ORACLE

Workshop · 4 of 7 · Loading context into AI

Most prompts fail before the model reads a word of them — they fail at what's missing. ORACLE is the checklist for what a working request contains. Six slots. Fill them and the same model, the same day, gives you a different class of answer.

• OObjective. What does done look like? The outcome, not the topic. “Research our competitors” is a topic. “A one-page comparison of our three closest competitors, ending in a recommendation I can defend” is an objective.
• RRole. Who should the model be? An experienced analyst writes differently than a friendly explainer. Say which one you're hiring.
• AArchitecture. How should it work, and what shape should the answer take? Steps to follow, sections to produce, a table, a memo, a decision with reasons.
• CContent. What material does it run on? The documents, the data, the constraints — and your examples. An example of what “good” looks like is worth a paragraph of adjectives.
• LLeverage. What multiplies it — the right model for the job, the tools it may use, the skills or frameworks it should apply?
• EEvaluation. How will you judge the result, and how will you correct it? Say what you'll check. The model that knows the test writes to the test.

Fill the slots in any sequence that suits you; just make sure, before you press enter, that all six are on the desk — and that the question you actually want answered is the freshest thing on it.