idea-gen

Idea generation

Start from an evidenced limitation, not from an unconstrained technique. State the affected setting, why existing approaches fail there, and what observation would distinguish a real improvement from a plausible story.

Propose one idea at a time. Define:

• the mechanism being changed;
• the closest prior approach and the material difference;
• the expected observation;
• the assumption most likely to fail;
• the smallest experiment that can falsify the idea;
• implementation cost and likely confounders.

Reject ideas whose novelty depends only on renaming, combining unrelated components, or assuming an unavailable dataset or metric. Return one primary idea and briefly record rejected candidates so they are not repeated without new evidence.