Context Management & Reliability

Manage the context window, prompt caching, long conversations and production reliability.

1. 5.1The Context Window and Token BudgetingWhat counts towards the context window, how to measure it, and how to divide it between instructions, history, retrieved content and output.9 min
2. 5.2Prompt CachingHow prompt caching works as a prefix match, where to place breakpoints, what it costs and saves, and the silent mistakes that invalidate it.10 min
3. 5.3Compaction, Summarisation and Long ConversationsThe techniques for keeping a long-running conversation or agent inside its context window, what each one loses, and how to design for that loss.9 min
4. 5.4Retrieval, Citations and GroundingSupplying Claude with the right source material at request time, asking it to cite that material, and designing so that answers stay anchored to it.9 min
5. 5.5Errors, Rate Limits, Retries and TimeoutsThe error classes the API returns, how rate limits are measured and reported, and how to build clients that recover rather than fail.9 min
6. 5.6Observability, Evaluation and Regression TestingWhat to log from a Claude system, how to measure whether it works, and how to stop changes from making it worse.9 min