Search papers, labs, and topics across Lattice.
This paper introduces PatchOptic, an innovative interface designed to enhance shared-state workflows in large language models (LLMs) by enabling projected reads and verified structured updates. By establishing a contract between local updates and global validity, PatchOptic improves the management of state fragments during model invocations, addressing the limitations of traditional methods like keyword search and retrieval-augmented generation. Evaluation using PatchBench demonstrates that PatchOptic significantly reduces leakage and token costs while maintaining output quality, with runtime verification effectively preventing workflow-contract violations.
Projected reads in PatchOptic not only cut token costs but also ensure that local updates remain valid in the context of shared-state workflows.
Agentic workflows often operate over shared, structured state. Because LLM context windows are limited, each model invocation is typically shown only the state fragment needed for the current workflow step, a pattern commonly known as progressive disclosure. Modern systems construct such model-facing views using grep-like keyword search, retrieval-augmented generation (RAG), abstract-syntax-tree (AST) queries, and task-specific agent skills. These methods make the read side manageable, but they do not define when a locally proposed rewrite is valid after it is applied back to the full state. The missing piece is a contract between local updates and global validity. We introduce PatchOptic, an optic-inspired interface for shared-state LLM workflows. Optics are compositional bidirectional accessors that describe how views of structured data are read and updated. PatchOptic borrows this view/update intuition and realizes it through projected reads and verified structured patches. Each workflow step declares a projected read view, an authorized write region, and a patch-source region. Beyond runtime enforcement, the same declaration yields a path-level footprint that supports delegation, sub-workflow composition, and static certificates for reordering independent steps within the same phase. We evaluate this design with PatchBench, a benchmark with 46 cases across domains. The results show that projected reads reduce reported leakage and token cost while preserving accepted-output quality under the strong actor. Runtime verification blocks declared workflow-contract violations before commit, and patch-read enforcement rejects compromised patch artifacts that use hidden sources.
MIT CSAIL