Computational workflows FAIR principles and best practices

[kinow]

Sorry the generic title, this is a follow-up from the last WCI meeting. We discussed moving actionable items to GitHub issues, and I remembered in our first meetings it was asked what were the goals of this project. And in the document we tried to list some, e.g.

• Define FAIR principles for computational workflows that consider the complex lifecycle from specification to execution and data products
• Define metrics to measure the FAIRness of a workflow
• Define recommendations for FAIR workflow developers and systems
• Define processes to automate FAIRness in workflows by recording necessary provenance data

It would be good to have a good document that could be cited and that answer all/some of these questions. This issue is both to gather papers that may answer these questions (we can update back the Google Doc with the answers), or to discuss the possibility of gathering material to write a paper or whitepaper about this.

[stain]

So let's start writing this paper!

[stain]

Some ideas for paper from 2023-09-14:

• What are the extra things that is not in FAIR Research Software?
• Need a community agreement on what is a “workflow”
• Is the Workflow Engine part of the FAIR Workflow?
  • Yes, it can be – for example, Parsl is imported directly into the script to make it a workflow
• Differences in talk about workflows in the abstract and workflows in execution
• Workflow Description vs Workflow Enactment
• Difference between (distributed) algorithms (and program of programs) vs workflows
  • More than one step – the composition of these steps is itself meaningful and help explain the process at a higher level than an algorithm.
• Complexity of coordination
• A workflow should be understandable by the domain scientists even if they are not themselves programmers (-- they don’t necessarily need to understand the underlying algorithms - e.g. image recognition)
• Scientific Workflows vs. Computational Workflows.
  • Exclude business/control flow workflows such as Airplane Reservation Systems.
  • Scientific Workflows are more data-centric, not event-driven