Workflow Cards: New Structured Summaries for AI Workflow Executions

A recent publication on arXiv (arXiv:2608.11022) presents Workflow Cards, a new format designed for documenting machine learning workflows. The authors highlight that while Model Cards and Data Cards effectively document static ML components such as datasets and models, they fail to account for the dynamic nature of workflow executions that generate, modify, and assess these components. These executions hold vital details regarding data preparation, parameter selections, runtime behavior, resource allocation, and intermediate changes, which can lead to bias, variations in performance, and issues with reproducibility. Workflow Cards seek to summarize machine-readable provenance information from workflow executions into a format accessible to both humans and large language models (LLMs). The paper proposes a Workflow Card template based on various examples and discusses its potential uses, addressing a significant gap in AI documentation by stressing the need to document the complete lifecycle of ML models. This work is classified as a cross-type announcement and is accessible on arXiv.