EAGER: Data Debugging

如今，系统对数据的依赖程度与对处理该数据的软件的依赖程度一样严重。 与软件不同，数据无法轻松测试或分析其正确性。部分问题在于很难判断数据是否有误。 印刷错误可以按数量级改变数据项。例如，当正确值为12.34时，可以输入数字1234。不幸的是，通过手动数据审计发现这种错误是繁重的，不可扩展的，并且容易出错。 数据错误可能代价高昂：电子表格数据中的错误导致了数百万美元的损失，据估计，数据质量差每年给美国经济造成的损失超过6000亿美元。数据调试是一种利用数据和程序之间的交互来定位可能的数据错误的新方法。由于不可能先验地知道数据是否不正确，数据调试的目标是做下一件最好的事情：识别对计算有异常影响的数据。直觉上，对计算结果有过度影响的数据要么非常重要，要么是错误的。相比之下，错误数据的存在对最终结果没有特别异常的影响，这并不重要。 通过引起对具有非常高影响的数据的注意，数据调试可以提供对数据和计算的洞察，并揭示错误。数据调试特别适合于数据密集型编程环境，如数据库和电子表格，这些环境将数据和程序进行交互。 数据调试可以大大降低人为数据输入错误或数据损坏的风险，提高数据计算的可靠性，并可能为美国经济节省数百万美元。

Today, systems rely as heavily on data as on the software that manipulates that data. Unlike software, data cannot be easily tested or analyzed for correctness. Part of the problem is that it is difficult to decide whether data is wrong. Typographical errors can change data items by orders of magnitude. For example, the number 1234 might be entered when the correct value is 12.34. Unfortunately, finding this kind of mistake via manual data auditing is onerous, unscalable, and error-prone. Data errors can be costly: Errors in spreadsheet data have led to million dollar losses, and poor data quality has been estimated to cost the US economy more than $600 billion per year. Data debugging is a new approach for locating likely data errors by leveraging the interaction between data and the programs that operate on it. Since it is impossible to know a priori whether data is incorrect, data debugging aims to do the next best thing: identifying data that has an unusual impact on the computation. Intuitively, data with an inordinate impact on the result of a computation is either very important, or it is wrong. By contrast, wrong data whose presence has no particularly unusual effect on the final result does not matter. By calling attention to data with inordinately high impact, data debugging can provide insights into both the data and the computation and reveal errors. Data debugging is especially well-suited for data-intensive programming environments like databases and spreadsheets that intertwine data and programs. Data debugging can dramatically reduce the risks of human data entry errors or data corruption, increase the reliability of computations over data, and potentially save the US economy millions of dollars.