Papers
- Proof-or-Stop:Don’t Trust the Agent, Trust the Evidence Loop Engineering for Verifiable Evidence-Gated Lifecycle Control
https://arxiv.org/pdf/2607.14890 arXiv:2607.14890
Lifecycle transitions (reviewed, tested, done, ready-to-merge) stay mere agent claims unless a downstream system can check them against evidence. Proof-or-Stop admits a transition only when fresh, source-state-bound, mechanically verifiable evidence satisfies the gate — agent outputs may propose lifecycle claims but never constitute them. In a 9,240-cell ablation it cut visible-pass/hidden-fail amplification from 31/1800 to 2/1800, and a self-hosted corpus (565 stories, 94.8% of review findings resolved) shows the system producing auditable evidence on its own development.
- An Empirical Study of Harness Design for Coding Agents
https://arxiv.org/pdf/2609.20804 arXiv:2609.20804
A component-level empirical study of coding-agent harnesses — holding the execution loop fixed while varying planning, action space, and context management across 176 matched settings on SWE-Bench Verified and Terminal-Bench 2.1. Standout findings: rule-based elision staged before LLM summarization is the most efficient context strategy, planning shifts from accuracy scaffold to cost saver as models strengthen, and bash-capable models do better with a bash-only interface than with predefined tools, at substantially lower cost.
- The Log is the Agent: Event-Sourced Reactive Graphs for Auditable, Forkable Agentic Systems
https://arxiv.org/pdf/2605.21997 arXiv:2605.21997
Abstract page for arXiv paper 2605.21997: The Log is the Agent: Event-Sourced Reactive Graphs for Auditable, Forkable Agentic Systems
- Solving a Million-Step LLM Task with Zero Errors
https://arxiv.org/pdf/2511.09030 arXiv:2511.09030
Abstract page for arXiv paper 2511.09030: Solving a Million-Step LLM Task with Zero Errors
- On the design of text editors
https://arxiv.org/pdf/2008.06030 arXiv:2008.06030
Abstract page for arXiv paper 2008.06030: On the design of text editors
- Supporting Our AI Overlords: Redesigning Data Systems to be Agent-First
https://arxiv.org/pdf/2509.00997 arXiv:2509.00997
Abstract page for arXiv paper 2509.00997: Supporting Our AI Overlords: Redesigning Data Systems to be Agent-First
- Self-Consistency Improves Chain of Thought Reasoning in Language Models
https://arxiv.org/pdf/2203.11171 arXiv:2203.11171
A technique for improved Chain of Thought Prompting by sampling different approaches and picking the ones that arrived at consensus.
- Large Language Models Are Human-Level Prompt Engineers
https://arxiv.org/pdf/2211.01910 arXiv:2211.01910
On using LLMs to improve/optimize prompts.
- Should We Respect LLMs? A Cross-Lingual Study on the Influence of Prompt Politeness on LLM Performance
https://arxiv.org/pdf/2402.14531 arXiv:2402.14531
On the Influence of Prompt Politeness on LLM Performance.
- Bitcask: A Log-Structured Hash Table for Fast Key/Value Data
https://riak.com/assets/bitcask-intro.pdf
A simple implementation of a K/V store where only the keys need to fit into memory.
- Delta Lake: High-Performance ACID Table Storage over Cloud Object Stores
https://people.eecs.berkeley.edu/~matei/papers/2020/vldb_delta_lake.pdf
In this paper, we present Delta Lake, an open source ACID table storage layer over cloud object stores initially developed at Databricks.