Order prompts static-first with byte-identical prefixes to actually hit the 90%-off cache tier

Digital Applied·medium signal

Prompt caching is now the highest-ROI LLM optimization (Anthropic bills cached prefixes at 0.1x input, a 90% discount), but the work is 90% prompt hygiene: put all static content first, keep the prefix byte-identical across calls, and push per-request dynamic content to the tail. If a different user, document, or system message leaks into the prefix, the cache has nothing stable to hold and hit rates collapse. Reorder your prompt template once and the static portion drops ~10x in cost with zero quality change.