I Asked ChatGPT to Analyze 3 Datasets. It Made the Same Mistakes Every Time

An experiment tested GPT-5.6 Terra and Luna on three small datasets with typical business questions. The AI made repeated mistakes, including miscounting rows and reversing conclusions, even after self-review prompts.