Every idea. One place.

Each average uses the six actual scores given to that idea. The combined order uses the unrounded average, then higher median, higher lowest score, and idea ID. A dash means that reviewer did not participate in that round; it is never counted as zero. “Over 75” is strictly greater than 75.

The first two rounds used a broad gameplay rubric. The look-and-feel round emphasized visible quality, breadth and payoff for effort. Their original scores are preserved here; the combined order is a browsing aid, and the round filter keeps those different briefs visible.

• Original concepts: GLM, Luna, Sol, Opus, Gemini and Codex.
• Model brainstorm and look & feel: DeepSeek, Luna, Sol, Opus, Gemini and Codex.
• Self-ratings are included. Codex added its original scores after seeing that panel, and recorded its later scores before reading peer ratings. The exercise was not fully blind.
• DeepSeek used its direct API in the later rounds; Gemini and the original GLM review used OpenRouter, and Opus used Claude. These are source-based proposal ratings, not proof of implementation or measured player benefit.