There's a witching hour approximately around 3:30-6:30 ET when East Coast people and Bay Area people are actually working based off when LLMs slow down broadly and Cursor throws errors like 'our model provider is experiencing high demand right now'