The SSBSE Challenge Track is an exciting opportunity for SBSE researchers to apply tools, techniques, and algorithms to real-world software. Participants can use their expertise to carry out analyses on open source software projects or to directly improve the infrastructure powering research experiments. The principal criterion is to produce interesting results and to apply your expertise to challenge the state of the art and inspire future SBSE research.

SSBSE 2026 challenge winners will be awarded with a signed certificate and 700 USD price.

Call for Solutions

SBSE for Agentic Systems

We invite contributions to the SBSSE 2026 Challenge Track that explore Search-Based Software Engineering (SBSE) techniques in the context of agentic AI systems. Agentic AI—LLM-powered agents equipped with memory, tool usage, planning, and autonomous decision-making—is increasingly used to support complex software engineering (SE) tasks. However, the agent configuration itself (memory usage, context construction, tool invocation strategies, reasoning depth, autonomy level, cost controls, adaptation mechanisms, etc.) remains largely unexplored and not optimised.

We invite the use of SBSE to optimize agent architectures, memory strategies, interaction policies, and prompting or planning configurations that maximize usefulness for SE tasks (e.g., debugging, code repair, documentation generation, model explanations, safety, hallucination avoidance).

What You Can Do

• SBSE to Discover Optimal Agent Configurations: Use search to find or adapt agent behaviors and configurations rather than hand-crafting them. Examples of search targets include: what information the agent should access, how much context or memory it should maintain, and in what form information should be represented or consumed.
• SBSE for Knowledge Selection & Compression: Instead of equipping agents with unbounded or persistent memory, search for the smallest and most useful knowledge footprint for an SE task. Possible search goals include: minimal memory with maximal task performance, minimal hallucination-inducing knowledge sets, and task-specialized agent memory profiles.
• Multi-Objective Agent Optimization: Use SBSE to balance trade-offs such as: accuracy vs. inference cost, context richness vs. latency, reasoning depth vs. token budget, and autonomy vs. controllability.
• Using Agent Knowledge for Better SBSE Heuristics: Agentic AI assists SBSE by enabling: context-aware mutation and crossover, reuse of past agent experiences or search traces, and adaptive or history-aware evolutionary operators.
• SBSE-Enhanced Continuous Adaptation: Agentic systems adapt over time, not once. Examples include: incremental re-optimization of agent configurations, decay or forgetting strategies, and adaptation to task drift or dataset drift. This is particularly useful for active repositories or continuously evolving software ecosystems.

From Defects4J to SWE-Bench

Much of the existing research in search-based automated program improvement has concentrated on Java, with benchmarks such as Defects4J dominating the field. In parallel, recent advances in generative AI have driven the creation of new benchmarks such as SWE-Bench and its variants (Lite, Verified, etc.), which primarily target Python and have spurred rapid progress through competitive leaderboards.

In this challenge, we invite the community to extend SBSE techniques to the SWE-Bench setting, encouraging cross-fertilization between SBSE and generative AI research and fostering broader participation in this emerging benchmark ecosystem.

What You Can Do

• Benchmark SBSE techniques for bug repair on SWE-Bench
• Utilize the issue reports and additional metadata present in SWE-Bench for search-based test oracle generation or fault localization
• Empirically assess Defects4J against SWE-Bench using lenses such as diversity, complexity, and localization of provided faults, and discuss implications for the SBSE community
• Explore hybrid approaches that combine SBSE with generative AI methods for solving SWE-Bench tasks
• Provide insights into the strengths and limitations of SBSE techniques when applied to modern AI-driven benchmarks

SBSE for Security and Robustness

Modern software systems must operate reliably in the presence of malicious inputs, unexpected usage patterns, and evolving security threats. Both the security research community and the SBSE community have developed powerful yet complementary techniques to analyze, test, and improve software systems.

This challenge aims to foster collaboration between these communities by encouraging the application of SBSE techniques to problems in software security and robustness, including vulnerability discovery, robustness testing, and security-oriented program improvement.

What You Can Do

• Apply SBSE techniques to generate security-focused test cases that expose vulnerabilities and robustness weaknesses
• Use SBSE to automate the generation of security patches
• Study trade-offs between security, performance, and correctness using multi-objective SBSE
• Provide empirical insights into the effectiveness and limitations of SBSE techniques for improving software security
• Benchmark SBSE-based security and robustness techniques on real-world open-source projects

Submitting to the Challenge Track

A challenge-track participant should:

• Perform original SBSE research using or enhancing the challenge programs and/or their artefacts
• Report the findings in a six-page paper using the regular symposium format
• Findings must not have been previously published in any peer-reviewed venue
• Submit the challenge-track report by the deadline
• Present the findings at SBSSE 2026 if the submission is accepted

It is not mandatory to implement a new tool, technique, or algorithm. However, applying existing or new SBSE tools/techniques/algorithms to the challenge programs is expected to lead to relevant insights and interesting results.

Acceptance Criteria

Papers will be evaluated based on:

• Application of an SBSE technique to analyze or enhance the challenge program and/or its accompanying artefacts, or interesting findings on the application of selected SBSE topics to SBSE problems (and vice versa)
• Technical soundness
• Readability and presentation

Submission Details

• Submissions must be at most six pages (including references), in PDF format, and must conform at submission time to the LNCS format and the general submission guidelines provided by the “Format and Submission” section of the Research Track
• Submissions must not have been previously published or be under consideration for any journal, book, or other conference
• Please submit your challenge paper before the Challenge Solution deadline
• At least one author of each paper is expected to register for SBSSE 2026
• In-person presentations are desirable, though online presentations may be allowed subject to circumstances
• Papers for the Challenge Solution track are required to follow double-anonymous restrictions
• All accepted contributions will be published in the conference proceedings

Papers can be submitted through the following link: https://ssbse2026.hotcrp.com

Further Information

If you have any questions about the challenge, please email the Challenge Track Chairs.

Accepted Papers