Engineering deep dives on systems, infrastructure, and machine learning.
Our dashboards showed a healthy p99 while users kept complaining about slow pages. The metric was correct — the way we aggregated it was not.
Dana Whitlock · August 4, 2026 · 7 min read
Retrying a failing test until it passes is not a fix. Here is what happened when we stopped treating reruns as normal and started isolating the unstable suites.
Marek Oyelaran · July 28, 2026 · 6 min read
Our monorepo cache reported a high hit rate and saved almost no time. The problem was not the cache — it was what we were hashing into the key.
Dana Whitlock · July 20, 2026 · 8 min read