Week 2 - Awakening

I was on vacation from work this week, so I had a lot of free time. I ended up watching a lot of Mahjong videos on YouTube.

Among those, I watched several videos from the guy who played with Northernlion, yu_song. After having watched ないおトン so much, going to watching a middling player was quite interesting. The difference in play was immediately obvious. And as my goal is above yu_song's level, knowing where those differences lie is valuable. Watching critically with the mentality of, "I must surpass this person."

Outside of simple skill/knowledge difference, I think the most core difference is "flexibility." This was immediately obvious with his, quote, "gigantamax folding," but also manifested in some other places as I continued to watch.

Let me show you some examples about what I mean. Maybe they're obvious to you, or maybe you can learn something too.

Initial Discard Order

Here's a screenshot of a starting hand from one of the videos.

Here, yu_song said, "I don't think this is a half flush either, we only have 3 shapes in the circles suit." He cut the east. Then, his next draw was east, to which he said, "I cut the east wind because it's unequivocally the right decision and then I get blasted by it."

But, I think it is very obviously correct to cut 9s from this hand. It's not honitsu yet, but it could easily become one with the right draws. The benefit of that is much greater than the 9s turning into a bad block. Even putting aside that he's very far behind and needs to create value in the hand, leaving the flexibility for honitsu is some great risk/return, especially with 5s already covering the 7s draw.

The next draws will determine if the hand can go in that direction or not, deciding it now is a bit premature. Similarly, I haven't seen him consider toitoi (all triplets) from the starting hand, nor have I seen him cut yakuhai first in pinfu hands. But, I suppose toitoi dash isn't in Riichi Book 1.

Folding Order

This comes up a lot. I think when yu_song decides to fold, it's a binary choice, with all safe tiles becoming equal. Obviously I can't read his mind, but that's what it seems like from his play. Here's some examples.

Here, yu_song cut 3s. This breaks apart a block and goes back to 2-shanten. The 2s and 8m are also completely safe and leave you at 1-shanten. You can start with those and see how things unfold, with the added bonus that 3s is also a safe tile for the left player in case they chase. It's at least two turns of keeping tenpai chance for free, and in those two turns, more safe tiles could be revealed.

There is a world where you cut 8m, draw one of 47m67p, cut 2s without calling riichi, and then tsumo, escaping the situation without ever putting yourself at risk, but only if you play in a way that leaves that world as a possibility. Of course, you can also get there in a more winding manner. Cut 8m, draw a dangerous tile, cut 2s, the riichi player discards the dangerous tile, you cut it, and so on.

Here, yu_song cut 6m. The hand can be viewed a few ways. Cutting 6m makes it 55m + 789m + 24p + 45p + 234s + 7m. The only tiles that get to tenpai are 36p, though 7m is pretty safe. The 2p is also safe against both riichis. Cutting that, the blocks are 5567m + 789m + 445p + 234s, with 5m and 5p mostly acting as floating tiles. The tiles that get you to tenpai are 345678m34567p. If the tile you need to cut to take tenpai gets shown as safe, some outcomes can be a dama pinfu, which would even allow you to ron.

Even if your next draw is dangerous and makes you fold, that opening is valuable. If you fully fold each time, you will encounter games where you unavoidably get fourth. You never get tenpai, other players keep winning, and you fall behind. By maintaining potential like this, you can claw back some of those formerly-unavoidable fourths.

I think this is the biggest difference between his folding and ないおトン's folding. When I asked an 8-dan player what was the difference between them and tenhoui, they said, "Really high level players feel like they never actually are truly out of a hand like betaori." When shown a discard pool full of safe tiles against a riichi, they said, "In Houou you see this and go, 'well that makes sense,' and then they call riichi or ron you." 

It is not really mawashi or pushing, but folding in a way that preserves potential in a hand. This sort of play also leaves you open to keiten, taking tenpai at the end of the round to avoid the noten payments.

Anyway, enough about that, let's go onto the report.

This Week

Efficiency

- WWYD301 is quite a hard book. I think I could probably read through it 3-5 times and still be getting good value out of it. I've read the first 170 problems once, the first 90 twice, and the first 60 thrice. My success rate is increasing and I can feel my efficiency transforming as I play.

Awareness

- Watching the tedashis is not a hard task anymore. Remembering them is a different story, but I can see when something is tedashi as it happens now.

- I can pretty much immediately identify when a full flush is tenpai, but picking the best discard when 1-shanten is still hard.

- Keeping my hand in my head is still pretty hard, especially with starting hands. I'll usually mix up some floating tiles or shift something by one.

I also read through the rest of zeRo's book. The book gave me the feeling of, "20% of the effort gives 80% of the results." So, I wanted to use it to develop initial calling and push/pull systems to have something to go off of before those arcs. I took notes and also referenced the push/pull charts from the sidebar here and painted with broad strokes to determine when it's correct to push and pull. That leads into my next subject.

Trust the System

I have been giving some thought to how I want to review my games. There's lots of AI reviews available now. In Mahjong Soul, there's Maka built in. There's Mortal, and Naga. But, at least early on, I think I'll use them only as highlighters and not put much thought into their actual opinions. I want understandable, actionable feedback.

If an AI says my decision was wrong, but my system says it was right, I will trust the system. Alternatively, if something comes up repeatedly, I will need to alter the system to include it.

Let's take a look at this push/pull decision from a yu_song video since I already have them open.

A Houou player I asked and Maka both want to push 6m here. The push/fold charts also label a non-suji 6 a must push when you have 3 han against a non-dealer riichi.

However, yu_song follows the principles written in Riichi Book 1, which are very conservative. That's fine, you gotta follow something. So, let's say Riichi Book 1 is the source of truth and go entirely off that.

We'll also ignore that this is South 3 and he's behind. Anyway, yu_song cuts 9s here. According to his system, this is a good fold. Riichi Book 1 defines a high-scoring hand as 7700+, so this hand does not count as high-scoring and is a bad wait. (Most things I've read put high-scoring at 5200, which this also doesn't meet)

Then it comes to this.

Now, we're ready with a good wait, so according to Riichi Book 1, we should push here. A Houou player says, "I would just close my eyes and cut 6m." 3 han good wait is a push vs a dealer on the charts, but the charts don't include multiple riichis. Maka is divided.

With my "trust the system" system, I would consider the 3p to be a bad discard, with pushing 6m being the correct answer. Of course, this system doesn't consider multiple riichis, point distribution, live suji, and so on. But, the system will be fleshed out more in the third month, where I will be focused on improving push/pull decision making.

The things the AI marks will be places that, to me, it is saying, "Hey, you sure about this one?" and then maybe I'll just end up going, "Yeah, I'm sure."

Even if Maka had 3p at 90, I would still consider 6m to be correct. Or 3m, both maintain tenpai and I'm not sure whether double non-suji or a dora single non-suji is more dangerous. Seeing Maka say 3p would not be very actionable. Is it saying 3p because there seem to be three tenpais? Is it saying it because of the point distribution? Is it saying it because it's South 3? Is it because the left player dealing into the across player would be good for us? Maka will not tell me, so I cannot know.

At best, it will become a point of research. I could look up push/fold articles about facing multiple riichis and then try to adjust my system to account for it. But, until I do that, 3/6m will be correct here in my mind.

Next Week

Awareness

- Okay, so I can see when someone does a tedashi, so what? This has pretty limited usefulness on its own. If a taatsu otoshi happens on consecutive discards, it doesn't matter if the first tile was tsumogiri or tedashi. But, if it happens with a few tsumogiris in between, it gets hard. I'd have to remember, at minimum, the previous tedashi. I have some ideas on how to achieve this and I'll see what ends up being easier.

1. Have some sort of visual mental marker on the last tedashi tile. This might be easier to evolve into tracking every tedashi, but I don't think that's necessary.
2. Keep a running count of consecutive tsumogiris for each player. This is basically remembering a three digit number. 203, 014, 120... I tried doing this in one round and it was extremely hard, but I think I could get used to it. When a number changes to zero, I can count back the number of tiles the number was at to see what their previous tedashi was.
3. A suggestion from ChatGPT, remembering the two digit coordinate of each discard for each player. Third tile on row two would be 23. According to the robot, humans are very good at remembering two digit numbers. I guess we'll see if that's true.

- Mahjong Soul gives a rather generous 5+20 time control. I'll take advantage of this and make some checkpoints where I am supposed to use time. This is also adapted from zeRo's book. At each of these, I could use up to 10 seconds, but ideally that shrinks and I can also think about it while others are thinking. I'll write these down on sticky notes and line them up at the bottom of my monitor.

1. On the first discard:
• Check the dora
• Might this hand become a bad wait nomi? (if yes, do something about that)
• Am I calling?
2. On the seventh discard (row 2):
• Anyone being weird?
• Do I have safe tiles for each player?
• How are my bad shapes doing? (eg are there two 3s visible for my 12s penchan?)
3. On the 13th discard (row 3):
• Do I have to fold, or can I aim for tenpai?
• Which tiles would I call if going for keiten?
• How scary are the other players?
4. When someone calls riichi:
• How many dead suji are there?
• Which of my tiles are safe, and what order do I discard them in?
• Start thinking about whether the other players are discarding safe tiles or not

Efficiency

- Keep reading through WWYD301. This will be my "Happy learned how to putt" moment.

Morning session: Reading

Evening session: Checkpoint practice

Night session: Tedashi tracking practice

Maybe after this next week I can start playing some real games. Well, as real as ippan/bronze games can be, at least... 

If you're curious, these are my tentative notes for calling and pushing, that will be refined in the later months:

This errs on the side of pushing, because I like pushing

Pretty much all from zeRo's book