Topic
Cache TTL selection and separating static from per-session prompt blocks so cached prefixes survive long conversational pauses.
A developer parsed his own Claude Code transcripts and found 1.99 billion cached reads against 62 million writes, which turns the length of your context into a recurring charge on every remaining turn rather than a setting you pick once.
Reality
A dev.to walkthrough of the run graph finds one append-only history per run, copied whole into each model call. The step count sets the exponent; your tool outputs set the price.
Reality
Prompt cache is scoped per upstream endpoint, so round-robin routing turns every agent turn into a full-price cache miss. One gateway writeup puts the sticky-routing saving at 50-70%.
Reality
The platform pricing page converts token spend into Claude Consumption Units at $0.01 each for a single AWS line item, and adds a 1.1x multiplier for US-only inference on Claude 4.6 and later.
Reality
A developer's account of a real-time transcription overlay says two of four candidate fixes did nearly all the work. One of the two was deleting code that seemed obviously correct.
Reality