Field notes on building believable synthetic businesses — the realism engine, the dashboards it powers, and what it takes to generate data an expert can't tell from the real thing.
Stripping names off real data doesn't make it safe — re-identification is a solved attack. Why data that never touched a real person is the only clean answer, and the one caveat that matters.
Read the article →Revenue in millions next to a fill rate in percent: the classic setup for a chart that lies. Dual axes, indexing, log scales, and small multiples — and when each one is the honest choice.
Read the article →A field guide to matching charts to questions, varying the form so a dashboard stays legible, and using color that means something — illustrated with the distributor data.
Read the article →Forecasting, churn, price elasticity, LTV, anomaly detection, and more — the signal is real because the business behind the data is coherent.
Read the article →Load testing, multi-dialect migrations, dashboard QA, and training — the jobs a realistic synthetic dataset does better than real production data.
Read the article →Anyone can generate a million rows. Making them behave as if a single coherent business produced them is where the real work — and the value — is.
Read the article →Two numbers that should match but don't will sink a dashboard. The two failure modes — mixed revenue bases and mixed period grains — and how to fix and guard against them.
Read the article →Gross margin hides the discounts, rebates, and freight that decide whether a deal actually made money. How the waterfall exposes every leak — and why net merchandise is the revenue anchor.
Read the article →Postgres, SQL Server, MySQL, Oracle, and Snowflake from one canonical schema — the transpilation decisions that keep them in lockstep, and why views are a separate add-on.
Read the article →How the pipeline builds hundreds of millions of interrelated rows with bounded memory — and keeps the derived analytics correct by accumulating summaries as the data streams past.
Read the article →Fake geography is easy to spot. How five census regions drive DC assignment, delivery performance, freight cost-to-serve, and a revenue map that actually means something.
Read the article →More articles are on the way. Want a topic covered? Let us know.
