myronkoch/reminisce-production-postmortem

Reminisce v2数据集是一个关于AI代理的提取性记忆架构的生产后分析报告。该数据集伴随了论文《The Placeholder That Became Production: A Postmortem of Extractive Memory Under Continuous Multi-Agent Ingestion》（Koch, 2026），讨论了在连续多代理摄入下提取性记忆架构的超线性噪声增长问题。数据集包含了4个月的生产关键指标，如146个事件、452个提取事实，以及噪声模式分类和计数等。数据文件包括完整的6个快照时间序列分析，但不包含原始数据库文件（.db），这些文件包含私人会话记录。

The Reminisce v2 dataset is a production postmortem report of an extractive memory architecture for AI agents. It accompanies the paper "The Placeholder That Became Production: A Postmortem of Extractive Memory Under Continuous Multi-Agent Ingestion" (Koch, 2026), discussing the superlinear noise growth in extractive memory architectures under concurrent multi-agent ingestion. The dataset includes key metrics from 4 months of production, such as 146 episodes, 452 extracted facts, and noise pattern taxonomy with counts. Data files include a complete 6-snapshot time-series analysis, but raw database files (.db) containing private session transcripts are not included.