Own research
Operational handoff evaluation
What the recorded AI-worker probes reveal about context loss, recovery and unnecessary blocking.
Latest
From the
research desk.
Our own experiments, public AI measurements and reading that shapes the next question. Each item keeps its source and evidence type visible.
Data release
277 sourced model records
A searchable snapshot with provider, source and coverage information. External scores retain their original measurement scale.
External paper
AgentDojo: A Dynamic Environment to Evaluate Prompt Injection Attacks and Defenses for LLM Agents
Selected methodological reference
External paper
Why Do Multi-Agent LLM Systems Fail?
Selected methodological reference
External paper
MAS-FIRE: Fault Injection and Reliability Evaluation for LLM-Based Multi-Agent Systems
Selected methodological reference
External paper
$τ$-bench: A Benchmark for Tool-Agent-User Interaction in Real-World Domains
Selected methodological reference
External paper
$τ^τ$-Bench: An Environment for End-To-End, Realistic Agent Construction
Selected methodological reference
External paper
3DCodeVerse: Exploring 3D Worlds as Code
Automated discovery; editorial review pending