data

# Data Studio Deliver an answer the user can inspect, revise and rerun with their next data file. Do not stop at a starter screenshot or a watch-only visualization. Read [the analysis contract](references/analysis-contract.md) before changing a schema, query, join, unit, missing-value rule or chart. ## Understand the question Use the user's files and known intent. Ask only for materially missing facts: what decision/question, field meanings, units/currency, period boundaries, join keys, duplicates and missing-data treatment. Never ask again for answers already supplied. If example data was approved, keep example=true and make the synthetic-data label visible in the UI and every report. Inspect a bounded sample and counts. Preserve original CSV text; map columns explicitly in analysis.json. IDs that look numeric remain text. ISO dates are strictly parsed. For money, use decimal with a declared scale and unit: values such as 12.30 become integer cents, with no implicit rounding. ## Build the useful workflow 1. Write the question, method, source mappings, keys, joins, controls and limitations in analysis.json. Add readable queries/*.sql and checks/*.sql. 2. Make the queries answer the actual question. Rates come from matched totals, not sums/averages of ratios. Missing remains missing unless the question explicitly defines an absent record as zero. Do not infer causality. 3. Add source-record links and a matching detail query. Check the detail rows reconcile to...