Articles, newest first
Numerical Verification Comes Before Optimization
A solver can conserve mass to machine precision and exhibit second-order self-convergence while approaching the wrong answer. A small conversion model shows ...
Read article
Monitoring Without Labels: What Is Actually Identifiable?
Two production environments can have identical inputs, predictions, and confidence scores while one has 90 percent accuracy and the other 10 percent. A monit...
Read article
More Subjects and Longer Trajectories Solve Different Problems
One thousand observations can mean twenty people measured fifty times or two hundred people measured five times. The better design depends on whether we need...
Read article
What a Before-and-After Testimonial Can Establish
A genuine improvement does not identify its cause. A worked probability model explains how selected starting measurements, natural variation, and selective r...
Read article
A Database for Analysis: Rows, Columns, Indexes and the Planner
The same five million orders answer an analytical query in 2 milliseconds or in 11 seconds, and a key lookup in 17 microseconds or 570, depending on how the ...
Read article
A Data Lake Is a Directory With Rules
A data lake is files in folders plus the conventions that make them usable. The same six million rows answer a question in 10 milliseconds or in 2.7 seconds ...
Read article
Why Exact Post-Selection Confidence Intervals Can Be Enormous
An exact 95% interval of [-71.7, 2.5] for a unit-variance Gaussian mean is not a bug. It is what conditioning on a selection event costs when the selected va...
Read articleWriting Statistical Software as Executable Mathematics
In statistical software, many of the strongest tests are not input-output examples. They are equations: test inversion must agree with pointwise decisions, p...
Read articleThe Trouble With Smooth Curves in Small Simulation Studies
Simulation studies often evaluate a method on a coarse parameter grid and then draw a smooth curve through the results. The plot looks persuasive. The proble...
Read article
How Often to Retrain: A Square-Root Rule and Its Limits
Retraining every week spends 15,000 a week to prevent decay worth nothing yet. Retraining twice a year spends nearly nine times as much in lost quality as th...
Read article