What's this about?
ExperienceIndex records how individual documents and pairs of documents helped solve earlier tasks. Future queries retrieve those experiences to find the complete relevant set faster.
What exactly is artifact-grounded memory?
Instead of remembering a vague lesson such as “search precedents,” it remembers that a particular case supported one issue and how it relates to another case. It is an index of reusable experience tied to source material.
Why does this matter?
The authors report up to 11 points better quality and 50.5% lower online cost, plus transfer between task types over the same corpus. These gains depend on corpus stability and await independent replication.