| I created a dataset which contains forecast data which therefore can't be created retrospectively. For ~38 days, a cronjob generated daily forecasts: - 10-day horizons - ~30 predictions/day (different stocks across multiple sectors) - Fixed prompt and parameters Each run logs: - Predicted price - Natural-language rationale - Sentiment - Self-reported confidence I used stock predictions as the forecast subject, but this is not a trading system or financial advice, it's an EXPERIMENT! Even though currently I didn't find something mind-blowing, visualizing the data reveals patterns I find interesting. Currently, I just plotted trend, model bias, and ECE - more will come soon. Maybe you also find it interesting. The dataset isn't quite big, so I'm actually building a second one which is bigger with the Gemini Flash and Gemini Flash-Lite model. PS: If you are interested in the dataset or the MVP with a dashboard to crawl data quickly, just mention it in the comments. [link] [comments] |