4thwithme

DORA, SPACE and DX Core 4 measure teams, not people. Surveys show what logs miss. A ladder, a rubric and a brag doc measure the person.

Part 1's models only look back, so they cannot tell a fad from a trend. Two models that look ahead: Fourier waves to remove the season, and LightGBM to predict next month. Same fake data, real numbers, and no clear winner.

Four simple models that use only past sales (window count, time decay, velocity, z-score), run on the same fake data so you can watch them disagree.

How to build synthetic customers without LLMs. For each step: what to try, what can go wrong, and what worked for us. Made-up store, real method.

A/B tests are slow and offline metrics mislead. Researchers build fake customers to test for you. Four ways to build them, and the open problem.

A cron job is five fields and a command. It's also where backups stop without a sound, reports run twice, and jobs pile up. How to read the syntax, how to run it in GitHub Actions, Docker, and Kubernetes, and how to know when it breaks.

Two RGB colors ten units apart can look identical, or nothing alike, depending on where you are in the space. Fifty years of color science fixed that, step by step.

A clean p-value doesn't mean the test was clean. A/A tests, sample ratio mismatch, and Simpson's Paradox are the checks that catch a broken testing setup before you trust anything it tells you.

The standard test answers one narrow question. When you're maximizing revenue instead of detecting a difference, or your users aren't independent, or there's only one of you to test on, that question is the wrong one.

Everyone thinks they know what statistical significance means. Almost nobody does. A generic ecommerce scenario, the actual math behind peeking, and why this is an org problem wearing a stats costume.

Karabiner remaps that skip the guesswork, zsh + Warp with zero multiplexers, three editors plus Claude Code for the other half of the time, and Spotlight beating Raycast out of the lineup.

Nuclear physics to Engineering Management, who I am, what I do, and why I'm starting a weekly blog on ML, agentic development, and the occasional controversial take.