My apologies for the delay in replying, due in no small part to the rapidly descending "silly season" (and the longer hours at work) and the time difference which I had neglected to remind everyone of.
Thank you Ricky, RJOH and hypersonic for your advice. Our data here seems to be similar to that of the 6/42 Belgian lottery in that these "crazy" draws seem to occur less than 1% of the time.
The consensus and common wisdom seems to be to leave these draws in and I must admit that, that was what I originally thought. However, my gut tells me that giving weight to something which happens less than 1% of the time, devalues something which happens more than 99% of the time and yes hypersonic, I have taken on board your advices on "smoothing" and the subsequent cumulative "ripple effect". Thanks for that.
I did a quick test by removing the two "offending" recent draws and ran VRA. The resultant graph-shapes under various parameters, closely mirrored that of the drawn numbers (post-dicting, as NemeSys calls it) although intersecting at no more than three points. My repeated attempts to "shift" either the non-intersecting post-dicting data points or all of them globally using different neighbouring dim and del settings, threw the whole lot out of whack.
My thinking is that, as the majority of those testing VRA are more likely to be using all the drawn numbers, I will remove all these "outriders" from the draws I'm presently testing (I don't know if anyone else has done it), and see how this effects the accuracy (or lack of) its predictions.
While it may take me a little while to run the tests, I will get back to you with the results of my little experiment.
Again, thank you guys for your input.
Regards
Tony