Debate feedback is useful only when it explains why a side lost and how that side can revise to become more competitive. However, LLM-generated feedback often sounds polished while remaining generic and vague. We introduce ReFeedGraph, a graph-based method that turns prior debate feedback into reusable principles. It collects multi-framework judge feedback, extracts weaknesses and actionable suggestions, abstracts them into principles, and stores them in a graph. For a new debate, ReFeedGraph retrieves relevant principles and adapts them into targeted revision feedback. We evaluate the generated feedback by recording whether it improves the weaker side after rewriting. Across three existing datasets (DebateArt, Intelligence Squared, and British Parliamentary debates), ReFeedGraph outperforms direct LLM and structured-prompt baselines, producing more specific and actionable feedback that leads to winning more debates.

Argument Quality Detection is an emerging field in NLP which has seen significant recent development. However, existing datasets in this field suffer from a lack of quality, quantity and diversity of topics and arguments, specifically the presence of vague arguments that are not persuasive in nature. In this paper, we leverage a combined experience of 10+ years of Parliamentary Debating to create a dataset that covers significantly more topics and has a wide range of sources to capture more diversity of opinion. With 34,890 high-quality argument-analysis pairs (a term we introduce in this paper), this is also the largest dataset of its kind to our knowledge. In addition to this contribution, we introduce an innovative argument scoring system based on instance-level annotator reliability and propose a quantitative model of scoring the relevance of arguments to a range of topics.