Two seeded failures in one small graph, and what each logging approach actually recorded. When a LangGraph run dies, the log usually tells you the same thing twice: a node started, a node stopped, an exception was raised. It almost never tells you what the node was given. I wanted to know what a log has to contain before the failure happens, so I broke a three-node graph two different ways and checked what each run actually recorded. I wrote this with AI assistance, and I fact-checked and...