DodgeScore - An Earnings Call Q&A Analyzer

Every quarter, analysts and retail investors have to rewatch an hour-long earnings call or comb through raw transcripts just to answer one simple question: did management actually answer what investors asked, or did they circle around it with jargon and numbers that never touch the actual question? The problem is this judgment is subjective — two people can read the same answer and disagree on whether it was a "dodge" or a complete response, because there's no objective standard to measure it against. Yet the dodging pattern repeats predictably: executives get asked for concrete numbers (guidance, margins, timelines), and respond with long-term strategy narratives that never touch the number requested. DodgeScore turns this manual, subjective process into a measurable pipeline. Gemma (via Fireworks dedicated server or serverless API) decomposes each investor question into concrete sub-questions and classifies whether the executive's answer actually addressed what was asked. But the final scores DodgeScore and PrecisionScore are computed by deterministic Python, not the model, so anyone can recompute the result by hand from the transcript at any time. The entire pipeline runs on AMD Developer Cloud, showing how AMD's infrastructure supports auditable, scalable real-time financial analysis.