This question comes up from time to time: You've got a bunch of analysis rays and are running multiple case. If you are using ZPL keywords or Merit Function operands, like NSTR, it is not enough to set the seed to zero. This is necessary, but it is not sufficient.

You must also go to the properties of our source and in the Raytrace group, under Sampling Method, choose the Random option. There is also a Sobol option that has many useful properties for getting pseudo-random and less clumpy data, but Sobol will always produce the same sequence of rays, all else equal. This is a property of the source itself, and it will overrule any random seed settings in operands that call the source.