MonoTune: Analyzing Trend Deviations through Database Repair

Datasets often violate expected monotonic trends. For instance, average income is typically expected to increase with education level, and housing prices to rise over time. Recent work studied how to quantify such deviations by introducing Aggregate Order Dependencies (AODs), an aggregation-centric extension of order dependencies, and by measuring the extent of violations via minimal repairs. Building on the AOD framework, this paper presents a demonstration of MonoTune, an interactive system for detecting, quantifying, and explaining violations of monotonic trends. MonoTune enables users to upload a dataset, specify an expected trend using an AOD, visualize violations directly in the results, compute repairs, and summarize the changes induced by these repairs. The system supports both optimal repair algorithms with high computational cost and fast heuristic alternatives that trade accuracy for efficiency. To enable interactive analysis, MonoTune further provides an incremental algorithm that produces a sequence of progressively improving repairs, starting from a heuristic solution and converging to an optimal one. We demonstrate MonoTune on several real-world datasets through a collection of scenarios illustrating how the system helps analysts explore expected trends, quantify deviations in both directions, compare the effectiveness and efficiency of different repair algorithms, and gain insights into the nature of the proposed repairs.