Philosophy Behind BoCoFlow

Let's explore the real-life activities of computational scientists. To complete a computational assignment, they might run Python code with complex conda environment dependencies, process files using various installed software packages, and submit heavy computational tasks to a remote supercomputer center. Some may document these steps in a file for sharing or future reference. However, this back-and-forth trial process can be tedious and error-prone. While some might attempt to write a Python script to automate the entire procedure, the complexity of scenarios often spirals out of control. Sometimes, a simple automation works adequately, but when new assignments arise, they must rewrite the entire script from scratch. Debugging? That's a real challenge. Different software versions, environment conflicts, remote job failures... It's frustrating.

While any computation can be wrapped as a node, the key is finding the right level of abstraction. We treat nodes at a conceptual level that aligns with how computational scientists naturally think about their work, avoiding unnecessary complexity while maintaining flexibility. Moving nodes to less abstract levels would compromise both performance and conceptual clarity.

We organize computational tasks into four distinct layers, each building upon the previous to enable increasingly complex scientific workflows.

The node layer bridges workflow design and execution. To maintain its conceptual integrity in design, we choose to delegate hardware dependencies in execution to the user. Rather than forcing users into a rigid environment, BoCoFlow empowers them to configure their own computational setup through executable paths, conda environments, or Docker images, ensuring maximum flexibility while maintaining reproducibility.

When computational tasks are wrapped as nodes, they gain powerful workflow capabilities that transform them from isolated operations into interconnected, manageable research components. These capabilities emerge naturally from the node architecture, enabling seamless integration into scientific workflows.

This transformation from isolated computations to workflow-enabled nodes revolutionizes how research is conducted. By handling operational complexity, the system frees researchers to focus on scientific discovery rather than technical overhead. Manual, error-prone processes become reliable, reusable components that can be shared with the scientific community - accelerating research progress and fostering open science collaboration through reproducible workflows.