Prompt Repetition Experiments
收藏Zenodo2026-04-27 更新2026-05-26 收录
下载链接：
https://zenodo.org/doi/10.5281/zenodo.19825521
下载链接
链接失效反馈官方服务：
资源简介：
Session logs, scoring rubrics, and analysis outputs from three pre-registered controlled experiments (30 agent sessions, 4,306 messages) testing whether instruction repetition improves task-success rates in parallel LLM-agent pipelines. Each experiment pairs a control group (single-copy instruction) against a treatment group (repeated instruction, verbatim duplication of the delegate system prompt) under blinded binary rubric scoring. Data include per-run conversation traces (JSONL), structured delegate outputs (JSON), blind scorer justifications, group assignment maps, and token/timing metadata.
提供机构：
Zenodo创建时间：
2026-04-27
