Data Engineering Quiz and Assessment

Data Engineering & Modern Data Platforms · Lesson 67 · Production Data Engineering and Capstone

Lesson Overview

Capstone week does not begin when you open a blank architecture doc. It begins when you can sit for a timed assessment, miss a question honestly, and know which lab to reopen. This lesson is the TechLearningPro evaluation design: multiple-choice sets with explanations, short architecture prompts, scoring rules, timeboxes, and a method to self-assess without lying to yourself. It is also teaching content. You will learn how strong assessments are built so you can study the shape of questions, not only their answers.

Lessons 65 and 66 trained interview speech and graded labs. Data Engineering System Design practiced whiteboards. Here the clock is smaller and the coverage is wider: foundations, ingestion, Spark, Kafka, lakehouse, cloud cost, reliability, and governance. The Monday board still matters. completion-v4 still pins the grain. Northline still appears when finance isolation is the cleaner story. Sample items with answers live in the Interview and Hands-On sections so you can grade immediately. Core Concepts and Technical Explanation teach blueprints, distractor design, and scoring math so the quiz is not a mystery box.

If you only memorize keys, you will fail the architecture prompts. If you only draw boxes, you will fail the MCQs that test null-safe SQL and offset commits. Balance the study plan the way you balance a pipeline: every stage gets cycles.

Learning Objectives

By the end of this lesson you will be able to:

1. Describe the modules of the TechLearningPro data engineering capstone assessment and their weights.
2. Sit a timed MCQ block and interpret scores by domain, not as a single vanity percentage.
3. Answer short architecture prompts with grain, failure modes, and refusals under a timebox.
4. Use a self-assessment protocol that separates knowledge gaps from fatigue and misreads.
5. Explain why each MCQ distractor is wrong without needing the answer key first.
6. Map wrong answers to lesson 66 labs and lesson 65 interview banks for remediation.
7. Apply scoring rubrics for architecture prompts consistently across two practice attempts.
8. Timebox study: retrieval practice over rereading.
9. Design a personal mock exam day that mirrors the official cadence.
10. Recognize assessment anti-patterns: second marts, auto-commit myths, raw board scans.
11. Include cost meters and SLO language when prompts touch cloud or reliability.
12. Produce a remediation plan with dates before starting the capstone project.
13. Use sample quiz items in this lesson as calibration, not as the only pool.
14. State what score bands mean for readiness: remediate, conditional go, go.
15. Enter Enterprise Data Platform Capstone Project with evidence, not optimism.

Prerequisites

• Lesson 65 drill log with domain scores.
• Lesson 66 SCORES.md and exercise pack evidence.
• Comfort with TechLearningPro completion grain and provisional versus certified paths.
• Lessons 58 and 62 for cost meters and SLO vocabulary.
• Willingness to take a practice exam before peeking at answer keys.

Detailed Core Concepts

Assessment as a product

A good assessment measures decisions under constraint. It does not measure whether you can Google a config flag with unlimited time. TechLearningPro's assessment product has four parts:

1. MCQ block A — Foundations and SQL/modeling (20 items)
2. MCQ block B — Processing and streaming (20 items)
3. MCQ block C — Platforms, ops, cost, governance (15 items)
4. Architecture prompts (3 short designs, written)

Total seated time: about 150 minutes including breaks between blocks. Personal practice exams may split across two evenings. Official cadence assumes one sitting with a fifteen-minute break after block B.

Weighting and score bands

Overall bands.

These thresholds are teaching defaults. An employer loop may differ. The self-assessment skill transfers.

How to self-assess without cheating yourself

1. Print or freeze the question set.
2. Enforce the timebox with a visible timer.
3. Mark confidence: H/M/L per item while answering.
4. Score with the key only after the block is closed.
5. Classify each miss: knowledge, misread, carelessness, or second-guess.
6. For every knowledge miss, write one remediation task (lab step, interview drill, or lesson reread section).
7. Retake only a parallel form later; do not memorize this lesson's samples as the universe.

Confidence calibration matters. If you mark H and miss, your knowledge model is wrong. If you mark L and hit, you may be under-selling in interviews. Track both.

Distractor design you should learn to see

Strong MCQs use distractors that match real anti-patterns:

• Dropping DQ to hit an SLO
• Pausing capacity without measuring resume
• NOT IN with nullable columns
• Auto-commit before sink success
• Second warehouse as "isolation"
• Hourly partitions for a daily board
• Claiming exactly-once into arbitrary APIs
• Broadcasting huge dimensions

When you study, argue why each wrong option fails. That study method beats highlighting prose.

Architecture prompt anatomy

Each prompt states business goal, volume sketch, freshness, and one twist. Your answer must include:

• Grain and definition versioning
• Happy-path stages
• Failure modes and checks
• Serving choice and isolation
• Refusals
• One meter or SLI

Rubric mirrors lesson 65's five layers, scored 0–5, then averaged.

Domain blueprint (what the MCQs aim at)

Foundations / SQL / modeling. Grain, windows, anti-joins, SCD choices, late resets, definition_version.

Ingestion / orchestration. Idempotent DAG tasks, backfills, sensors on data not clocks only.

Spark. Skew, shuffle, file sizes, broadcast misuse, stage evidence.

Kafka / streaming. Offsets, poison, watermarks, provisional vs certified.

Lakehouse. Merge idempotency, commits before delete, time travel drills.

Cloud / cost. Scans vs capacity, dry runs, idle hours, single publisher.

Reliability / governance. SLI/SLO/budget, least privilege, PII minimization, audit scope.

Sample MCQ set 1 — Foundations (with answers)

1. Monday board must count distinct learners completing a lesson under a pinned definition. Which filter is non-negotiable?

A. WHERE year(event_time) = 2026
B. WHERE definition_version = 'completion-v4'
C. WHERE course_id IS NOT NULL
D. WHERE event_type <> 'reset'

Answer: B. Version pin is the contract. A is a pruning hazard and weak. C is useful but insufficient. D may be wrong depending on reset modeling.

2. Which anti-join pattern is safest when completion keys can be null?

A. WHERE learner_id NOT IN (SELECT learner_id FROM completion)
B. LEFT JOIN completion c ON ... WHERE c.learner_id IS NULL
C. WHERE learner_id NOT IN (SELECT learner_id FROM completion WHERE learner_id IS NOT NULL) only, forever
D. EXCEPT ALL on unsorted JSON blobs

Answer: B (or NOT EXISTS). A fails with nulls. C is a partial patch people forget. D is nonsense for this grain.

3. Type-1 SCD is acceptable when:

A. Lesson eligibility for certification changes and history matters
B. Correcting a typo in a lesson title that never affects metrics
C. Always, because SCD2 is expensive
D. Never in lakehouses

Answer: B.

4. A window dedupe must break ties. Best ORDER BY?

A. ORDER BY event_time
B. ORDER BY event_time DESC, event_id DESC
C. ORDER BY random()
D. ORDER BY learner_id

Answer: B.

5. Partition column is event_date. Which predicate prunes?

A. DATE(event_time) = CURRENT_DATE
B. event_date = CURRENT_DATE
C. CAST(event_time AS STRING) LIKE '2026-11-11%'
D. Filter in the BI tool after SELECT *

Answer: B.

Sample MCQ set 2 — Spark and Kafka (with answers)

6. One Spark task runs 40 minutes, others 2. First investigation?

A. Double the cluster
B. Inspect stage task metrics for skew
C. Switch languages to Scala
D. Disable AQE permanently

Answer: B.

7. Broadcasting a "small" table OOMs the driver. Likely cause?

A. Table was not actually small after filters
B. Parquet is too compressed
C. Kafka lag
D. Missing primary key in Postgres

Answer: A.

8. Target Parquet file size for analytical reads in this course's guidance?

A. 1–2 MB
B. 128–512 MB range
C. One file per day always
D. Exactly 1 GB always

Answer: B.

9. Consumer writes to DB then crashes before commit with auto-commit false and idempotent sink. On restart:

A. Data must duplicate
B. Offsets rewind; idempotent sink prevents duplicate business rows
C. Topic is lost
D. Exactly-once is impossible forever

Answer: B.

10. Poison message handling that blocks a partition forever is:

A. Best practice
B. An anti-pattern; park and continue with alerting
C. Required for exactly-once
D. Only needed for Avro

Answer: B.

11. Provisional streaming tile and certified mart share one cursor. Risk?

A. None
B. Mixing honesty deadlines; certified numbers become unstable
C. Lower cloud cost always
D. Better SCD2

Answer: B.

Sample MCQ set 3 — Platforms, cost, reliability (with answers)

12. First redesign when forty analysts scan raw JSON Mondays?

A. Bigger warehouse
B. Point board at pruned mart; measure dry-run bytes
C. Disable DQ
D. Replicate to three regions

Answer: B.

13. Pausing warehouse capacity is wrong when:

A. It is Saturday
B. Measured resume plus merge misses 06:00 SLO
C. Idle hours exist
D. Finance asks

Answer: B.

14. Two warehouses publish the same mart after a bake-off. Primary risk?

A. Only cost
B. Cost and definition drift
C. Only network latency
D. None if both are cloud

Answer: B.

15. Pipeline green but board wrong. Missing?

A. More executors
B. Data checks: uniqueness, version pin, reconciliation
C. A new logo on the dashboard
D. Disabling alerts

Answer: B.

16. Error budget is:

A. Infinite retries
B. Allowed SLO miss rate over a window
C. Kafka retention
D. Spark shuffle buffer

Answer: B.

17. Cross-region replica without restore drill is:

A. Proven backup
B. Cost line with unverified recovery
C. Required for Iceberg
D. A substitute for DQ

Answer: B.

Building parallel forms

When you remediate, rewrite each missed item into a sibling: same skill, new story. Example: change completion to Northline refund status codes; change course_mega to seller_mega. Parallel forms prevent memorization theater.

Retrieval practice schedule

Sleep beats a third reread of lesson notes at midnight.

Additional MCQ set 4 — Ingestion and orchestration (with answers)

18. An Airflow task that succeeds when a file lands, without checking mart freshness against the SLO, is:

A. Sufficient production control
B. Incomplete; green compute can still miss the board promise
C. A replacement for DQ
D. Only wrong on Kubernetes

Answer: B. Sensors and landing checks are necessary but not sufficient. The publish path still needs data checks and a freshness SLI.

19. Backfilling completion-v3 days after v4 is live requires:

A. Overwriting v4 silently
B. Pinning definition_version per run and avoiding accidental cross-version merges
C. Dropping uniqueness
D. Pausing Kafka forever

Answer: B.

20. Idempotent daily ETL for a partition should:

A. Append blindly every run
B. Replace or merge the target day so reruns converge
C. Delete the entire table every morning
D. Depend on humans not clicking twice

Answer: B.

21. Catchup=True on a DAG that is not idempotent causes:

A. Nothing
B. Duplicate side effects across intervals
C. Automatic exactly-once
D. Lower cloud bills

Answer: B.

22. The best dependency for "start merge" is:

A. Wall clock 03:00 only
B. Upstream data presence + prior DQ for the interval
C. Random sleep
D. Manual Slack emoji

Answer: B. Clock-only schedules race late producers.

Additional MCQ set 5 — Lakehouse and governance (with answers)

23. Deleting small files before the table commit points at compacted files risks:

A. Nothing
B. Readers seeing a partial day
C. Better SLOs always
D. Free backups

Answer: B. Commit-then-expire ordering matters.

24. Time travel / version restore drills exist to:

A. Impress auditors with screenshots of settings
B. Prove recovery meets RPO/RTO with a named owner
C. Replace uniqueness checks
D. Justify dual marts

Answer: B.

25. Board service account should:

A. Read raw and mart
B. Read only published mart with least privilege
C. Own cloud root keys
D. Share personal user credentials

Answer: B.

26. PII in progress payloads: first design move?

A. Replicate to more regions
B. Minimize collection and restrict raw access
C. Print payloads in logs for easier debug
D. Store only in screenshots

Answer: B.

27. A column gate that blocks forbidden fields on publish fails open when:

A. It exits non-zero and blocks merge
B. It only logs warnings and merge continues
C. It is tested in CI
D. It pins definition_version

Answer: B. Fail open means bad data ships.

How to write your own sibling items quickly

Template:

1. Pick an anti-pattern from Common Mistakes in any Module 5–6 lesson.
2. Write a one-sentence scenario with TechLearningPro or Northline nouns.
3. Write the correct action as option B (vary letter positions when assembling forms).
4. Write three distractors: tool worship, silent correctness spend, and irrelevant true fact.
5. Write a one-sentence explanation.

Ten siblings per evening is enough to build a parallel Block A. Quality beats quantity; discard items where two options are arguably correct.

Misread reduction tactics

• Underline negatives: not, never, except, least, first.
• Restate the ask in five words before choosing.
• Watch for "first" versus "best long-term" mismatches.
• If two options seem right, reread the qualifier in the stem.

Many "knowledge" misses are literacy misses under time pressure. Track them separately so you do not waste lab hours.

Architecture and Design Diagrams

Assessment flow for one sitting:

mermaidCopy
flowchart LR
  WARM["Warm-up refusal list 5 min"] --> A["MCQ Block A 35 min"]
  A --> B["MCQ Block B 40 min"]
  B --> BRK["Break 15 min"]
  BRK --> C["MCQ Block C 30 min"]
  C --> ARCH["Architecture x3 45 min"]
  ARCH --> SCORE["Score and classify misses"]
  SCORE --> PLAN["Remediation plan"]

Self-assessment feedback loop:

mermaidCopy
flowchart TD
  ITEM["Missed item"] --> TYPE["Classify knowledge misread careless guess"]
  TYPE --> MAP["Map to lab or interview bank"]
  MAP --> TASK["One concrete task with date"]
  TASK --> RETEST["Parallel item later"]
  RETEST --> BAND["Update readiness band"]

Technical Explanation

Scoring MCQs fairly

Mark machine-scored. No partial credit on single-select. For multi-select items (if you add any in personal forms), require exact set match. Do not argue with the key during scoring; argue during remediation design afterward if an item is ambiguous, then fix the item.

Scoring architecture prompts

Use the 0–5 layer rubric:

Average three prompts. A beautiful diagram without checks caps at 2.

Timeboxing tactics

• MCQ: first pass answer all; flag L-confidence; second pass only flagged.
• Do not spend three minutes on one item in Block A.
• Architecture: outline bullets first for all three prompts (10 minutes), then deepen (35 minutes). Empty third prompt is worse than three medium answers.

Explaining answers out loud

After scoring, explain five random correct answers aloud. If you cannot, you guessed. Guessing is fine; unmarked guessing is dangerous for readiness bands.

Connecting to prior modules without reopening everything

Use targeted rereads: only the section that owns the miss. Cost miss → lesson 58 meters section. Skew miss → Spark shuffle lesson and Lab B. Offset miss → Kafka consumer lesson and Lab C. SLO miss → lesson 62. Governance miss → lesson 57/61. Capstone will not give you time to reread the course linearly.

Sample architecture prompt A — TechLearningPro board

Prompt (15 minutes). Design the certified path for Monday 06:00 completion under completion-v4 at ~40 GB Parquet/day progress, with a provisional ops tile. Twist: product wants one combined number on one tile.

Hire-bar sketch. Separate tiles or labels; Kafka → compact → batch merge → DQ → published mart → board; stream → provisional tile; single publisher; SLI freshness; refuse combined unlabeled metric; refuse raw JSON board scans; meter dry-run bytes.

Sample architecture prompt B — Northline finance noon close

Prompt. Portal p95 five seconds with RLS; finance needs refund-adjusted revenue by noon. Twist: team proposes a second warehouse.

Hire-bar sketch. One order_line grain; workload isolation; finance daily table from same publisher; contract test on refund codes; refuse second publisher; cost sheet notes idle vs required capacity.

Sample architecture prompt C — Replay after bad commit

Prompt. Consumer auto-commit advanced before sink write for two hours. Design recovery.

Hire-bar sketch. Stop consumers; estimate gap; restore sink from idempotent replay of offsets; fix commit order; add crash test; DLQ for poisons; communicate provisional uncertainty window to stakeholders.

How official distractors map to course anti-patterns

Keep a personal table while studying:

Written answer length norms

Architecture answers in this assessment should fit one page each: roughly 250–400 words plus a tiny diagram. Graders punish novels that never state grain. Practice writing short.

Integrity rules for self-assessment

No keys during the attempt. No second browser with lesson text for MCQs. Architecture may reference a one-page personal crib of refusal lists if you will have notes in your real interview take-homes; otherwise practice closed-book. State which mode you used in ASSESS.md.

Worked scoring example with mixed results

Suppose:

• Block A: 16/20 = 80%
• Block B: 12/20 = 60%
• Block C: 12/15 = 80%
• Architecture: 4.0, 3.0, 2.0 → average 3.0/5 = 60% on the 0–100 scaled arch component if you map linearly, but this lesson weights architecture as rubric average converted by arch_pct = (avg/5)*100 = 60%

Weighted overall: 0.25*80 + 0.30*60 + 0.20*80 + 0.25*60 = 20 + 18 + 16 + 15 = 69%.

Band: below 70% overall → remediate. Even if you rounded emotionally to "almost 70," Block B floor failed. Remediation focuses on Spark/Kafka labs and a parallel Block B, not on rereading governance.

Architecture clinic format (90 minutes)

1. Warm-up refusals (5 min)
2. Prompt B timed (15 min)
3. Peer rubric score (10 min)
4. Rewrite to include checks and meters (15 min)
5. Prompt C recovery timed (15 min)
6. Compare against hire-bar sketch (15 min)
7. Commit three personal rules to card (15 min)

Clinic goal is raising average above 3.5, not polishing handwriting.

Oral defense add-on (optional but recommended)

After MCQs, spend twenty minutes answering aloud:

• Explain your Lab C crash test.
• Defend why provisional and certified are separate.
• Compute a tiny dry-run ratio with made-up bytes.
• Name your SLI for 06:00.

Record it. Playback once. Cringes are curriculum.

Linking assessment to capstone acceptance

Capstone mentors should ask for ASSESS.md before reviewing architecture docs. A go band does not guarantee a good capstone, but a hold band predicts thrash. Treat the assessment as a merge gate for your own project branch of career time.

Practical Implementation or Design Walkthrough

Problem

Build and sit a personal mock exam weekend that mirrors the 150-minute cadence, then produce a remediation plan tied to labs.

Step 1 — Assemble forms

• 20 items Block A (use samples here plus siblings you write)
• 20 items Block B
• 15 items Block C
• Prompts A–C above

Step 2 — Create ASSESS.md

markdownCopy
# Mock exam record
Date:
Mode: closed-book / notes-allowed
Block A: xx/20  confidence misses:
Block B: xx/20
Block C: xx/15
Arch scores: p1 p2 p3 avg:
Overall weighted:
Band:
Top 5 remediation tasks with dates:

Step 3 — Sit Block A with timer 35 minutes

Score immediately after. Classify misses.

Step 4 — Break 5 minutes; sit Block B 40 minutes

Same protocol.

Step 5 — Break 15 minutes; sit Block C 30 minutes

Step 6 — Architecture 45 minutes

Outline all three, then deepen.

Step 7 — Compute weighted overall

Example: A 80%, B 70%, C 75%, Arch 3.6/5 = 72%.
Weighted: 0.25*80 + 0.30*70 + 0.20*75 + 0.25*72 = 20+21+15+18 = 74% → conditional go.

Step 8 — Remediation plan example

Step 9 — Retake policy

Retake parallel form no sooner than 48 hours. Same form invalidates the band.

Step 10 — Gate to capstone

Paste band into capstone kickoff notes. If hold, schedule architecture clinic with peer using lesson 65 rubric cards.

Enterprise Real-World Scenarios

TechLearningPro academy certification day

Cohorts sit the assessment before receiving production repo access. One engineer scores 85% overall but 2.0 on architecture by drawing tools first. Hold band triggers a clinic. Two weeks later architecture averages 4.0; capstone quality jumps. The assessment saved a messy project, not just a grade.

Northline vendor checkpoint

Vendors must pass a shortened Block B+C and one architecture prompt before touching refunds. A vendor aces MCQs, fails recovery prompt by proposing DELETE FROM sink without replay plan. Contract paused. The prompt measured production judgment better than a resume keyword match.

Candidate take-home plus onsite quiz hybrid

Some employers send a take-home (like lesson 66 pack) and then a short MCQ onsite to prevent ghostwriting. If your pack is real, MCQs feel easy. If the pack was copied, Block B exposes you. Treat this lesson as insurance.

Faculty calibration session

When multiple graders score architecture, they must co-score three anchor answers: weak (1), medium (3), strong (5). Disagreement beyond one point triggers rubric discussion before cohort grading. Self-assessors can simulate this by asking two peers to score the same prompt blind. If peers diverge, your answer is probably vague on checks or refusals—the layers that create shared certainty.

Progressive difficulty inside a block

Official forms should sort items roughly easy → medium → hard to reduce early panic, but not strictly. Personal mocks can shuffle. If you always miss the last five, it may be time, not knowledge. Track unanswered items separately from wrong items.

Accessibility and timing accommodations

If you use extended time in real exams, practice with the same accommodation so bands stay meaningful. Do not compare your extended-time score to a peer's standard-time score without noting the mode in ASSESS.md. The readiness question is "can I perform in the conditions I will actually face?"

Best Practices

• Timebox every practice block.
• Classify misses; do not only count them.
• Study distractors as anti-patterns.
• Keep architecture answers short and grain-first.
• Link every knowledge miss to a lab or drill.
• Track confidence calibration.
• Sleep before D0.
• Use parallel forms on retakes.
• Enforce honesty rules you would respect in a real exam.
• Review refusal lists the morning of the exam, not new tools.
• Write ASSESS.md every mock.
• Prefer retrieval practice to passive rereading.
• Treat conditional go as a schedule, not as shame.
• Bring lesson 66 evidence to oral follow-ups if offered.

Common Mistakes and Anti-Patterns

Rereading notes instead of timed retrieval. Feels productive; predicts poorly.

Memorizing sample keys only. Parallel forms will move. Learn skills.

Skipping architecture practice. MCQ-only high scores still hold at arch <2.5.

Arguing with the timer. Empty answers score zero.

Inflating self-scores. Capstone reality grades harder.

Ignoring L-confidence hits. Still verify; lucky guesses fade.

Remediating everything. Prioritize misses that are knowledge-typed in weak domains.

Studying tools not used in prompts. Stick to blueprint domains.

Combining provisional and certified in answers "to save tiles." Instant rubric cut.

Recommending second marts for isolation. Classic fail distractor—do not author it in arch answers.

Treating illustrative prices as quotes. Lesson 58 labeled ratios; your cost MCQs and Lab F must too.

Skipping confidence markers. You lose calibration data that would have prioritized H-misses.

Peer grading without a rubric card. Scores become personality contests.

Cramming twenty new tools on D-1. Raises anxiety; rarely raises architecture scores.

Ignoring per-block floors. A shiny overall percent can hide a Kafka hole that the capstone will fall into.

Cognitive load notes for exam day

Eat, hydrate, and use the break to stand, not to open Twitter or reread Iceberg specs. If a question triggers a spiral, mark L, answer your best, move on. Spirals spend the timebox that would have saved three later items. After the exam, spiral in remediation—not during Block B.

Troubleshooting and Design Review

Scores volatile across mocks. Likely sleep, timing, or form difficulty mismatch. Stabilize conditions.

Always miss cost items. Rebuild the five-meter list from memory daily for a week.

Architecture runs long. Practice outline-first with a 10-minute hard stop.

Peer grades differ by more than a point. Recalibrate with the rubric table side by side on one sample answer.

Anxiety spikes. Start with a five-minute refusal-list warm-up; write three refusals before Block A.

You finish MCQs early and second-guess half the paper. Stop. Early finishers often degrade calibrated answers. Use leftover time only on flagged L items, then stop again.

Architecture handwriting or typing turns into a tool catalog. Physically cross out tool names until grain and checks exist. Reintroduce one serving engine after.

Remediation plan is a wish list without dates. Add calendar days; undated plans do not change bands.

Parallel form is accidentally identical. Change nouns, numbers, and at least one constraint (for example, add GDPR delete). Same answer key text means invalid retake.

Extended remediation cookbook by miss theme

Each cookbook item should produce an artifact path listed in ASSESS.md. Reading without artifacts does not clear conditional go.

Capstone risk register seeded from assessment

Convert remaining weak themes into capstone risks:

markdownCopy
| Risk | Likelihood | Impact | Mitigation from labs |
|---|---|---|---|
| Duplicate attempts ship | M | H | Lab D gate in publish DAG |
| Hot course skew misses window | M | H | Lab B salt path + max task alert |
| Offset commit bug | L | H | Lab C pattern + crash test in CI |
| Board scans raw | L | H | Query lint + dry-run budget |

Bring this register to capstone kickoff. Assessment then becomes project input, not a discarded grade.

Design review before capstone:

• What is your band and date?
• Which block is weakest?
• Which lab will you reopen?
• Can you state completion grain in one sentence?
• What three designs do you refuse under time pressure?

Interview Questions and Answers

Sample quiz-style oral items with answers. Use as a live drill; covers assessment literacy plus technical checks. Distinct from lesson 65's deep bank.

Q1. Why weight architecture at 25% if MCQs are easier to score?

Because production judgment lives in designs under constraint. MCQs catch sharp edges; architecture catches system thinking.

Q2. What is a conditional go?

Overall pass-ish with a required remediation window before capstone, especially when one domain is weak.

Q3. Give two distractors for a cost question and why they are wrong.

"Disable DQ to save runtime" — spends correctness. "Pause without resume measure" — may break 06:00.

Q4. How do you classify a miss where you knew the content but misread "not"?

Misread, not knowledge. Remediation is slower reading practice and underlining negatives, not a full lab reopen—unless chronic.

Q5. What SLI fits the Monday board?

Fraction of weekdays with certified mart ready by 06:00 with checks green.

Q6. Why include confidence markers?

To calibrate self-knowledge and to prioritize review of H-misses.

Q7. When is a retake invalid?

When it reuses the same items within memory window; use parallel forms.

Q8. How does lesson 66 evidence help assessment day?

Oral follow-ups and architecture answers gain concrete meters, crash tests, and fail logs.

Q9. What refusal belongs in almost every TechLearningPro design answer?

No second certified publisher; no unlabeled mix of provisional and certified; no board on raw JSON.

Q10. Score this one-liner arch answer: "Use Kafka and Snowflake."

0–1. No grain, checks, or refusals.

Q11. Why teach distractor design to learners?

It turns MCQs into anti-pattern review and improves interview trap recognition.

Q12. What should you do the night before D0?

Light refusal review and sleep—not a new Spark whitepaper.

Q13. Block B 60%, overall 82%. Band?

Not a clean go if policy requires each block ≥70%; treat as conditional until Block B remediated. This lesson's suggested rule flags per-block floors.

Q14. How do you parallelize a null anti-join item?

Change entity names and table labels; keep the null trap; new story about refunds without completion keys.

Q15. What artifact gates capstone start?

ASSESS.md with band, plus remediation completion notes for conditional goes.

Q16. Why convert architecture scores via (avg/5)*100 when weighting?

So architecture can sit beside MCQ percentages in one weighted formula without inventing a separate opaque scale. Document the mapping in ASSESS.md.

Q17. A learner scores go but cannot demo Lab C. Ready?

No. Go assumes evidence still exists. Re-demo or hold. Assessment without lab proof is a paper victory.

Q18. How many refusals belong on the warm-up card?

Three is enough: second publisher, raw board scans, unlabeled provisional+certified. Add a fourth for commit-before-sink if Block B is weak.

Q19. What makes a sibling item invalid?

Two correct options, or a stem that depends on memorizing this lesson's exact sample wording rather than a skill.

Q20. Should Block C include Spark skew items?

Keep skew in Block B. Block C should stay platforms, cost, reliability, governance. Blueprint fidelity helps remediation mapping.

Hands-On Exercises

Exercise: Full mock assessment weekend

Problem statement. Sit the 150-minute cadence using expanded items (samples plus at least ten sibling items you author), score with keys, classify misses, and produce a dated remediation plan. Include three architecture prompts scored on the 0–5 rubric.

Tasks.

1. Author sibling items so Blocks A/B/C meet counts (20/20/15).
2. Sit timed blocks with confidence markers.
3. Score and compute weighted overall and band.
4. Score architectures with peer if possible.
5. Write remediation tasks mapped to labs/drills.
6. Schedule parallel retake if band < go.
7. File ASSESS.md and a one-page readiness recommendation for capstone.
8. List three refusals written from memory in under two minutes as warm-up evidence.

Expected deliverables. Item set, timed answer sheets, ASSESS.md, remediation plan, readiness recommendation, refusal warm-up.

Evaluation rubric.

Exercise: Mini quiz lab (answers included)

Sit in 12 minutes, then check.

M1. Idempotent merge key for completion attempts should be:
A. Row number on insert
B. (attempt_id, definition_version)
C. Dashboard refresh id
D. Random UUID each run

Answer: B.

M2. Dry-run bytes exist to:
A. Decorate notebooks
B. Measure scan cost drivers before running heavy queries
C. Replace SLOs
D. Prove exactly-once

Answer: B.

M3. Consumer lag on two partitions only suggests:
A. Global broker death always
B. Key hot spots or stuck poisons on those partitions
C. Need for hourly lake partitions
D. Dropping uniqueness checks

Answer: B.

M4. Green DAG with wrong board number implies:
A. Airflow is useless
B. Missing data-level checks
C. Spark must be abandoned
D. Cost sheet is wrong by definition

Answer: B.

M5. Combined unlabeled provisional+certified tile:
A. Recommended UX
B. Assessment fail pattern
C. Required for Kafka
D. Fixes skew

Answer: B.

If you miss any, remediate the matching domain before claiming go.

Exercise: Distractor autopsy worksheet

Pick ten items you got right and ten you got wrong across mocks. For each, write the anti-pattern each distractor encodes. Submit a table:

This exercise converts a key into a study guide. Learners who only check "B was right" forget by Thursday. Learners who name traps retain them in architecture prompts.

Exercise: Architecture outline drill (30 minutes, three prompts)

Using only bullet outlines (no paragraphs), answer prompts A–C from Technical Explanation in ten minutes each. Score outlines with a reduced rubric: grain present? failure modes? refusal? meter/SLI? Each yes/no worth 1 point (max 4). Goal: ≥3 on every outline before you write prose. This drill fixes the empty-third-prompt failure mode.

Answer key index for this lesson's samples

Do not mistake this index for a complete exam pool. Sibling authoring remains mandatory for a full-length mock.

Study group protocol (three people, two hours)

1. Separate rooms or muted calls for a 40-minute Block B mini (10 sibling items each authored beforehand and shuffled).
2. Swap keys; score.
3. Each person teaches one miss for five minutes max.
4. Joint architecture prompt (Northline) with three colors of pen—one per person—on a shared diagram, fifteen minutes.
5. Rubric debate ten minutes; update personal refusal cards.

Groups that only discuss feelings waste the slot. Groups that produce scored artifacts win.

When to stop studying and start capstone

Stop when:

• Two parallel mocks land in go or conditional-with-finished-remediation
• Architecture average ≥3.5 on timed prompts
• Lab C and Lab D proofs still exist and can be demoed in five minutes
• You can recite three refusals cold

Continuing to collect MCQ percentage points past that yields diminishing returns versus building the capstone's real publish path.

Sample MCQ set 6 — Mixed rapid-fire (with answers)

28. Freshness SLI should be defined on:

A. CPU percent of Spark workers
B. User-visible readiness of the certified mart by the promised clock
C. Number of Git commits
D. Kafka broker disk only

Answer: B.

29. AQE skew join, when available, still requires:

A. No verification
B. Correctness checks on aggregates and a glance at stage metrics
C. Turning off all other optimizations forever
D. Dropping definition_version

Answer: B.

30. The Monday board query should project:

A. SELECT * for future-proofing
B. Only columns the board needs, with prune-friendly filters
C. Raw JSON strings for flexibility
D. Every region’s replica

Answer: B.

31. Workload isolation for finance versus portal means:

A. Two divergent marts
B. Separate pools/reservations or schedules on one certified grain
C. Shared unconstrained slots always
D. Disabling RLS

Answer: B.

32. A take-home that omits duplicate handling should be scored:

A. Full marks if charts are pretty
B. Down on correctness regardless of README eloquence
C. As a pass if Kafka is mentioned
D. Only on font choice

Answer: B.

These five close common last-mile gaps: SLI honesty, AQE verification, projection, isolation without dual publishers, and take-home grading aligned to labs.

One-page personal crib (allowed only if your mock mode permits notes)

textCopy
Grain: attempt + definition_version
Paths: certified batch vs provisional stream (labeled)
Checks: unique, version, null, freshness, volume policy
Commit: sink then offsets; DLQ poisons
Skew: measure stage; salt/split; verify totals
Cost: dry-run bytes; H vs H_idle; label illustrative $
Refuse: 2nd publisher; raw board; unlabeled mix; pause w/o resume
SLO: 06:00 certified; error budget explicit

If you cannot recreate this crib from memory on D-1, you are not ready—even with a high practice percentage.

Mapping lesson 66 HANDOFF.md into ASSESS remediation

If you authored HANDOFF.md in the practical exercises lesson, paste its "labs below 70%" list into the D-7 plan before writing new sibling items. Assessment prep that ignores fresh lab weakness repeats the same Block B failure. Conversely, if HANDOFF shows Lab F weak, author three cost-meter siblings before Block C. The course is a pipeline: interview drills → labs → timed assessment → capstone. Skipping a stage shows up as thrash in the next.

Examiner script for peer-administered mocks

Peer says only:

1. "Timer started."
2. "Five minutes left in this block."
3. "Pencils down / hands off keyboard."
4. After scoring: "Classify each miss; do not debate the key until classifications are written."

Peers who coach mid-block invalidate the band. Save coaching for remediation hours.

Final readiness sentence

Write one sentence at the bottom of ASSESS.md:

"I am [go / conditional go / hold] for the TechLearningPro capstone as of [date] because [evidence]."

Examples: "conditional go because Block B is 68% and Lab C crash demo is scheduled tomorrow." Vague optimism is not a sentence this lesson accepts.

Lesson Summary

Assessment readiness is a band with evidence, not a feeling after rereading. This lesson defined a capstone-oriented exam: three MCQ blocks, architecture prompts, weights, timeboxes, self-assessment protocol, and sample items with answers. Distractors encode the course's anti-patterns—second marts, reckless pauses, auto-commit, dropped checks—so studying the quiz also rehearses production judgment.

Take your ASSESS.md band into the Enterprise Data Platform Capstone Project. Remediate first if the band says hold or conditional. The Monday board still needs a single publisher, a pinned definition, and a 06:00 honesty you can defend under a timer—the same timer a hiring loop and an on-call night both use.

Further Reading

• Course practice spine: Data Engineering Interview Questions and Answers, Data Engineering Practical Exercises.
• Data Reliability and SLA/SLO Design for error-budget language in prompts.
• Data Platform Cost and Performance Optimization for meters appearing in Block C.
• Spark Performance Optimization and Kafka Producers and Consumers for Block B remediation.
• Retrieval practice research summaries from reputable learning-science sources of your choice; apply the schedule table rather than debating citations during exam week.
• Your ASSESS.md history. It is the authoritative readiness document.