Issue JP-4168 was created on JIRA by Ned Molter:

In wfss_contam, the wavelength spacing used to disperse sources is determined based on an attempt to find the "native" wavelength spacing at a given point on the detector.

The computation of the native spacing is as follows:

• convert direct image pixel of interest to sky, then to the "direct-image" plane of the grism detector
• find the location of that pixel in the dispersed frame for the minimum wavelength of interest (wmin)
• find the location of that pixel in the dispersed frame for the maximum wavelength of interest (wmax)
• figure out how many pixels apart that is
• use that to determine the number of wavelengths you'd need to adequately sample that many pixels
• make that delta wavelength smaller by the oversample factor (typically 2)

The native spacing doesn't change very much across the detector, so when we moved to vectorized computations of the dispersion in chunks, it was decided to just take the 0th x,y value in each chunk and assume it's representative enough for purposes of determining the native spacing.

However, this has the unintended consequence that changing the chunk size changes the numerical output of the step by up to ~1%.  This should be fixed.