FeaturedEssays
Accuracy tells you how often a model is right. Calibration tells you whether its confidence means anything, and most dashboards never look.
Demo AuthorRead “Calibration is the forecast quality nobody checks” →
Notebooks
Population stability index, a rolling window and one chart: a feature drift monitor you can run before buying a platform.
Demo AuthorRead “Notebook: a drift monitor in forty lines of pandas” →
Measuring what matters
Every metric is a proxy for something you care about. The first part of this series catalogues the ways proxies break.
Demo AuthorRead “Measuring what matters, part 1: proxies and their failure modes” →
Measuring what matters
Longer sessions look like engagement. Often they are confusion. A closer look at a heavily skewed distribution.
Demo AuthorRead “Measuring what matters, part 2: session length is a trap” →
Causal inference notes
With twenty metrics on the dashboard, one will move by chance in almost every test. Writing down the primary metric first is the cheapest fix.
Demo AuthorRead “Causal inference notes, part 1: pre-register your primary metric” →
Methods
Heatmaps of correlations are everywhere in exploratory analysis. Here are the four mistakes they invite and how to avoid them.
Demo AuthorRead “Reading a correlation matrix without fooling yourself” →
Methods
A forecast without an interval is a guess dressed as a fact. How to produce intervals that grow with the horizon, and how to check them.
Demo AuthorRead “Forecast intervals that widen honestly” →
Notebooks
A co-citation network of 600 papers, community detection in three lines, and what the clusters say about how a field grew.
Demo AuthorRead “Mapping a research field from its citations” →