safety
The Agent's Memory Keeps Its Receipts
WikiTrace links distilled lessons back to trajectory summaries and raw evidence so an agent can challenge bad advice before acting.
Summary
WikiTrace links distilled lessons back to trajectory summaries and raw evidence so an agent can challenge bad advice before acting.
External agent memory can degrade performance when a concise lesson is wrong, misread or applied outside its original context. WikiTrace stores each lesson with a summary of why its source trajectory succeeded or failed and a link to the raw trajectory. When a lesson is underspecified, conflicts with another or comes from failure, the agent can descend through those links before acting. Across five agent benchmarks, the authors report that evidence-linked memory outperformed lesson-only baselines and reduced memory-induced degradation.
Why it matters
WikiTrace links distilled lessons back to trajectory summaries and raw evidence so an agent can challenge bad advice before acting.
Limits and context
- Across five agent benchmarks, the authors report that evidence-linked memory outperformed lesson-only baselines and reduced memory-induced degradation.
Key claims
WikiTrace links distilled lessons back to trajectory summaries and raw evidence so an agent can challenge bad advice before acting.
Qualification: Across five agent benchmarks, the authors report that evidence-linked memory outperformed lesson-only baselines and reduced memory-induced degradation.
Evidence: source-2026-11-11-015
Sources
- arXiv:2611.14320v1arXiv / Cornell University · primary research
Corrections
No corrections have been recorded for this story.