The Subtle Relationship Between Prompts and Caching
Change one character in System Prompt and the entire KV Cache is invalidated. Minimizing accidental cost
THE QUESTION THIS PAGE ANSWERS
ANSWER FIRSTWhat is the key idea behind “The Subtle Relationship Between Prompts and Caching”?
Change one character in System Prompt and the entire KV Cache is invalidated. Minimizing accidental cost
Make the claim earn its place. Use this page as a decision aid, not a definition to memorize. Connect the idea to one real task, one observable result, and one failure that would change your mind.
Write one question you could answer with evidence after trying this idea.
A conclusion that sounds complete but leaves the key assumption untested.
Every request, the system computes a fingerprint (hash) of the System Prompt
Different time each second → different fingerprint → cache never hits → recomputed every time → costs double.
This isn't just a timestamp issue: any dynamic content injected into the System Prompt has the same effect.
Switching Skills changes System → cache invalidated
System never changes → prefix cache always hits
Cache Hit Simulator
Assume 3 out of 10 requests switch to a different Skill. Observe the cache difference between the two injection approaches:
Why “Changed fingerprint = Cache invalidated” depends on the operation
“Different time each second → different fingerprint → cache never hits → recomputed every time → costs double.” makes the structure concrete. The useful comparison is not which name sounds more advanced, but how the data is arranged and how far the most common operation has to travel.
Read a structure through access and change
“Switching Skills changes System → cache invalidated” exposes a trade-off that is easy to miss: reading by position, looking up by key, adding at either end, inserting in the middle, and traversing relationships do not favor the same organization. A structure that is fast for one operation is not automatically fast for all of them.
Count scale and update frequency together
Use “Assume 3 out of 10 requests switch to a different Skill.” as a boundary check. Write down the data size, the dominant operation, and the latency you can accept before deciding whether an AI-generated structure actually fits.
From “Changed fingerprint = Cache invalidated” to “Skill injection method also affects caching”
“Changed fingerprint = Cache invalidated” grounds the problem in “Different time each second → different fingerprint → cache never hits → recomputed every time → costs double. This isn't just a timestamp issue: any dynamic content injected into the System Prompt has the same…”. “Skill injection method also affects caching” then moves it toward “Switching Skills changes System → cache invalidated”. Together, they show that the lesson is not just a conclusion to remember, but a claim with conditions.
Carry the judgment into the next situation
When you meet a new data structure, do not begin by memorizing its definition. Write down the most frequent operation, estimate scale and update behavior, and check whether the structure satisfies all three conditions.
- “Changed fingerprint = Cache invalidated”: Different time each second → different fingerprint → cache never hits → recomputed every time → costs double. This isn't just a timestamp issue: any dynamic content injected into the System Prompt has the same…
- “Skill injection method also affects caching”: Switching Skills changes System → cache invalidated
- “The closing point”: System never changes → prefix cache always hits
The final “The closing point” brings the discussion to “System never changes → prefix cache always hits”. The useful thing to carry forward is knowing which judgments must be revisited when input, scale, or risk changes.
I turned one judgment from this article into a small experiment I could run today. Knowing what to observe next is more useful than simply remembering the conclusion.
After reading this, I first looked for the conditions behind the idea instead of copying the method into a project. That order made the later trade-offs much clearer.
When this judgment reaches real work, which constraint should be added first? I am curious which step matters most between reading and the first practical attempt.
No discussion on this article yet.