What it means
Planned. The unit that matters is tokens, cost and time per accepted outcome — not per prompt — computed from measured telemetry and the ledger's outcomes. A task run under the right profile may spend more on one prompt and less on the task, because it needed two sessions instead of three and no retry.
How it works
Nothing is implemented; it depends on telemetry landing first. The ledger already carries outcome and verification per task, so once tokens and cost are recorded the comparison is a query over one file.
How to see it
There is nothing to run.
What it does not cover
No percentage of any kind is claimed anywhere on this site or in the repository. None has been measured, so none is stated — that rule is checked by the site build, which fails on a number in the marketing copy.
Why it exists
"Saves tokens" is the wrong metric and the easiest to fake. Cost per accepted engineering outcome is the number a team can act on, and it can only be stated after twenty to fifty real tasks have been measured.