{"arxiv_id":"2608.19013","incoming":{"papers":0,"sentences":0,"groups":[],"unlabelled_sentences":0},"outgoing":{"papers":21,"sentences":22,"groups":[{"kind":"data_used","label":"Uses data or values from","papers":1,"sentences":1,"quotes":[{"arxiv_id":"2110.14168","title":"Training Verifiers to Solve Math Word Problems","url":"/paper/2110.14168","section":"Textual Reasoning","sentence":"The textual stream follows the order MuSiQue [ref] , ProofWriter [ref] , GSM8K [ref] , and HotpotQA [ref] .","intent_p":0.86,"contests_p":null}]},{"kind":"comparison","label":"Compares with","papers":2,"sentences":2,"quotes":[{"arxiv_id":"2601.03192","title":"MemRL: Self-Evolving Agents via Runtime Reinforcement Learning on Episodic Memory","url":"/paper/2601.03192","section":"ALFWorld","sentence":"We compare HCL with a Static Harness, a RAG baseline, MemP [ref] , and MemRL [ref] .","intent_p":1.0,"contests_p":null},{"arxiv_id":"2511.15164","title":null,"url":"/paper/2511.15164","section":"Multimodal Perception","sentence":"We additionally compare with DGG [ref] , a recent adaptive method for sequential multi-task continual learning whose setting aligns with this controlled multimodal stream.","intent_p":1.0,"contests_p":null}]},{"kind":"background","label":"Cites as background","papers":7,"sentences":8,"quotes":[{"arxiv_id":"2310.08560","title":"MemGPT: Towards LLMs as Operating Systems","url":"/paper/2310.08560","section":"Harness Engineering","sentence":"MemGPT, Reflexion, and Voyager retain experience as memory, feedback, or executable skills [ref] .","intent_p":0.83,"contests_p":null},{"arxiv_id":"2205.00445","title":"MRKL Systems: A modular, neuro-symbolic architecture that combines large language models, external knowledge sources and discrete reasoning","url":"/paper/2205.00445","section":"Harness Engineering","sentence":"Toolformer, MRKL, and HuggingGPT expose and coordinate external capabilities [ref] .","intent_p":0.86,"contests_p":null},{"arxiv_id":"2605.20247","title":"CP-MoE: Consistency-Preserving Mixture-of-Experts for Continual Learning","url":"/paper/2605.20247","section":"Model-Centric Continual Learning","sentence":"Architecture-based approaches isolate, expand, or select model components to reduce interference between tasks [ref] .","intent_p":0.81,"contests_p":null},{"arxiv_id":"2605.26112","title":"From Model Scaling to System Scaling: Scaling the Harness in Agentic AI","url":"/paper/2605.26112","section":"Harness Engineering","sentence":"Environment adapters execute actions, and task-specific validators check outcomes at the boundary of the pipeline [ref] .","intent_p":0.81,"contests_p":null},{"arxiv_id":"2404.14829","title":"Revisiting Neural Networks for Continual Learning: An Architectural Perspective","url":"/paper/2404.14829","section":"Model-Centric Continual Learning","sentence":"Architecture-based approaches isolate, expand, or select model components to reduce interference between tasks [ref] .","intent_p":0.81,"contests_p":null}]}],"unlabelled_sentences":11},"claims":[],"thresholds":{"intent_p":0.8,"supports_p":0.95,"contests_p":0.7},"method":"Citing sentences come from each paper's LaTeX source; every \\cite marks the sentence and the works it cites. Pith's classifier typed each sentence. A label is shown at intent probability 0.8 or more (0.95 for support), a contest at 0.7 or more. Measured accuracy against two blind model families is at https://pith.science/accuracy. The quoted sentences are the citing authors' own.","held_up":null,"_links":{"html":"https://pith.science/paper/2608.19013"}}