What is it about?

Software is checked for mistakes by running thousands of automated tests, which can take hours, too slow for teams that update their software daily. We reorder tests so the ones most likely to catch mistakes run first. This is done by automatically drawing a "map" of how a program's building blocks depend on each other and scoring each block on its risk of containing bugs. Because research shows most bugs come from a few trouble spots, we amplify the scores using the exponential function, a simple function from high school math that makes high values count enormously more. This consistently found bugs faster on three real-world projects, at virtually no extra cost.

Featured Image

Why is it important?

To our knowledge, this is the first test prioritization technique that exploits a well-established but underused fact: bugs cluster in a small number of trouble spots, following a so-called power-law pattern. Mirroring that pattern in how tests are ranked turned out to beat the standard approach on every system and metric we evaluated, a one-line change costing virtually nothing extra. This matters now because teams releasing software multiple times a day need fast methods that work straight from the code, and ours plugs directly into such pipelines. Our implementation is freely available, so developers and researchers can apply or extend it immediately.

Perspectives

What surprised me most while working on this paper was how much we got from such a simple change. We expected the enhanced graph model to do the heavy lifting, but the exponential weighting, which was one line of math motivated by an old empirical observation about where bugs live, improved every single configuration we tested. It's a nice reminder that, sometimes, simplicity beats complexity.

Olav Blaak
Associatie KU Leuven

Read the Original

This page is a summary of: Exponentially-Weighted Test Case Prioritization Based on an Enhanced Graph Representation of Software, March 2026, ACM (Association for Computing Machinery),
DOI: 10.1145/3748522.3779859.
You can read the full text:

Read

Resources

Contributors

The following have contributed to this page