Reproducibility

Reproducibility

Preserve for each final experiment:

• Git commit
• experiment configuration
• model profile and identifier
• repetition
• mutation and test IDs
• raw evidence
• generated summary tables
• software environment
• failure logs

Before public release, inspect evidence for credentials, .env values, proprietary inputs, machine usernames, absolute paths, internal endpoints, and sensitive prompts.

Use tagged releases or external artifact storage for large machine-readable evidence when appropriate.