Make the evidence inspectable.

Required research record

• Question: the precise hypothesis or engineering question.
• Methodology: task definition, population, conditions, and procedure.
• Artifacts: exact BOUND documents, benchmark inputs, code, and configuration.
• Dataset: source, format, inclusion/exclusion rules, and version.
• Raw results: observations before aggregation.
• Analysis: transformations, statistical procedures, and uncertainty where applicable.
• Environment: software versions, hardware/runtime assumptions, and relevant configuration.
• Limitations: known threats to validity and scope of inference.

Publication structure

research/
├── question.md
├── methodology.md
├── artifacts/
├── dataset/
├── results/
├── analysis/
├── environment.md
└── limitations.md

Version everything that affects comparison

Comparable results require stable identifiers for the benchmark definition, BOUND artifact versions, implementation revision, analysis procedure, and dataset version. If any of these change materially, the experiment record should make the revision visible.