Here's what I did:

1) made a note of my server seed hash, as shown on the provably fair tab

2) made up a random client seed

3) played some rolls

4) requested a new server seed, revealing the old one

5) ran my script, which:

  a) checked that the server seed's hash in 1) matches the hash of the server seed in 4)

  b) checked that the client seed in 2) and the server seed in 4) combine to give the exact numbers I rolled in 3)

The generated numbers are the numbers I got when I rolled the dice.

The 5 steps above guarantee that the site had no way of picking which numbers I rolled (since step 2 happened after step 1) and that they actually used the client seed I provided.

It's a little complicated, but it does work.

In the event that I had a run of horrible rolls, I would much rather be able to check that I really was just unlucky rather than harbouring doubts about the site being "rigged". Bad runs happen, and so does cheating. The above process allows you to tell which is which.