Welcome to the board. Three dials in front of you — Role, Task, Format. That's RTF, one of the classic prompt frameworks. Role is who Claude should be. Task is what you want. Format is how it comes back.
There are a lot of these frameworks — CRAFT, RISEN, RACE, APE. They all do the same job: a checklist so you don't forget to tell the AI something that matters.
Here's the test. Turn the dials. Hit record. Read what comes back. Then try a different mix. Somewhere on this board there's a combination that sounds clean. There are also mixes where the track falls apart. Find both. That's the lesson.
On authenticity: Each output is what Claude produces for the exact prompt shown. Copy the prompt and test it yourself — exact wording will vary slightly run-to-run, but the structural pattern (bloat, placeholders, pompous voice, over-formatting) is consistent.
RTF — and every other framework — is a checklist that helps you remember the pillars. But each pillar has a ceiling. Turn Role too high and Claude opens with credentials and technical jargon. Turn Task too high and it tries to cover everything you mentioned — history, formulas, edge cases — instead of doing the one thing you needed. Turn Format too high and you get headers, bullets, quizzes, and callouts whether the content warrants them or not.
The ladder metaphor: frameworks help you climb out of vague-prompt territory, but they don't tell you where to stop. That's your job — and it's why testing matters.
The sweet spot on this board: R2 · T2 · F2. A role specified clearly (not stacked with credentials). A task targeted at a specific audience (not sprawling). One format cue (not a whole production template).
Try the extremes too. R3 · T3 · F3 is a 600-word credentialed lecture with emojis, a quiz, and a fake Einstein quote. Useful only as a cautionary tale.