Cert Drills Aws Cfa Bar Item Writer Prompts

A generic "make me a study plan" prompt produces a calendar. It does not produce the thing that actually moves a score: items that look like the exam, with stems, options, and a rationale you can argue with.

If you are sitting AWS, CFA, or a bar exam, you already have a content outline. The missing piece is a drill factory. You want a prompt that writes items in the exam's voice, then a second prompt that grades your answer the way a picky item writer would. Matching cert-drill prompts are landing in the PromptDig library (Browse more prompts). Use them as a shell. Paste your own outline in.

This is not a substitute for the official bank. It is a way to practice the shape of the question when you have twenty minutes and a domain you keep missing.

Steal the item shape, not a topic list

Every serious exam has a house style. AWS items are short scenarios with one best action. CFA items mix vignettes and standalones, with distractors that are almost right. Bar items (MBE-style) hang on a rule, a twist, and four answers that all sound lawyerly.

Your prompt has to name that shape. "Write practice questions about S3" will give you trivia. "Write a four-option single-best-answer item testing when to use S3 Object Lock in a compliance scenario" will give you a drill.

A skeleton that works across the three exams:

You are an item writer for [exam], not a tutor.
Write ONE item.
Stem: a short scenario. No trivia. Test [objective from my outline].
Options: A-D, similar length, one best answer, three plausible distractors.
After the item, wait. Do not reveal the key until I answer.
When I answer, give: correct key, why the key is best, why each distractor fails, the outline citation I should review.
Do not invent official item IDs, pass rates, or "this is a real exam question."

The "wait" line matters. If the model dumps the key under the stem, you will peek. Make the drill a two-turn conversation.

AWS: scenario, constraint, one action

AWS items punish laundry lists. The stem should include a constraint that knocks out two of the four services people always guess: cost, a region, an existing VPC, a compliance hold, a team that cannot rewrite the app.

Feed the prompt a single task from the exam guide, not a whole domain. "Design a highly available architecture" is a study plan. "Choose a storage class for a 30-day backup that must be restorable in minutes" is an item.

Ban marketing language in the output. If the explanation says a service is "robust and scalable," the item writer prompt failed. Demand the tradeoff: latency, blast radius, operational load, or a limit you actually have to remember.

Keep a miss log. After five items, paste the miss log into a second prompt: "These are the distractors I keep picking. Write the next item so that the tempting wrong answer is the one I usually choose, then explain the distinction in one paragraph." That is studying. A 90-day calendar is not.

CFA: vignette first, then the question

CFA drills go wrong when the model writes a textbook paragraph and slaps a question on the end. Real vignettes bury the number you need in a fact set. Your prompt should require a fact table (rates, holdings, dates) and then a question that uses one of those facts.

Tell the model which level and which reading family you are in. "Level II, equity valuation" is usable. "CFA finance" is not. If you paste a formula sheet, say "you may use only these formulas." Otherwise it will mix frameworks.

Distractors should be the classic CFA kind: right formula, wrong input; right input, annual vs. periodic mixup; a number from the vignette that is irrelevant. Ask for that explicitly. If you do not, you get four random percentages.

Bar: rule, twist, no fan fiction

MBE-style items need a governing rule and a fact that triggers an exception or a cross-claim. Your prompt should name the subject (Evidence, Torts, Civ Pro) and the issue, then forbid extra facts that turn the stem into a novel.

A second skeleton for the grade pass:

I will paste an MBE-style stem and my answer.
Grade like an item writer. Do not coach first.
1) State the issue in one line.
2) State the governing rule in one line. If you are not sure of the rule, say UNSURE. Do not invent a restatement citation.
3) Apply the facts to the rule.
4) Why my choice is right or wrong.
5) The one fact that, if changed, would flip the key.

"UNSURE" is load-bearing. Models like to sound like a supplement. You want a drill partner that will stop when it is guessing. You still check the rule in your own outline or a licensed Qbank. The prompt is for reps, not for authority.

Run drills, then retire the prompt that started moralizing

A drill prompt that starts with "believe in yourself" is a study-plan prompt in disguise. Cut the pep talk. Cut "as an AI." Cut fake statistics about pass rates.

Work in sets of five. One objective per set. When you can explain the distractor you used to love, switch objectives. Save the item-writer prompt and the grade prompt as a pair. If you want that pair to survive the next exam cycle, share it (Share a prompt) with the exam name in the title so nobody pastes an AWS stem template into a CFA vignette slot.

The outline tells you what to cover. The drills tell you whether you can choose under time. Build the second thing.