Original title: An Evidence-Centered Decision-Making Framework for Explainable Autonomous Recovery in Self-Healing Software Systems
Oleh Pitsun · September 2026
Modern distributed software applications must fix their own bugs, but typical self-healing systems rely on preset rules or black-box models. These setups make it hard to see why the software chose a specific fix. To solve this, researchers designed the Evidence-Centered Decision-Making framework. This system turns runtime observations into structured evidence and maps them onto an Evidence GraphA structured map of observations and their relationships, used to reason about what went wrong.. It then creates recovery ideas and tests alternative actions using a Hierarchical Decision Confidence Model. In an evaluation across 110 decision episodes, the framework reached an overall decision accuracy of 85.45%. Its built-in Safety GateA filter that decides whether the system can handle a recovery autonomously or should ask a human. approved 49 decisions for autonomous handling, which included 34 active recovery actions. Every single one of those 34 actions met the recovery-verification standards. Overall, the framework provides clear justifications and traceable evidence for every autonomous repair choice.
Imagine you're a chef in a busy kitchen and a dish comes out wrong. Normally you just taste it and fix it, but you can't explain why. Now suppose you must write down every clue—the color, the smell, the texture—and then draw a map of how those clues connect. You use that map to decide whether to add salt, change the heat, or start over, and you give each option a confidence score. In this paper, the software does the same: it turns error signals into evidence, builds a map, and then chooses a repair with a clear reason you can check.
When software fails, autonomous systems usually just act—leaving operators in the dark about why. This framework makes those decisions transparent and traceable, so engineers can trust and verify automated repairs. It matters for any critical service that must stay available, from banking to cloud infrastructure.
The abstract does not report performance on real-world production systems, only simulated episodes. It does not specify how the framework scales to larger or more complex systems, nor does it compare against other recovery methods. Autonomous coverage varied across scenarios, suggesting some failures still require human intervention.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
An Evidence-Centered Decision-Making Framework for Explainable Autonomous Recovery in Self-Healing Software Systems — Oleh Pitsun — Automation — doi:10.3390/automation7050152
Abstract licence: CC BY
Explanation generated by Synthesis from the abstract — it is not the original paper.