SWE-Future: Forecast-Conditioned Data Synthesis for Future-Oriented Software Engineering Agents

2026-06-17

AI Summary

Existing benchmarks for evaluating code agents either rely on replaying historical GitHub issues—risking overlap with training data—or employ purely synthetic tasks that may diverge from real-world developer needs. This work proposes a forward-looking approach to synthetic code task generation by leveraging repository evolution prediction as a conditioning signal: given a repository snapshot up to time T₀, a semantic model predicts future task types, which then guide the creation of synthetic tasks aligned with actual development demands. Experiments across 80 repositories show that 58.1% of the generated tasks exhibit high relevance to future real tasks. Building on this, the authors construct a high-quality dataset of 200 synthetic tasks spanning 61 repositories, effectively balancing novelty and practical utility.

Abstract

Realistic coding-agent benchmarks often replay public GitHub issues and pull requests, making them vulnerable to overlap with model pretraining, fine-tuning, synthetic-data generation, or benchmark-driven model selection. Fully synthetic tasks avoid direct historical replay, but can drift away from real repository needs. We propose SWE-Future, a forecast-conditioned data synthesis method for future-oriented coding tasks. Given a forecast snapshot at time $T_0$, the method uses only pre-$T_0$ repository evidence to forecast future feature implementation/enhancement, bugfix, and refactor task families. We first validate this forecasting step retrospectively: after forecasts are fixed, later pull requests are used only to measure whether the predicted task families match future repository work. In an 80-repository study, the forecaster achieves 58.1\% future-work relevance under the main semantic matching metric. We then use validated forecast families as conditioning signals to synthesize a 200-task coding-agent dataset across 61 repositories from a task-generation snapshot, rather than replaying the later pull requests used for validation. SWE-Future shows that repository-evolution forecasts can guide realistic, future-oriented coding-task synthesis while reducing direct dependence on historical pull-request replay.

Problem

• coding-agent benchmark
• data overlap
• synthetic task
• future-oriented software engineering
• repository evolution

Innovation

• forecast-conditioned synthesis
• future-oriented software engineering
• coding agent benchmark
• repository evolution forecasting
• synthetic task generation