Loading
Case studies
Every engagement here arrives with the instrument attached: the prompt set, the competitor set, the engines, and the protocol it was measured under. You can check the number rather than take it.
1 published · every one re-tested on the same instrument
What goes in the register
The prompt set, the competitor set, the engines and the sampling window are all stated. A study that will not say what it measured is a screenshot with a logo on it.
Answer engines are non-deterministic. The same prompt returns a different source set on a different day, so every figure is a rate across repeated samples rather than one lucky run.
Where a reading sits on too few mentions to be stable, the study says directional and declines to score it. We would rather under-claim a number than headline one that will not survive a re-test.
Every study locks its protocol so a re-test measures against the same instrument rather than a new one. Movement is credited only when the new interval separates from the old.
Start your own
An audit measures where you actually stand and names the work that would move it. Nothing moves on its own: the reading changes when the fixes ship. You get the baseline either way, whether or not it flatters anyone.