Buy Me a Coffee

Five Checks to Validate AI-Generated SQL

This technical guide explains five quick checks to verify results produced by AI-generated SQL queries. It warns that a running query only proves syntactic correctness and outlines practical tests: (1) compare row counts before and after joins to detect fan-out, (2) watch for NULLs breaking NOT IN filters (use NOT EXISTS), (3) ensure filters sit in WHERE vs HAVING appropriately, (4) confirm the denominator used by averages or percentages, and (5) ask the AI to read the query back clause-by-clause. The author cites the BIRD benchmark (Li et al., 2023) showing large gaps between model and human execution accuracy and provides examples and remediation patterns to avoid incorrect analytics numbers.

How to Build a Tableau Dashboard and Story

Step-by-step tutorial showing how to create a published Tableau dashboard and a three-point narrative story from a real dataset. The guide uses the Telco Customer Churn dataset (7,043 customers) and a public GitHub repo for data-shaping code. It stresses shaping data upstream (one row per entity, 1/0 outcome column, readable names, ordered buckets), creating a single calculated field for rates (Churn Rate = AVG([Churned])), building four focused worksheets (one point each), assembling them into a dashboard, and sequencing three story points (problem, mechanism, action). It explains Tableau Public publishing requirements (workbooks must use extracts) and gives practical UI steps and common error fixes. The guide also covers visual rules (one-accent color, bar chart accuracy) and advises documenting limitations when publishing.

Data-driven method for defensible analysis cutoffs

A practical guide describing a four-step, data-driven method for choosing defensible cutoffs (Measure, Price, Defend, Record). The author illustrates the approach with a 68-year Billboard chart example (57% of charting artists appear once) and shows how candidate thresholds (3+, 5+, 10+) map to survivor counts. The piece warns against the small-sample trap and argues every ratio or average-based ranking needs a minimum-denominator floor chosen from the measured distribution. References include Wainer (on variability), Tversky & Kahneman (on small-sample bias), and Tukey (exploratory data analysis).