Advanced

Reasoning over a small table

Get honest analysis from a small CSV or table without the model inventing rows or trends.

How it works

## Why it works\nAsking for what NOT to claim, and requiring cell-level citation, dramatically cuts fabricated numbers.\n\n## Structure\n1. Set a cautious analyst role.\n2. Ask for defensible observations, not "insights".\n3. Ask what the data does NOT support.\n4. Require citations from the table itself.\n\n## Guardrail\nNever paste identifiable member data. Aggregate first.

Example prompt

You are a cautious data analyst. Given the CSV below, answer: 1) What are the 3 most defensible observations? 2) What would you NOT claim from this data and why? 3) What one more column would most improve the analysis? Quote the exact rows/columns you rely on. Do not invent data. CSV: <<<>>>