When describing the accuracy of the flat field, one describes it over two length scales: on the scale of individual pixels and on the scale of the entire array. As discussed above, the pixel-to-pixel accuracy is around 0.2-0.3%. On the scale of the entire array, the accuracy of the flat field is of the order of 1-3%. This is measured by placing a bright standard over several positions of the array and by measuring the flux of the standard after the data has been flat fielded (this test is sometimes called ``the thousand points of light test''). Alternatively one can use the science data itself, if dithering has been used and if there are enough sources that are sufficiently bright. The large scale accuracy of the flat field can be improved to the 1% level by using an illumination correction.
Scientific data should be normalised with flats that were taken within one week of the scientific data. Beyond this, the relative accuracy at which pixels can be normalised slowly degrades.