Archives

Splitting 'what changed' from 'is that acceptable' into two commands — and why compare has to run first.

Six validators, three file types, one command — catching a broken policy file before it breaks check.

A policy file, a stability gate, and 21 benchmarks that correctly refused to give an answer.

BenchmarkDotNet gives you excellent performance measurements — but no opinion on whether a number is acceptable, no memory of yesterday's baseline, and no way to fail a build. Here's the gap, and what I'm building to close it.

How CedarRecon's exception classifier went from 1,368ms to 646ms per million transactions, and the four optimizations that made things worse before it got better.