This is just a link post. The excellent and excellently-named Data Colada blog has a brilliant analysis of scientific fraud exposed by the raw data. Figures can obscure flaws that are immediately obvious in the numbers. (Although, Matt Terry's awesome and hilarious Yoink might alleviate this.) In this case, averages of four numbers turning out to be integers every single time, and two independent experiments giving almost exactly the same distribution of values. (Frankly, if you can't simulate random sampling from an underlying distribution, you don't belong in the fraud world!)
The post demonstrates the importance of publishing as much data (and code) as possible with a paper. Words are fuzzy; data and code are precise.
See here for more.