Ask a research question the model cannot answer from vibes
Lesson 1 of 6 · Tax research that survives review
1 · Learn the move · RTO. Role · Task · Output
'Is a home office deductible?' gets an answer from the general shape of the internet, which is where the model learned to sound like a tax professional. The question that survives review names the taxpayer type, the facts the answer turns on, the year, and the output you need: the authorities to open, in order, each marked VERIFY, and the open questions the facts leave. Role: a researcher who lists what to read, not an advisor who concludes. Task: the de-identified facts and the single question. Output: a reading list with the provision, why it matters, and what fact it turns on; then the facts you have not given it that would change the answer. A model that is told to list authorities and mark them VERIFY produces something you can check in an afternoon. A model that is asked 'is this deductible' produces something you can only believe.
You are a tax research assistant who lists what to read; you do not conclude. Facts, de-identified: [paste]. Year: [year]. Question: [one sentence]. Output: (1) the authorities to open, in order of weight, each as: provision | why it matters here | the fact it turns on | VERIFY; (2) the facts I have not given you that would change the reading; (3) the question restated as narrowly as the facts allow. Never state a threshold, a rate or a dollar limit; write [NEEDS: the figure, with source and year]. Never say what the answer is.
2 · Your turn. You write the prompt
Taxpayer A, a self-employed consultant, uses a room in her home for client calls three days a week and as a guest room the rest. She asks whether she can deduct it. The staff accountant's first prompt was 'can my client deduct a home office she also uses as a guest room' and the answer came back fluent, confident and uncited.
Write the prompt for this scenario as given, or swap in a real one from your own work. The score works the same either way.
Remember: the AI sees only your prompt, not this page. If the situation isn't in your prompt, it doesn't exist.
Optional. These shape the output when you run your prompt below, not your score.