Is there something funky with the algorithm that picks starting 960 positions?

Having all bishops start in the corners would happen one sixteenth of the time, but it seems a lot more common than that. My impression is that if one bishop starts in the corner, it's at least 50-50 the other one does as well, instead of it being 1/4.

How large is your sample size?

Did a little digging, taking this game as an example. I looked at the data sent over the network and there's a json file containing the player moves. The first move contains a field called "jmove" with value "266". I believe this indicates which of the 960 starting positions to use. Cross referencing against this table and it appears to be correct. I feel like the random number generator producing the values in the 0-959 range is unlikely to be biased towards placing the bishops in the corners, perhaps you just saw an unlikely sequence?

I counted 54 but should be 60 out 960

Could not see your last 16 games because my android on Google Chrome fails due to expired certificate

Also game text does not include starting position,
ideally this should match a format that can be fed into a chess engine

of your last 16 chess960 games only one had bishops in corners (average)

P.S. had to navigate back to start using millions of left clicks, the move list does not work (maybe the lack of 8 characters to indicate fischer sequence for whites back row)

The move list links only do what they're supposed to in the old design scheme, which you can switch back to if you want. I have played or am playing 336 games of 960 here. If it uses a random number generator for 0-959, I'm sure it's correct.

If coding for chess960 maybe just start with 8 characters eg RNBQKBNR

and if chess960² happens in the future ... eg RNBQKBNR-RNBKQBNR