🧪 Symetra

Visual Analytics for the Parameter Tuning Process of Symbolic Execution Engines

A visual analytics system that supports Human-in-the-Loop parameter tuning of symbolic execution engines (e.g., KLEE). Symetra helps analysts understand why certain configurations work well — not just which ones — by revealing how parameters affect branch coverage across tuning trials.

Published at the Eurographics Conference on Visualization (EuroVis) 2026 · Computer Graphics Forum, Vol. 45 (2026), No. 3.

Symetra turns thousands of tuning trials into an interactive workflow: Overview → Compare Configuration Groups → Inspect Hyperparameters → Trace Branches to Code.

---

✨ Key Features

• Coverage View

  • Visualizes branch coverage achieved across trials against the base and total branch counts.
  • Surfaces which branches are newly covered, hard to reach, or consistently missed.

• Hyperparameter View

  • Compares the distribution and impact of each of the 61 KLEE parameters (boolean / numeric / categorical).
  • Relates parameter values to coverage gains over the baseline.

• Trial Group View (Collective Analysis)

  • Contrasts groups of configurations to identify differences that affect branch coverage.
  • Helps discover complementary configurations whose test cases cover different sets of branches.
  • Area / Bidirectional / Overlapped charts for side-by-side group comparison.

• Code View

  • Links a selected branch back to its source location (file + line) in the target program.
  • Lets analysts move from a coverage signal to the exact condition responsible for it.

---

🎯 Supported Targets

Tuning experiments run KLEE against real C programs, optimizing configuration for branch coverage.

Target	Program	Metric	Branches (covered / total)
grep	GNU grep	Branch Coverage	base ≈ 1135 / 8225
gcal	GNU gcal	Branch Coverage	— / —

Targets are defined in src/data/targetConfig.json; trial and branch data live in src/data/.

---

🔌 Add Your Own Target

1. Add an entry to src/data/targetConfig.json (name, base, total, max, …).
2. Drop the tuning results as src/data/tuned_parameters_<name>.json and branch metadata as src/data/branch_info_<prefix>.json.
3. Describe parameters in src/data/parameter_descriptions.csv.

The app loads these dynamically — see src/App.tsx and src/model/experiment.ts.

---

⚡ Quick Start

# 1. Install dependencies
pnpm install

# 2. Launch the dev server (HMR)
pnpm run dev

# 3. Build for production
pnpm run build

# 4. Preview the production build
pnpm run preview

Symetra is a client-side application — the experiment data is bundled from src/data/, so no backend server is required.

---

⚙️ Configuration

Deployment base path

The public base path is set in vite.config.ts. For the GitHub Pages project site it is /Symetra/; for root/custom-domain hosting use /.

// vite.config.ts
export default defineConfig({
  plugins: [react()],
  base: "/Symetra/",
});

Experiment metric

Defined in src/data/config.json:

Field	Description	Example
metric.name	Optimization metric	Coverage
baseValue	Baseline (default-config) coverage	1473.37
totalBranch	Total branches in the target program	8225

---

📊 System Overview

   src/data/*.json ──▶  Model layer  ──▶  React + visx/D3/Plotly views
   (trials, branch       (Experiment,       • Overview
    info, config)         Hyperparam,       • Coverage View
                          Trial, Target)    • Hyperparameter View
                              │             • Trial Group View
                         Zustand store ────▶ • Code View

Tech stack: React 18 · TypeScript 5 · Vite 5 (SWC) · Chakra UI · D3 / visx / Plotly.js · Zustand

---

🔗 Related Resources

• KLEE → KLEE: Unassisted and Automatic Generation of High-Coverage Tests
• ATMSeer → Increasing Transparency and Controllability in AutoML
• HyperTendril → User-driven Hyperparameter Optimization

---

📚 Citation

If you use Symetra in your research, please cite:

@article{hong2026symetra,
  title     = {Symetra: Visual Analytics for the Parameter Tuning Process of Symbolic Execution Engines},
  author    = {Hong, Donghee and Kim, Minjong and Cha, Sooyoung and Jo, Jaemin},
  journal   = {Computer Graphics Forum},
  volume    = {45},
  number    = {3},
  year      = {2026},
  note      = {Proc. Eurographics Conference on Visualization (EuroVis)},
  publisher = {The Eurographics Association and John Wiley \& Sons Ltd.}
}

---

👥 Authors

Name	Affiliation	Email
Donghee Hong	Sungkyunkwan University	dh.hong@skku.edu
Minjong Kim	Sungkyunkwan University	minjong.kim@skku.edu
Sooyoung Cha	Sungkyunkwan University	sooyoung.cha@skku.edu
Jaemin Jo*	Sungkyunkwan University	jmjo@skku.edu

* Corresponding author

---

🙏 Acknowledgments

This work was supported by the Institute of Information & Communications Technology Planning & Evaluation (IITP) grant funded by the Korea government (MSIT):

• RS-2019-II190421 — Artificial Intelligence Graduate School Program (Sungkyunkwan University)
• RS-2024-00438686 — Development of software reliability improvement technology through identification of abnormal open sources and automatic application of DevSecOps

and by the National Research Foundation of Korea (NRF) grant funded by the Korea government (RS-2025-24873100).

---

📄 License

Released under the MIT License.