Abstract: Embodiments are directed to visualizations. A visualization may be provided based on data from a data source such that the visualization includes marks that are associated with from the data source. A mark-of-interest may be determined from the marks based on characteristics of the marks or the visualization. A snapshot of the data may be generated from the data source that may be associated with the visualization and a time that the mark-of-interest is determined. Mark evaluators may be employed to generate evaluation results based on the mark-of-interest and the snapshot data such that the evaluation results may include an explanation narrative or an explanation visualization and such that each evaluation result may be associated with scores that may be based on the evaluation. Evaluation results may be ordered based on their association with the scores. A report that includes the evaluation results may be provided.

Abstract: Embodiments are directed to visualization data. A visualization may be provided based on data from a data source such that the visualization includes marks that are associated with from the data source. A mark-of-interest may be determined from the marks based on characteristics of the marks or the visualization. A snapshot of the data may be generated from the data source that may be associated with the visualization and a time that the mark-of-interest is determined. Mark evaluators may be employed to generate evaluation results based on the mark-of-interest and the snapshot data such that the evaluation results may include an explanation narrative or an explanation visualization and such that each evaluation result may be associated with scores that may be based on the evaluation. Evaluation results may be ordered based on their association with the scores. A report that includes the evaluation results may be provided.

Abstract: Embodiments are directed to managing visualizations of data. Visualizations to monitor may be determined based on metrics associated with the monitored visualizations. Mark values may be sampled from each monitored visualization such that the sampled mark values may be stored with a timestamp. If an amount of the sampled mark values for a monitored visualization exceeds a threshold value mark models may be trained to classify a portion of the sampled mark values associated with the monitored visualization such that the mark models may predict ranges of values for the classified mark values associated with the monitored visualization. If the portion of the classified mark values associated with the monitored visualization has a value outside of the predicted range values, indicating that the monitored visualization may be associated with an anomalous mark value.