Most people study by reading. They go through their notes, highlight things, maybe summarize a chapter. Then they close the book feeling ready. Then they sit in the exam, stare at a blank page, and realize something is very wrong.
The problem is not intelligence or effort. It is a well-documented cognitive phenomenon: recognition and generation are two completely separate skills. You can recognize a correct answer when you see it and still be completely unable to produce it from scratch under time pressure. Exams, unfortunately, only test the second one.
I started thinking about this seriously after bombing a topic I was sure I had "covered." I had read through every relevant chapter. I had watched the lectures. Everything felt familiar. But when the exam question appeared, I could not generate a coherent answer without cues. That failure sent me looking for something better than re-reading and hoping for the best.
That is what led me to the Minimum Viable Clue Prompt Pack, and what I want to share here is what actually happened when I used it.
What the Pack Is and Why It Exists as a System
The Minimum Viable Clue (MVC) Method is built on a single cognitive insight: the gap between what you recognize and what you can independently generate is almost always much wider than you think it is. The pack is designed to measure that gap precisely, simulate real exam conditions, and close the gap systematically.
It contains three prompts, each addressing a distinct phase of preparation:
- Minimum Viable Clue Knowledge Audit — the diagnostic engine
- Full Mock Exam Generator — the simulation tool
- Generative Drilling Session — the reinforcement loop
What makes this work as a system rather than three separate tools is the sequential logic. You cannot effectively drill what you have not diagnosed. You cannot self-assess your mock exam without a proper mark scheme. Each prompt feeds directly into the next one, which means the output of Phase 1 becomes the input configuration for Phase 3. That chain is where the real value sits.
Running a single prompt in isolation gives you something useful. Running all three in order gives you an exam preparation loop that mirrors how actual examiners think about assessment: start with diagnosis, move to simulation, reinforce the weak spots, repeat.
You can explore the full system at appliedaihub.org/prompts/minimum-viable-clue/.
The Knowledge Audit: Where It Gets Uncomfortable
The first prompt is the one that changes how you think about studying. I ran it on an Economics topic I had spent about four hours reviewing the previous week. I set Subject: Economics, Topic: Comparative Advantage and Trade Theory, and QuestionCount: 8.
The AI then asked me a question like: "What happens to global output when countries specialize in what they do relatively better?"
Notice what that question is not doing. It is not asking me to define comparative advantage directly. It is not offering four options to choose from. It is giving me the minimum clue — just enough framing to make the question fair — and then waiting for me to produce the answer from memory.
That design choice is the entire point. The prompt's system instruction tells the AI to act as a rigorous academic examiner "specialising in diagnosing the gap between recognition memory and genuine generative knowledge." The question calibration rule is explicit: a bad question gives the answer away in the phrasing; a good question gives just enough orientation without enabling passive recall to substitute for generation.
After each of my answers, the AI produced three things: a full model answer, a checklist of exactly which points I had included versus missed (with ✅ / ❌ markers), and a running sub-score for that question. By question five, I could already see the shape of my actual knowledge versus my perceived knowledge. The final output was an Audit Report with a Generative Accuracy Score and a structured breakdown of concepts I could truly generate versus concepts I only recognized.
My score on that Economics topic was 61%. I thought I was ready. I was not.
That number is not demoralizing in a useless way — it is diagnostic in a useful one. The report told me exactly which concepts were solid and which ones were sitting in the dangerous zone where I could recognize the right answer but not write one under pressure. Cognitive science research on retrieval practice, including work by Roediger and Karpicke published in Psychological Science, has consistently shown that active recall outperforms passive re-reading for long-term retention. The MVC audit operationalizes that finding into a score you can actually act on.
The Mock Exam Generator: Practicing Like an Examiner
Once you have your audit results, the second prompt lets you simulate a real exam on the same topic. I configured it with Subject: Economics, Topic: Comparative Advantage, QuestionFormat: Mixed (exam-style), QuestionCount: 10, and DifficultyLevel: Intermediate.
The output structure is worth noting. Section A is a properly formatted exam paper, complete with mark allocations per question, an exam header with total marks and recommended time, and question types calibrated to the difficulty setting I chose. No answers, no hints — exactly what a student would receive in an exam hall.
Section B follows immediately: the full official mark scheme, point-by-point mark allocations for every question, distractor explanations for multiple-choice items, and a grade boundary table scaled to the marks of the paper generated.
That combination is what makes this prompt genuinely useful rather than just a curiosity. Self-grading your own practice answers without a mark scheme is guesswork. With the mark scheme and grade boundaries, you are running the same evaluation process an examiner would run. You know not just whether your answer was correct but which specific points earned marks and which ones did not.
The prompt's role instruction tells the AI to act as "a professional examiner with 20 years of experience setting and marking high-stakes academic papers." That framing matters because it pushes the output toward examiner-calibrated language rather than generic quiz content. Difficulty ramp calibration is explicit: Foundation questions are single-step recall; Advanced questions require synthesis and evaluation; Exam-simulation distributes approximately 30% recall, 50% application, and 20% evaluation.
The Drilling Session: Closing the Gap With Escalating Pressure
The third prompt is where the actual remediation happens. I took the weak concepts identified in my audit report and fed them directly into the Drilling Session as my WeakConceptsList. I set SessionRounds: 8 and DifficultyMode: STANDARD for the first pass.
STANDARD mode still uses minimum-clue framing. HARD mode gives only a category label: "Mechanism — go." BRUTAL mode provides zero clue at all and simply says "Concept [number] — explain it fully." That escalating ladder is not arbitrary. It mirrors the actual conditions you face in different exam formats, where sometimes context appears in the question and sometimes it does not.
The session tracks per-concept performance across rounds. Concepts that score below 70% get repeated more frequently. After all rounds, the AI generates a Drilling Session Summary with a per-concept progress table, most-improved concept, still-at-risk list, and specific instructions for what the next session should prioritize.
After two drilling sessions on my weak Economics concepts, I ran the Knowledge Audit again. My Generative Accuracy Score on the same topic moved from 61% to 79%. That improvement came not from re-reading my notes but from being forced to reconstruct the material from near-zero cues multiple times in sequence.
What This System Is Actually Good For (and What to Expect)
A few things worth knowing before you go in:
The prompts work best with frontier models. GPT-4o, Claude 3.5 Sonnet, Gemini 1.5 Pro, and DeepSeek-R1 all handle the role instructions and sequential output control reliably. Smaller or older models can produce the Knowledge Audit adequately, but the precision of the mark scheme and the consistency of the session tracking degrades noticeably. Use a capable model.
The subject coverage is broad. Built-in variable presets include Biology, History, Economics, Chemistry, Mathematics, Physics, Psychology, and Law. The Topic variable is free-form, which means you can run it on any specific syllabus area within those subjects, or any other discipline where written generation is assessed.
The system is designed for high-stakes formats. AP, IB, and A-Level board exams, university midterms and finals, and professional licensing exams (Bar, CPA, Medical, CFA) all share the common denominator that generation under time pressure determines your grade. If recognition is sufficient — open-book exams, multiple-choice only with no wrong-answer penalty — the system still helps, but the diagnostic pressure is lower.
The 70% threshold is meaningful. A Generative Accuracy Score below 70% on a topic consistently correlates with exam risk regardless of how prepared you feel subjectively. That number is worth taking seriously.
Importing the Full Pack Into Your Workflow
You can copy-paste these prompts directly from the landing page into ChatGPT or Claude and they will work. But if you are using all three prompts across multiple subjects and topics, doing that manually adds friction and variable-filling errors.
The better approach is to import the full Prompt Pack directly into Prompt Vault, the free browser-based prompt manager from AppliedAI Hub. The pack includes a ready-to-import prompts.json file. You open Prompt Vault, click Import, select the file, and all three MVC templates appear instantly with fillable dropdown menus for Subject, Topic, QuestionCount, DifficultyLevel, and every other variable.
No account required. No installation. The variable dropdowns eliminate the risk of formatting errors when configuring sessions, which matters more than it sounds when you are trying to run a consistent protocol across a multi-week exam prep schedule.
Whether This Is Worth Your Time
The MVC Method is not a productivity hack and it is not a shortcut. It is a more rigorous version of what good students already do instinctively: test themselves, identify gaps, and target practice specifically at what is weakest. The difference is that it systematizes and measures that process in a way that removes self-delusion.
If you have an exam that tests written generation — which is most exams worth taking seriously — and you currently prepare primarily by reading and reviewing, this system will show you that your actual readiness is lower than you think. That is uncomfortable but useful information to have before the exam rather than during it.
The full pack, including the 35-page PDF guidebook covering the cognitive science behind the method and subject-specific calibration guidance, is available at appliedaihub.org/prompts/minimum-viable-clue/. The PDF explains the architecture in enough depth that you could adapt the prompts for specialized exam formats if needed.
If you already have the pack, open Prompt Vault and import it now. Run the Knowledge Audit on one topic you are confident about. See what your actual Generative Accuracy Score comes back as.
The result may be more informative than any amount of additional re-reading would be.
Top comments (0)