Seki-Toto and Oracle game play issues

[Asashosakari]

As already alluded to during the basho, there are a few things that may merit changing in the Seki games to improve the game play a bit, and because the previous brainstorming sessions were pretty enlightening and I really don't want to introduce changes that nobody cares for, here goes again. Warning, long and boring post:

1) As Toto players will know, players who are absent for their first day are kept in the results lineup but score zero points and an automatic loss, and are only removed completely for their second consecutive absence. The thing is, since the scoring is median-based two more absent players also means one more daily winner so in any basho there are more wins than losses. The effect wasn't so bad in Kyushu basho just now (768-744), in part because the server worked almost perfectly aside from the totally lost Day 2, but any type of server issues generally increase the differential significantly because they increase the number of temporarily missing players. Last three basho: 760-711, 856-728 (oh, wow), 897-813.

The great majority of those absences are by makushita players. Not a huge surprise since on average the makushita players are less game-dedicated which means that they miss more days due to unrelated reasons, and are also less likely to keep trying (or emailing picks) in case of server problems. That part can't really be helped, short of praying for the server to hold up. The bigger issue is what this does to the other divisional scores:

a) makuuchi and juryo players tend to miss very few days, so the number of (banzuke-relevant) losses isn't increased much by that,

b) they tend to score better than the active makushita players, so they're usually above 7.5 wins on average, sometimes significantly,

c) because they score better, they also profit disproportionately from the additional wins created by absent players, which increases their records even further

Potential solution: remove absent players from the scoring immediately. I'm actually not sure what the original rationale was for keeping them in, especially since for the first year or so they even stayed in for 3 days instead of just one which obviously would skew the W-L's even more. Preparation for a future addition of auto-picks that never came, or maybe done to mimic real sumo? (Or perhaps simply done because ISP already handled it the same way?)

Anyway, unless there's a compelling reason I can't think of, I wouldn't mind changing that so missing players are marked as totally absent and un-scored immediately in the future. Kofuji seemed to indicate that this would actually simplify the automation a bit, so that's a bonus.

---

2) I'm a bit tired of seeing the same four or five people competing for the Toto kanto-sho basically every basho. (You know who you are! And for those who don't, just look at the sansho stats...) I was trawling through the archives of the defunct Toto mailing list earlier this year, and it seemed that several years ago Zenjimoto once reached the same point of frustration. Anyway, his suggestion at the time was to only count the risk factor points for days on which a player actually scored a win, but nothing came of that idea (implementation issues?) and IIRC the KK qualifying requirement was introduced at that time instead. Unfortunately that's not having much of an effect; at least one of the players in question is able to reach KK nearly every basho so the prize rarely travels outside of that group.

So, unless I'm the only one who is bothered by it, I'd like to see that change. I still need to check if the "winning days only" idea actually produces desirable results (I'm a bit afraid it might give the sansho to yusho or jun-yusho players too often), so in the meantime maybe somebody has a better idea for how to adjust the kanto-sho scoring.

---

3) Another thing I'm a bit annoyed by is the records distribution for the makuuchi division in Oracle. Back after hakkeyoi.net keeled over I reverse-engineered Kashunowaka's scoring formulas for use in Golynohana's new automation, because they seemed to work pretty well at the time, and it was the quickest way to get the game revived. However, the makuuchi division has become so competitive (and maybe also risk-averse) over time that extreme records have become almost impossible, especially in the KK half - not a single player managed to get a better converted record than 12-3 in makuuchi this year, not even players whose score was strong enough to win the overall yusho. But it also matters on the MK side, because there's always a huge number of 6-9 and 7-8 records and few more extreme results. Put more simply, tons of players tend to score very close to the median, on both sides of it.

Who knows, perhaps I'm just overreacting because I'm tired of scoring 9-6 all the damn time. Anyway, obviously there are banzuke issues - really difficult to get a good handle on ozeki runs when it takes a yusho-quality performance just to get 12 wins and even 11 are a rarity, relatively few demotions to juryo because of all the 6-9's and 7-8's, etc. - but even beyond that the sheer notion that a pre-basho game never awards anything better than 12-3 just rubs me wrong somehow.

Easy solution, just make the near-median intervals smaller. Right now the KK records are all spaced at 3 points per win (e.g. if 8-7 starts at 550 pts, 9-6 starts at 553, etc.), and the MK records at 4 points. I'd probably try to reduce that by one point each for all records between 6-9 and 9-6, perhaps also for 5-10 and 10-5 or even further depending on how the results work out. Again, unless I'm the only one who sees a problem here.

---

Comments welcome, and thank you for reading if you actually made it down here. :-)

[Kofuji]

1) It was pretty expensive to look back 3 days to see if someone had played, not so much for 1 day and cheapest not to look back at all, so it's a fairly trivial change. I think since the scoring is done once per day and saved, this shouldn't affect the archives.

As far as auto-picks go, given the speed of the new server, I could entertain putting ISP-style auto-picking into ST. I don't think Max was particularly happy with that style of auto-picking and I'd have to dig out the "Maxi list" to check what he suggested a few years back. What would be easiest would be if all ST players also played ISP and used the same formula for both. In fact at one point we proposed that we could automatically pick the ISP bout for someone from their ST picks as long as they didn't X them ... could even highlight the ISP bout on the entry form.

2) Adding a winning day only would just make things worse IMHO. It is fairly trivial to implement. I think there was a whole separate new way of doing this sansho (I think) but I never implemented it online since it was very expensive to calculate from scratch each time and I don't have a mechanism for saving the daily sansho results built. I know I did program it. I'll have to check what it was and get back to you on it. It may be that with the better server, it won't be such a big deal.

3) I always suck at SO and only play it to set myself up for the rest of the daily games and my Bench squad. This tells you why I suck at most of the games. I was terrible in Hoshitori, UDH and RotoSumo this time because of bad decisions there and my Bench squad wasn't anything to write home about either. For some miraculous reason, I actually did OK at Fantasy Sumo and GISP ...

Looking at my scoring, I think this might be better solve with a non-linear score for how far off you are ... say 25-20-15-12-10-9-8-7-6-5-4-3-2-1-0 ... that should serve to spread things out more I would think, but I haven't tested it.

At least I don't have to worry about coding this one ;)

Warm regards,

Kofuji

[Gernobono]

1) why not only set the median at all entering guesses and giving all non-participants a kuroboshi?......this will of course lead to more mk than kk at the end, but i guess those not playing will not care.....

[Bogdanyama]

Why not put as an option for those players that are also playing ISP? If they choose at the beggining of the basho, when they loose one day entry the system apply the auto-picking option set by him in ISP. This of course will be in favour of the multi-games players that normally don't loose daily picking, but desasters happens...

Bogdan

[Saruwatari]

In fact at one point we proposed that we could automatically pick the ISP bout for someone from their ST picks as long as they didn't X them ... could even highlight the ISP bout on the entry form.

I don't think you should make that particular link between the games. You might wanna pick differently for these games. In general you choose the one you think will win, but in ST you might wanna pick the other guy for risk factor reasons, and in ISP you might wanna gamble on the other guy, because nobody else does.

Using the ISP auto-pick formula seems like a fine idea. Another idea for auto-picks is that a player gives a rating to all 70 wrestlers, but that may be too elaborate. A third idea is that a player selects 35 wrestlers. If one selected and one unselected face each other, the auto-pick will be on the selected. If two selected or two unselected face each other, the auto-pick will be an X.

[Kofuji]

Using the ISP auto-pick formula seems like a fine idea. Another idea for auto-picks is that a player gives a rating to all 70 wrestlers, but that may be too elaborate. A third idea is that a player selects 35 wrestlers. If one selected and one unselected face each other, the auto-pick will be on the selected. If two selected or two unselected face each other, the auto-pick will be an X.

I think your 3rd proposal is where Zenjimoto ended up years ago.

[Asashosakari]

Geez...one small off-hand comment about autopicks and everybody responds only to that. (Whistling...) FWIW, I'm not totally opposed to autopicks, but I don't think I'd favour a full integration with ISP. Maybe with lower limits though...up to 5 autopicked days seems okay in a game like ISP where it's a 50-50 proposition anyway, but I'd like autopicks to be an emergency thing for Toto, not something people can rely on for significant parts of the basho. Maybe 3 days maximum? I guess the ISP auto-pick structure would be fine if it's feasible to implement, but Saruwatari's third idea sounds fine, too.

Anyway, I've just finished a quick analysis of the Oracle results of the last two years (since the move to Golynohana's automation), with some more or less surprising results:

- The number of 8-7 and 9-6 records is actually right on the mark, compared to real-life makuuchi. Guess I only have myself to blame for scoring 9-6 all the time.

- Double-digit records are a bit too low, but since 8-7/9-6 are fine that's mostly because the overall KK frequency is lower (51.5% in Oracle, 53% in Ozumo). Nothing really to be done about that without screwing up the rest of the balance...

- As expected, not nearly enough records above 11-4.

- Too many 6-9 and 7-8 records, also as expected.

So, barring vocal opposition, the points schedule in makuuchi for each record "bracket" will be changed as following:

Rec	old	new
1-14	4	4
2-13	4	4
3-12	4	4
4-11	4	4
5-10	4	3
6-9	4	3
7-8	4	3
8-7	3	3
9-6	3	3
10-5	3	3
11-4	3	2
12-3	3	2
13-2	3	2
14-1	3	2

That will hopefully push a few 7-8's and 6-9's into larger MK records, and space out the really good results between 11-4 and 14-1, instead of concentrating everything at 11-4 and 12-3.

[Kofuji]

Well, I finally found the Maxi and Mini lists for Sekitori-Toto. They arise from the month before I got employment in UAE. Most of the mini-list got implemented, not much of the Maxi list.

Bet you thought these would never see the light of day again, Max! (Whistling...)

Mini list (17 Oct 2002)

Dear Kofuji

As threatened, the harmless email to remind you about adding the SO and

SB links on the sumogames.com links page also includes my ominously

growing "mini" wish list for various fixes... heh, the list started

with 3 items, but as time progressed has grown into a size that is

probably no longer a "mini" list :) I've had it on my computer at work

probably for a month now :)

Well, here it goes...if you have any time and/or spirit to chip away at

it, that would be great... if not, no problem either of course - some

of them are just my usual stickler thingies :)

"Mini" wish list for Kofuji:

**Please** :):

- Change all the front page, rules page and links page links back to my

server so I can update them and they don't need to be mirrored on to

sumogames.com all the time

- change the results page back to where pre-basho entries (newcomers

and people who registered for the next basho) are once again visible

- make the 'o's demarcating wins on the results page capitals ('O') for

better visibility, and change the 'x' demarcating that a player has

made their daily entry back to the 'p' - as it was at the very

beginning of the game.

- change the colors for wins and losses (in the individual

picks-breakdown page) back to what they are on the NSK site (dark grey

for a loss and red for a win), 'X' picks should be beige (as the

"losses" are now)

- add a link to the medals page (anasuya.com/toto/medals.html) and to a

new stats page (anasuya.com/toto/stats.html) on the main "archive" page

- change the shukun-sho to a new formula ... using the best daily

"percentage" of patrons (i.e. one day the winner may score 23 out of 30

= 76.66%) – the player with the highest such score from any given day

would get the shukun-sho (as it is a real outstanding performance) OK,

this is perhaps an item that belongs on the "maxi" wish list... :)

- don't show nations in the results list that are not participating

- add Sekitori-Oracle and SUPER BANZUKE sites to the sumogames.com

links page, and change the 'Makususuchi' to Zenjimoto

- send me a new list of all patrons and their email addresses

**Please** :)

I know what you're thinking... if this is the "MINI" list, then what

the heck is on the "MAXI" list, right? :D (answer: the auto-pick

system, and the far-away dream of the heya competition) :)

[ducking for cover]

Here is some text for the sumogames.com links, if you need it:

Sekitori-Oracle is the sister game to Sekitori-Toto. As in

Sekitori-Toto, you try to predict the outcome of bouts for each basho –

but this game is played before the basho (not during), guessing the

final win-loss records of the 66 Sekitori (Makuuchi and Juryo Rikishi).

The one closest to the real numbers wins the Yusho, special prizes and

a nations' competition add to the fun, and the Yusho winner gets a REAL

trophy sent to them! This game was started by Zenjimoto during Aki

Basho 2002.

The SUPER BANZUKE is an overall ("World") Ranking which includes all

the "prediction-type" internet sumo games currently active (all of

which can be found on this page!) , combining points players accumulate

in each individual game, to arrive at a ranking of players reflecting

their "overall" gaming abilities.

Better send me your mailing address down there, so I can send you more

briberies :)

Thanks again for all your help... when can *I* do something for

*you*??? :(

Cheers :)

Zenjimoto/Max

Maxi list (24 Oct 2002)

Dear Kofuji!

Well, since you asked for it, here is the infamous and much-feared

"maxi" wish list ... surely making you hope that you'd never asked for

it in the first place :D

I have taken a "throw it on the wall and see what sticks" approach here

:) ... well, here it goes:

"Maxi" wish list for Kofuji:

- auto-pick system per RTR scheme ... I still think that this would be the

most efficient way of providing auto-picks that allow for the

occasional while-away win ... - provided that you do not feel that this is

an unnecessarily huge amount of work cause the

Edited November 30, 2008 by Kofuji

[Asashosakari]

2) Adding a winning day only would just make things worse IMHO.

Turns out you were perfectly right. Counting winning days only for the kanto-sho results exclusively in winners who: 1) Are either tied for the yusho score or at most one win behind, 2) Are two wins behind but still 12-3, or 3) are the same player as with the old formula.

So that's no good. I'm still open for ideas...

Meanwhile I'll see if something like points multiplied by risk factor (either daily or for the whole basho) is any better.

[Asashosakari]

Meanwhile I'll see if something like points multiplied by risk factor (either daily or for the whole basho) is any better.

I give up. (Whistling...) Guess we'll leave everything as-is unless somebody has a flash of inspiration...

[Kofuji]

I hesitate to mention this, but I just discovered that in SQL Server 2005 there is an easy way of calculating medians by division ... The technique may also make calculating Nations standings viable again.