DISCOVER: Distributional Counterfactual Explanations Solver

In today's artificial intelligence ecosystem, the transparency of predictive models has become an indispensable requirement, especially in regulated sectors such as finance, health, or insurance. Counterfactual explanations allow us to understand what minimal changes in input data would modify a model's decision, offering a clear path for auditability and continuous improvement. However, traditional methods operate at the individual instance level and depend on gradients, making them useless against non-differentiable models, such as decision trees, random forests, or hybrid pipelines. DISCOVER emerges as a model-agnostic solver that extends the distributional counterfactual explanations (DCE) approach, replacing gradient-based optimization with a proposal-and-selection search scheme with a limited budget. This technique calculates per-row impact scores —leveraging a sample-based decomposition of the optimal transport objective— and applies a top-k intervention budget, concentrating edits on the most influential samples. To guide candidate generation without needing predictor gradients, DISCOVER introduces conical sampling guided by the transport geometry on the input side. Experiments on multiple tabular datasets demonstrate strong joint alignment of input and output distributions, bringing distributional counterfactual reasoning to modern black-box pipelines. From a business perspective, having tools like DISCOVER allows organizations to validate automated decisions without relying on specific model architectures. At Q2BSTUDIO, a software and technology development company, we integrate these advances into our AI solutions for businesses, helping our clients build custom applications that incorporate robust explainability. Our team combines artificial intelligence, custom software, and AWS and Azure cloud services to deploy systems that not only predict but also explain themselves. Additionally, we offer business intelligence services with Power BI and AI agents that facilitate the visualization and monitoring of these processes. Cybersecurity is also key when handling sensitive data in AI environments; that is why we include protection practices at every stage of development. In short, DISCOVER represents a significant advance for the practical adoption of counterfactual explanations in real-world environments, and at Q2BSTUDIO we are prepared to implement these capabilities in projects that require rigor, scalability, and full compatibility with existing infrastructures.