Some very early documents from FlameSpotting, an LLM pipeline designed to help moderators catch flamewars and harassment in online chat. To be updated soon with finalized prompts, experiment results, and scripts. (Medium article to be published soon with explainer, too!)