architecture deep-dive · reply arena pipeline
Each reply is scored by Claude Opus across three dimensions. The composite score is a weighted average.
Evidence of real building. Shipped projects, GitHub repos, live demos, technical depth. Did they make something or just talk about it?
Originality of approach. Novel pitches, unexpected angles, creative formats. Standing out from "I'm passionate about AI" noise.
Personality and memorable factor. Humor, bold takes, weird flex. The replies that make you stop scrolling.
COMPOSITE = (Builder × 0.40) + (Creativity × 0.35) + (Quirkiness × 0.25)
The entire pipeline — from fetching replies to scoring to deployment — is orchestrated through Claude Code with custom skills and plugins. No traditional backend. No database. Just structured JSON files and static HTML.
from first prompt to live leaderboard