The research introduces Specified-Foil Counterfactuals, a technique for answering the question of what past conditions would have led to an alternative prediction in temporal graphs. Existing counterfactual explanations typically focus on changing past events to invalidate a prediction. This new method addresses the user’s need to understand the specific conditions that would have resulted in a different outcome. The method identifies a low-cost past-event intervention where the predictor selects a specified foil. It uses trace-guided intervention search, contrasting completed and incomplete executions, mapping differences to DELETE, INSERT, REWIRE, RELABEL, and SHIFT operations, and verifying the foil through exact replay. The research instantiates this principle with LiFTER on continuous-time dynamic graphs and TLogic on temporal knowledge graphs. On CTDGs, the method retains 85.7-93.6% of black-box greedy successes while reducing predictor evaluations by 75.0-80.0%. On TKGs, it reaches the specified foil in 74.8% of 600 comparisons. Executable traces become computational structures for constructing conditions of unselected alternatives.
Source: https://arxiv.org/abs/2609.11170