Notes
What the critics keep catching
Writing about judging agent output rather than producing it — where a single reviewer goes blind, why anything measured against a fixed suite eventually games it, and what these runs really cost.
4 min
What a gauntlet run actually costs
The method has no round limit by design, which means it has no natural spend limit either. Here is where the tokens go, why Graph mode multiplies rather than adds, and the three habits that keep a run from quietly becoming expensive.
4 min
One critic has exactly one blind spot
A single reviewer per piece finds real problems and misses a predictable category of them — the ones it was not looking for. Replacing it with a graph of narrow critics is not about being thorough; it is about not sharing a blind spot.
