Deceptive and inaccurate reporting?

According to the readme and website, RTK promises to save 60-90% of tokens. In order to be more honest to the users, it should be clear that this represents only specific bash tool uses, and compare the value saved by RTK to the full session token usage. Statements like "High-performance CLI proxy that reduces LLM token consumption by 60-90%" in the readme header are deceptive at best.

The website has this statistic about usage saved over a 2 hour coding session:

In order to fairly portray the savings, you should include how many tokens were used in the 2 hour session overall. I suspect the number is in the millions, meaning the RTK savings overall are probably less than 10%.

If this is true, I don't see how it makes any sense that RTK extends sessions by a factor of 3x as well. Is this some kind of cherry picked session where the only tools used were RTK specific bash commands?

I can't find any information on how token use through bash tool and RTK savings compares to other tools. If most of the context comes from tools like read / write / web search, what does it mean for RTK to save 60-90% of only bash for your overall usage? Also, why are savings from things like cat file reads even advertised considering agentic harnesses always read files with the read tool which doesn't benefit from RTK?