Method

DeepSeek V4 Pro

DeepSeek-V4-Pro-0813 directly evaluates each source sample using the developer-guided zero-shot prompt adopted from Ouédraogo et al. The prompt covers structure, nesting, intent, length, action granularity, and reading flow. Scores average three independent runs, without tools or an agentic workflow; recorded explanations appear on sample pages.

Score scale: 0–20; mean of three runs.

Dataset results