Openbat Bestpractices Chatbot

Field-tested best practices and fundamental truths for making data/analytics LLM chatbots and agents over SQL, tools, or structured data reliable and hallucination-resistant. Reach for it when building, debugging, reviewing, or hardening such a bot whose answers can't be trusted — it makes up numbers (amounts, counts, percentages, averages, growth rates), invents competitors/entities/plans/terms not in the data, gets rankings backwards, pins a real number to the wrong entity or metric, returns empty answers, or leaks tool names — even if the user never says "hallucination" (they may just say it "makes stuff up" or "feels off"). Also use for design calls — how to shape wha…