View a markdown version of this page

RAIER01-BP03 For each system update, re-run the evaluation and update the system registry - Responsible AI Lens

RAIER01-BP03 For each system update, re-run the evaluation and update the system registry

Record evaluation activities in logs that capture test conditions, system configurations, data inputs, raw results, and methodological notes with sufficient detail to make the entire process reproducible. Establish version control for evaluation artifacts to assist builders to trace unique system builds and their corresponding evaluation results.

Level of risk exposed if this best practice is not established: High

Implementation considerations

  1. Log your evaluation runs, including information on which datasets you used, what system version you tested, what hardware and software configuration you ran on, and raw and intermediate outputs. Your logs should be detailed enough that someone else could reproduce your exact evaluation months later.

  2. Set up version control for your evaluation materials, including test scripts, configuration files, and result outputs.

  3. Link your evaluation materials to both your system and your dataset registry so that it is clear which data and system versions led to the evaluation results. This allows you to link each system build and dataset pair to its specific evaluation artifacts.

Resources

Related documents