The 0.8 Cosine Similarity Threshold

Ideas with cosine similarity ≥ 0.8 can be treated as functionally identical for brainstorming purposes.

This empirically-derived threshold was established through testing and provides a practical cutoff for automated de-duplication. Below 0.8, ideas are sufficiently different to warrant separate consideration. Above 0.8, the differences are typically superficial, same core concept expressed differently.

Example at 0.82 similarity (considered identical):

• “MiniMend Sewing Kit: A compact, travel-sized sewing kit with pre-threaded needles…”
• “QuickFix Clothing Repair Kit: A compact kit with needles, thread, buttons…”

Example at 0.36 similarity (considered distinct):

• A battery-powered heating mug
• An LED desk lamp with Pomodoro timer

The threshold enables automated measurement of idea diversity at scale, though it has limitations: it doesn’t consider similarity to ideas that already exist in the world, and isn’t perfectly calibrated to human judgments of semantic difference.

Related:, 05-atom—idea-space-as-landscape