Symetra: Visual Analytics for the Parameter Tuning Process of Symbolic Execution Engines

A research team from KAIST and Seoul National University has introduced Symetra, a visual analytics system designed to tackle a critical bottleneck in automated software testing. Symbolic execution engines like KLEE automatically generate test cases to maximize branch coverage, but their performance hinges on dozens of complex parameters. Users typically rely on suboptimal default settings because understanding each parameter's impact is notoriously difficult. While automated tuners exist, they act as black boxes, offering little insight into why certain configurations succeed. Symetra addresses this by enabling a human-in-the-loop approach, giving experts the visual tools needed to interpret and guide the tuning process.

Symetra provides two complementary visual overviews that map how different parameter configurations affect branch coverage values and patterns. This design allows users to collectively analyze and contrast groups of configurations, identifying subtle differences that influence performance. The system was validated through expert case studies, which revealed that users could not only interpret parameter impacts but also identify complementary configurations that automated methods missed. The result was a demonstrable improvement over fully automated tuning in both final branch coverage and the efficiency of the tuning process itself, proving that expert intuition, when properly supported by visualization, can still outperform pure automation for complex optimization tasks.