Deep Reads
5 papers and posts explained from first principles.
-
Quipu: A Governed Bitemporal Knowledge Graph Store
Quipu combines post-state write gates, bitemporal governance, conservative trust composition, and signed audit evidence to govern agent-written knowledge graphs while exposing the limits of policy compliance and historical replay.
-
Graph Engineering in the Era of LLM Agents: From Individual Intelligence to System Intelligence
This survey explains how explicit task, agent, and runtime-state graphs, grounded in shared ontology, can organize reliable agent systems and inform the design of a governed context platform.
-
EvoOntology: A Self-Evolving Ontology Layer for Data Agents
EvoOntology builds an evidence-grounded, tool-accessible ontology over heterogeneous data and improves it through trajectory-guided edits accepted by paired validation.
-
HySparse2: Hybrid Sparse Attention with Two-Level KV Sharing
HySparse2 shares attention memory across and within decoder blocks so long-context agents can prefill with less computation and KV-cache storage while improving retrieval.
-
Skill-Based Agentic Evaluation for Real-Time Data Science Tasks
A first-principles, interactive guide to grading agents when the right answer keeps changing.