1 My math-fu is weak these days, but I've had bad luck reseeding pseudo-random functions instead of just seeding them once at program start.  In my experience it seems to introduce quirks, such as favoring certain number ranges, or even producing pairs of numbers that often appear one after the other.

2 I would not do anything without collecting data first (e.g. a vector of counters for each picture). I haven't seen many dublicates during my visits, maybe even none. The set is small, dups are unavoidable, and your mind may be playing tricks on you.

3

I'm not talking about reseeding. I'm talking about using the number of seconds since midnight, modulo the number of images, in addition to a random offset.

Taking a histogram would be a pain, because the problem looks like the way that they are seeding. So I'd have to store it in a file, and then invoke the test program hundreds of times.

4 Pete, you may be right that I'm borrowing trouble.

5 Why not just use the number of seconds and skip the pseudorandom generator entirely? To eliminate bias from people checking at around the same time every day, just use the number of seconds since the system time epoch modulo the number of images.

6

Boviate, that would mean that if someone reloaded the site several times in close succession, they'd get a series of pictures from the same series.

Folks, this post was not a request for suggestions.

For the time being I'm not going to change anything.

7 Looks like the world didn't end, for one thing.

Are you sure it didn't?

8 This reminds me of how the BSD 4.x fortune program worked: when you built the indexed data file, it shuffled the fortunes once, and then wrote a counter into the file. Every time someone ran the program, it incremented the counter. This turned out to be really painful when 100 workstations mounted the same /usr/games partition, and everyone's dotfiles included a fortune call.

-j

9 I finally saw closely spaced duplicates. They happen, and one who still remembers C-by-n-out-of-m can probably calculate the probability.

Also, I found a link that used to be in the image: there's an equivalent "Home" link in the navigation area on the top right.

10 Actually, now that I think about it, every time you reload the page you're getting a fresh instance of the SWF, so you're always using the first random number generated from a particular seed. Poorly-chosen seeds would produce very un-random results.

-j

11

J, exactly so. (Except that I make 12 calls to the random function and throw them away, using the 13th for the image choice.)

Pete, the "Chizumatic" logo on the flash is the link you're looking for. Clicking that takes you to the base page of this site.

I'm reserving the rest of the flash display space for future expansion. I may want to put other controls in there some day.

12 Ah, lucky 13! I think varying the number of results you throw away each time would be the simplest improvement. Maybe let the RNG run in a loop for half a second or so?

-j

13 So, any complaints from iPad users yet? J?

14 What's an iPad?

-j

15

It's like my Slate, only worse.

16 I expect iOS users who care about being able to see the top rotation would come to the blog less often or stop reading it entirely. IOW, no downside?