Why it matters
It is one of the rare token-savings claims somebody actually metered. The plugin's creator advertises 65%; the test shows that number only survives if you count output tokens and let a conversation run long enough to amortize the instructions.
Tokenmaxxing read
Output-only savings are the oldest move in the token-optimization pitch deck. Averaged over ten prompts, caveman took the total from 1,330 to 820 tokens, a 38% cut rather than 65%, while plain terse prompting reached 1,020 with no plugin and no input overhead.
Source takeaway
One writer, no published method, no error bars, so read the tables as directional. The single-prompt case is the honest counterexample: caveman's input ballooned to 600 tokens against 180 for a normal prompt, erasing most of the win.
