To be clear, that was during recovery, not a root cause.

The closest to that is this:

Neither outage was caused by a code or configuration change. Both incidents were capacity failures at their core. We failed to scale critical components before demand exceeded their capacity. Since April, monthly commits have grown from 1.4 billion to 2.9 billion. That growth explains the pressure on our systems, but it does not excuse these outages.

One could argue they still DDOSd themselves by promoting tools that produce commits at a much higher rate.

LLMs looove to commit things. Every little thing. It is handy for them as they can use commits as their own memory store that they can scan easily, but it leads to very wordy-but-minor commits.

Things that humans would just lump together in a single commit like "updated docs" gets broken into commits for every single change made over the course of a few hour session.