We investigate the oracle problem in automated repair of multi-fault programs, focusing on validating partial patches produced during incremental repair phases. Given a multi-fault program P with faults (f₁, …, fₙ) and a test suite T triggering these faults, the goal is to design validation oracles (O¹, …, Oⁿ) for intermediate patches (pt₁, …, ptₙ) that address individual faults.

Validating partial patches raises two major issues. First, such patches are typically generated incrementally by automated program repair (APR) tools and may address only a subset of the program’s faults, leaving others unresolved. As a result, they often fail to produce correct outputs when evaluated in isolation. Second, validation must account for interactions with remaining faults to avoid rejecting valid fixes.

To address these issues, we introduce a new validation framework that blends the strengths of output-based, halting, and assertion-based oracles, some of which are informed by bug reports. The framework provides a modular approach to assess both partial and composite patches in programs with multiple defects by integrating formal reasoning and modeling fault interactions.

This approach lays the groundwork for more fault-aware and interaction-sensitive repair techniques and brings us closer to a deeper, more practical understanding of how to assess patch correctness in complex, real-world multi-fault scenarios.