How this exam works & how to attack every question
65 questions, 170 minutes (~2.6 min each), pass at 720/1000 (~72%). Four domains: data prep (28%), model development (26%), deployment (22%), monitoring & security (24%). Almost every question is a short business scenario plus four options, and it is a pattern-matching exam: scenarios contain trigger phrases that map to answers. This guide teaches the patterns.
The 5-step procedure, run it on every single question
- 1. Read the LAST sentence first. It holds the grading rule: "LEAST operational overhead," "MOST cost-effective," "which metric." Know the game before reading the story.
- 2. Name the topic in three words. "Production drift monitoring." "Scheduled batch ETL." Every question belongs to one of ~10 families, and naming the family loads the right chapter of your brain.
- 3. Extract the hard facts. Underline numbers (payload sizes, durations, frequencies) and absolutes ("no code," "no internet," "same framework").
- 4. KILL options, don't pick one. Hunt for reasons to eliminate: violates a numeric limit, wrong lifecycle phase, wrong tool for the verb, builds what AWS already sells. Two options usually die instantly.
- 5. Between two finalists: fewest moving parts wins (when the rule is least effort, overhead, or cost).
Three universal tie-breakers. (1) Never build what AWS sells: a purpose-built managed feature beats any Lambda or Step Functions assembly of the same thing. (2) Match the lifecycle phase: training-time tools vs production tools, because half of all traps are the right tool in the wrong era. (3) Match the verb: collect, track, store, and document are all different, and each verb has exactly one service whose job description contains it.