System One decision models are designed to act quickly: given a state and a set of options—such as intents, tools, or interface elements—they return a distribution over those options in one forward pass, with no generated reasoning. According to the abstract of the new arXiv paper, web systems increasingly build on this kind of model, which makes the stability of their scores important.
The paper introduces Prism-1, a scoring approach that is explicitly position- and rival-independent. That means the score assigned to an option should not shift merely because the option appears in a different position in the list, or because the set of competing options changes. The abstract frames this as a way to make System One models more reliable in practical settings.
The announcement does not include experimental results or implementation details, so the claims are limited to the proposal itself. Still, the motivation is clear: for web systems that repeatedly score changing option sets, a scoring method that ignores positional and rival effects could offer more consistent behavior.