Field notes.What the verdicts keep saying.
Every number on this page comes out of finished TRIAL reports, stripped of who wrote them. We print the sample size next to the figure. When the sample is too small to mean anything, we print nothing at all.
METHODS
Two graders. The founder, who marks each section AGREE or DISPUTE, and the internal bench, where a run gets an expected stamp and then gets replayed whenever a prompt or a model changes. A run counts as graded once either grader has touched it.
One of two things reported back after 60 days: they built it, or they killed it. Anything softer than that is an opinion, and opinions do not go in the numbers.
The time constant on the strongest edge in a report. Not how long the market takes to notice. How long a funded copier needs to reproduce the thing that compounds.
Publish a figure without its sample size. Round a thin number up. Show a category with fewer than 5 runs. Numbers move as runs land, so this page changes.