I wanted to know which kind of business context actually moves Genie's accuracy, so I ran a small benchmark on samples.wanderbricks: 30 questions (15 scenarios in 2 phrasings), each with ground-truth SQL, adding one layer of context at a time in a ne...
