stated requirement of the migration, which is why Domain 4 skills ride along with every Domain 2 choice.

• 📚 Did you know? Nasdaq's "70 billion records a day" and FINRA's "37 billion records" are not isolated curiosities, AWS's Kinesis customer page quotes Hearst ingesting 30 TB per day of clickstream data across 250+ sites ("I don't know how we could have made our clickstream data pipeline work without Amazon Kinesis services", Peter Jaffe, Hearst). The three together are why Domain 1 spends 12 skills on ingestion alone: at these volumes, ingestion design is the job (AWS customer pages, as of Oct 2026).

• 📚 Did you know? The migration trio behind cases like GE Aerospace each carries its own headline number, and they are not interchangeable: EOS Group reports a 50% reduction in infrastructure costs with zero data loss and minimal downtime after moving an on-premises warehouse with AWS DMS into Amazon S3 and Amazon Redshift (using dynamic data masking for PII), while PayU consolidated roughly 40 production databases into Glue → S3 → Redshift and cut query volume from 150,000 to 35,000 per month (−77%) in one month: "a 6-month exercise in the previous environment". Different baselines, different levers: one saved infrastructure, one saved queries (AWS case studies, as of Oct 2026).

---

Practice Questions