Show HN: Kronumos-Neurosymbolic code repair with a 5µs Rust Sub-Cortex (zenodo.org)

Kronumos 2 Kairos has made a significant leap in Automated Program Repair (APR) by introducing a cost-bounded, dual-brain cybernetic system that combines a powerful open-weight code model with a highly efficient runtime engine. This innovative system tackles the complex challenges of accurately diagnosing defects, generating precise syntactic fixes, and maintaining the integrity of existing functionalities. During evaluation against the Princeton SWE-bench Verified dataset, it successfully proposed 442 candidate patches while maintaining a 100% verification rate, marking a substantial advancement over traditional APR methods. The significance of Kronumos 2 Kairos lies in its innovative architecture, which features numerous advanced components, including an Issue De-Noiser and an Interlocking Causal Invariant Mesh. The system demonstrated remarkable efficiency—achieving bug resolutions at zero marginal inference cost and reducing token use by 93.5% compared to existing industry benchmarks. Notably, its AST Healer component contributed to a 33% improvement in bug resolution capabilities, underscoring the importance of deterministic post-processing within APR frameworks. This work not only enhances the robustness of software development practices but also propels the broader AI/ML community forward in addressing practical software maintenance challenges.