Prompt comparison playground

The setup

• Run the same task with one variable changed and watch the outputs diverge.
• Each preset isolates one lever — phrasing, framing, attached context, skills, model, or reasoning prompt.

What moves the output

• Phrasing and framing. Loaded questions get loaded answers.
• Signal-to-noise. A single relevant document beats burying it in a novel.
• Skills. A saved instruction set turns "ask nicely and hope" into a repeatable, reviewable process.
• Reasoning and model choice. Chain-of-thought prompts and stronger models materially shift accuracy on reasoning-heavy tasks.