Advice on Sekitori Oracle?

[Shiroiokami]

Was hoping if anyone had any advice on improving my SO results? I made a W/L matrix for it last basho but I only had middling success with it, not sure what else to do either than just eyeballing each player using H2H website.

[Randomitsuki]

Some comments from a former gamer...

A while ago Jejima created some controversy by predicting an 8-7 for each rikishi on the East side and a 7-8 for each rikishi on the West side, and getting pretty good results out of it. While this outcome made some people question the validity of the entire game, it rather showed to me that Jejima's move embodied two key principles that not-so-strong players easily overlook:

1) Being too daring does not pay off (only in one's dreams). If you think that someone will finish 2-13, do not enter 2-13 because it will kill your results if that rikishi ends up with going 6-9 or better. Using a purely mathematical approach, hardly any rikishi had less than 5 expected wins in my computations. This is particularly true for low-performing rikishi, and not quite so much for high-performing rikishi. During his prime, 13-2 or 14-1 was a nearly safe bet for Hakuho, but even for typical Ozeki I rarely predicted double-digit wins. Take home message: do not deviate too much from 8-7 and 7-8 unless you have very good reason for it.

2) I always find myself surprised at looking at SO standings for Day 1 as these (completely irrelevant) standings invariably feature the same bunch of players on the top. Why? Because they violated one of the most obvious rules - the average number of wins for all your rikishi in a divison should be 7.5 (or slightly lower if you figure in potential kyujo). Just for illustration: if you would predict all rikishi to finish with 5-10, you will be guaranteed to lead the standings after Day 1 as much as you are guaranteed to end up dead last on senshuraku. So keep an eye on your average points.

[Oshirokita]

If you don't have a good read on someone definitely likely to get more than 8-7, then guess 7 or 8. I often guess alternating 7s and 8s for the Juryo division, for example.

If you are absolutely certain a rikishi is likely to get 10 or more wins, guess one win less than you expect (i.e. 9 for a likely 10 wins). It is rare for even an elevator rikishi to outperform their expected results, but always possible to get injured and get less.

[Gurowake]

As someone who was one of the top Oracle players back when I was playing, I can just say that Randomitsuki basically said all the general high level advice that I would have given.

If you're more serious about the games in general, you'll need to develop a method of determining the probability that each rikishi wins in a match.  Determining which rikishi is going to win is generally less important than determining the probabilities that each rikishi will win.  My own experience, and this is generally played out by the Seki-Toto results, is that matches are only roughly 60% predictable, which isn't much better than guessing at random and most of the deviation from 50% comes from the few rikishi that are at the top of the banzuke, out of their depth at their rank, or coming back after having sat out a tournament.  As such, given that Oracle is an aggregate game, you always need to take into consideration that a rikishi that you think is going to win a match really will only win 0.6 matches or whatever per bout that you're favoring them if you're adding multiple results together to get a combined amount.

In conjunction with that matrix of probability of wins for each potential rikishi matchup, you could then move on to guessing how likely it is each rikishi bout actually happens, and make a consistent matrix of match probabilities to multiply by the win probabilities for each match and then sum by rikishi to get your total number.  If you think this sounds like too much work, that's exactly why I stopped playing the games.  I didn't feel like doing it any more, but I simply couldn't imagine trying to make any guesses without such a system after having used it for a few years to get very good results.

Edited May 7, 2024 by Gurowake

[Sumo Spiffy]

I have no idea what these win matrices look like, but you can eyeball something similar using the guidelines already mentioned. At first I used to just stick everyone at 8-7 or 7-8, with a rare 9-6, but I'd only do OK. The improvement I've made the last two bashos was to do the same thing, but allow myself to stretch things out a bit. If I was absolutely convinced a guy would have a losing record, I just didn't know what, I'd go to 6-9. Likewise, if I was equally convinced someone would have a winning record, I'd go to 9-6. The reason is that you want to aim for the middle of each wrestler's win range. For most of them, that's seven or eight wins, but if you see someone where you're absolutely positive it'll be different, then go that way. The end results are decided by very thin margins, so if you're looking for top-ten finishes, you have to be able to gauge which guys are outliers and give them appropriate credit (or "credit" for the low end, I suppose).

That said, I don't know if I've ever picked eleven wins for anyone (I wasn't around in Hakuho days). I've done a few tens, mainly for guys who looked set to completely steamroll the competition, but that's it. No matter what you expect, there's a point at which skewing too far off 7-8/8-7 is unnecessarily risky, especially when you see a situation like last basho where Kirishima was expected to do well but it turned out he was three broken bones in a trench coat.

[Jakusotsu]

Just like with many other games, it depends on what you aim for: low risk strategy for a simple KK or go out on a limb for the (unlikely) yusho.

[Ganzohnesushi]

rrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrr

Edited May 8, 2024 by Ganzohnesushi

[Ganzohnesushi]

seems my PC is kaputt.

Edited May 8, 2024 by Ganzohnesushi

[Oskanohana]

On 08/05/2024 at 01:17, Gurowake said:

If you think this sounds like too much work, that's exactly why I stopped playing the games.  I didn't feel like doing it any more, but I simply couldn't imagine trying to make any guesses without such a system after having used it for a few years to get very good results.

You could grow a pair and guess a little bit on the fly and simply have fun playing . (Insert meme of "the virgin spreadshitter" vs "the chad guesser").

In the past I thought of building that kind of spreadshit (that's not a typo), but apart from being hard work, I thought it'd suck the fun of it. Glad to be proven right.

 

To the question originally posted, Rando and Spiffy gave already pretty good answers. Still, you can always pick some other player you like and compare his past bets to yours (and you have all the historic data on the site if you want to go further down the rabbit hole) and maybe you can see where you're going wrong, if you're being too tame or too impulsive.

So why do all the job Gurowake said if you can just copy his results and tweak them to your liking? It's not corporate espionage or anything. Even that seems way too much work for me though, I'm too lazy for that. I'm a cold guesser type (with attention to my sums, of course), but what do I know, it's not like I'm the lone yokozuna on the current banzuke for the game or anything.

[Anjoboshi]

From all the games I have been playing in the last 20 years, I can safely say that Sekitori Oracle is the only game where I have absolutely no clue what I'm doing. :D

(Me, every time I fill out the entry form for SO: *insert the "I have no idea what I I'm doing"-dog-meme*)

[Pitinosato]

I play this game to see if my predictions were good or bad

not to win the yusho or do maths

So no advice from an average SO-Player 

Pitinosato