I'm sorry, but I'm still tinkering with the rating system off and on. In about one hour when the midnight (CET) update goes through, you will see some changes. The leaders will be about 100 points lower than before, for one. There will also be slightly more foreigners in the top 40. Hopefully most of you will not consider this a grave insult to your favourite Koreans.

The aim of this modification is to tweak the rates at which ratings adjust. Players who play rarely will now have their ratings adjust faster than before, and players who play more often will adjust somewhat slower. Overall, most of the players fall into the latter category, so I have also increased the uncertainty floor and the uncertainty decay rate to speed up adjustments over the whole player pool. You will see that uncertainties will be a bit higher than before.

There is yet another change that I am looking at making, but hopefully that won't have a big impact. More on that later when and if I get it to work.

The predictive power has not been affected, and is even slightly improved in some cases, such as very large skill gaps.

So, since last time I wrote a blog post, a lot has happened.

Most importantly, the rating system has been tweaked a bit following the discovery of a bug. Now it is much more conservative than before. The site will also provide continuous updates every six hours instead of every two weeks. The matches are still grouped in two week periods, but the preview of the next list is now kept up to date, so you can see the ratings change in "real time".

I have also opened for everyone to submit results to the site. Hopefully this will lower the bar for the general public. If you see anything missing there's literally nothing to it other than to submit it. Don't worry too much about the details, we can work that out ourselves when we review your submission.

The event catalogue has now grown quite large, and two thirds of the database has been catalogued. This is slow, hard work, and we need every man we can get. The result will be something quite unlike anything that exists.

Speaking of which, the Aligulac database is already unlike anything else. It's the most complete publically available pro/semipro SC2 database in the world. I only know of one other database which is bigger, namely the internal Blizzard one, which is hardly public and wouldn't be very useful even if it were (for these purposes, anyway).

Other than that, most other updates are in usability and minor features. The prediction system now supports four formats, and player/results search has been heavily improved.

Ferguson Mitchell of ESFIWorld was kind enough to publish an interview with me. It is not extremely current, but probably a good read anyway.

Since last time, some things have happend. First, we have a new rating list. Since there weren't so many games over the new year, it looks almost the same as the previous one. I have also changed the fonts a bit, and made the site fully zoomable (try it).

Several of the menu options should now have submenus, and a slightly red colour will indicate the currently selected pages.

We have a Hall of Fame page. On the rating list you can sort by any of the four ratings. (I will add filters in the future.) You can search the results database. (This feature has room for improvement, which will come eventually.)

Most excting for me is the event interface.

In the database there are more than 29,000 matches, the majority of which have no event information associated with them. This is a shortcoming that I hope to fix. New games are added with event information, but this is only textual, and may not be consistent over a whole event.

For this reason I have created the event system, where events are arranged in a hierarchical fashion (e.g. GSL -> Season -> Code S -> Ro32 -> Group A), and matches are linked onto this chain wherever appropriate.

The downside is of course that matches have to be assigned to events manually, and it's pretty daunting to search through 30,000 matches to find the ones you need. Currently, about 6.4% of the matches are assigned to an event object, and 20% have any kind of event information at all.

Hopefully these figures can be pushed close to 100% some time in the future.