AI Reasoning Traces and Coded Units for Software Engineering Tasks

7,000+ segmented reasoning units with qualitative codes and 100 raw reasoning traces from AI models like DeepSeek and Gemini. This dataset accompanies a research paper on interacting with AI reasoning models for software engineering, providing materials for replication and further analysis. It was last updated on 2026-05-22 by author Christoph Treude.

Use Cases

• Replicating the qualitative study on AI reasoning challenges based on the provided coded units and raw traces.
• Analyzing divergent reasoning paths and conflicting conclusions from AI models on software tasks.
• Conducting further quantitative analysis on the 100 reasoning traces collected from Stack Overflow questions.

Strengths

• Contains over 7,000 segmented reasoning units with assigned qualitative codes.
• Includes 100 raw reasoning traces from two AI models (DeepSeek and Gemini) for 50 Stack Overflow questions.
• Provides full transparency for replication with all example cases and the full corpus from the associated paper.

Limitations

• Row count for the primary data files is unknown, which may limit suitability assessment.
• Column-level documentation is absent; field semantics must be inferred after download.
• The dataset is small (716.3 KB), indicating a limited scope focused on specific examples and a qualitative study.

Provenance

Source

Research paper by Christoph Treude and Raula Gaikovina Kula.

Collection Method

Collected AI-generated reasoning traces from models like DeepSeek and Gemini for software engineering tasks and Stack Overflow questions.

Freshness

Last updated 2026-05-22 10:49:46; freshness should be verified.