1 Occult Academy is not a good candidate. Virtually no fanservice that I recall, no onsen, no beach episodes.

2 Oh, I dunno.  Occult Academy's heroine had one of the nicer looking behinds I can remember.  Not much fanservice, and she was really bossy, but she did have a nice behind.

BTW, take a word of advice from the translator of the show, and cross 11eyes off your list.  It's about as dark and gloomy as you can get, and there's barely any fanservice to be found.  Lots of blood, gore, angst, and all the things you hate.

3 Much as I like the show, I'd probably cross The World Only God Knows off your list for perhaps being a little too chaste for top rotation material.

As far as iM@S and SoraKakeGirl go, episodes 5 and 13 respectively would probably give you the most mileage even if the rest of the show don't prove to your taste.

4

Ninja Nonsense is pretty much a vehicle for Wakamoto-san to get his perv level over 9000, and the art style is pretty simple.  Not sure it has much potential.

Yumeria looks like the sort of show I'd see in the top rotation here, but actually watching it may damage brain cells.

Occult Academy: Mikaze (the waitress) is cute, and she's more in line with what I see in the top rotation than the protagonist is (she's proud of her chest), but she doesn't go past low-cut necklines until some spoilery stuff near the end of the show.  Maya (protagonist) does have a Psycho-shout-out shower scene in the 2nd episode.

Sora no Otoshimono: see Yumeria, but be doubly careful of those brain cells.  (And something like 2/3 of the show is done with chibi artwork.)

(The others I haven't seen, or just agree with what's already been said.)

5

I used Ninja Nonsense before. I own the show and have watched the whole thing. It's on the list, but I probably won't use it. That's one of the ones that have been on the list so long they've gotten dusty.

Yumeria was in the top rotation, too. Problem with that one is too many lolis.

6 Getting senile in my old age... those two are on your reviews page, too.

7 G-on Riders is going to be panty shots and glasses.  That's pretty much the extent of the entire series. 

8 Yeah, I watched about 3 episodes of it a couple years back.

9 As a sad note, I was helping the group that subbed it originally, haha.

It was okay as a comedy.  At least of the stuff that season.

10 Sqa, those features will not come as news to Our Host, although I'd add "broad parody" to your list; pretty much every scene spoofs something. FYI, the official manga adaptation adds lesbian foreplay, light bondage, tentacles, and frequent nudity, with the entire female cast nude for the final battle (lolis included).

-j

11 Suddenly I see a lot more new images, but not from any new series: Hoshizora, for example.

12 I suspect IdolM@ster has some potential. At least if this image from Sankaku is representative. Image safe, but site NSFW, as always.

13 Pete, I haven't done an update yet.

14 Oddly enough, it does feel like I'm seeing a lot of pictures that weren't coming up before, even though I know they've been in the rotation since the last update. For instance, I'd never seen Carnival Phantasm 9, Aika Zero 2, or Koihime Musou 10 until today.

-j

15 What J and Pete said. I noticed it over the weekend.  

16

It's not impossible. The last time I revamped the selection code, I tried to make it so that over the course of a year it would be able to deliver every single one of the images, even if it couldn't do so over the course of a day.

The primary "random" number is the number of milliseconds since midnight. What I don't know is what the granularity of that is. It almost certainly isn't updated every single millisecond; it's possible the granularity is 16.6 milliseconds, or maybe 33.3 milliseconds.

I do a bunch of math on that to eventually make a selection. But I also included an offset based on the calendar. Basically, after calculating an image number, I add another number which is the month number multiplied by 30 plus the day-in-the-month.

The sum of that plus the previous choice number is taken modulo the number of images, and that's how I make my choice.

So if the former calculation is grainy and can't actually pick every single one of the (currently 994) images, at least I can offset it gradually over the course of the year.