rnd-data-sciencelisted

# R&D Data Science ## Overview Analytical work fails silently. A wrong number in a table looks identical to a correct one. A chart with the wrong axis scale misleads without warning. **Core principle:** Every numerical result must be independently verifiable. Show your computation, not just your conclusion. ## When to Use - Financial calculations (P&L, cash flow, ratios, projections) - CSV or XLS data ingestion, transformation, or export - Chart and visualization generation - Statistical analysis, aggregations, or derived metrics - Data wiring between sources (APIs, files, databases) - Insight extraction from structured or tabular data - Any task where the output is numbers, tables, or charts ## The Iron Laws ``` 1. LOAD COMPUTATION TOOLS (JULIA OR DUCKDB) BEFORE ANY COMPUTATION 2. VERIFY EVERY NUMERICAL RESULT WITH AN INDEPENDENT CHECK 3. VALIDATE INPUT DATA BEFORE PROCESSING — GARBAGE IN, GARBAGE OUT 4. NEVER HARDCODE INTERMEDIATE VALUES — RECOMPUTE FROM SOURCE 5. DOCUMENT UNITS, CURRENCY, AND TIME ZONES EXPLICITLY ``` ## Tool Selection Choose the right tool before starting. Use the decision table below: | Task type | Preferred tool | Reason | |---|---|---| | SQL-expressible queries, aggregations, GROUP BY | DuckDB | Native SQL, fast, zero boilerplate | | Joins across multiple CSV/Parquet files | DuckDB | Join syntax is concise; Julia merge is verbose | | Data filtering, WHERE clauses, data exploration | DuckDB | Interactive CLI; no session state needed | | Parque