On Self-Appraisal

Type: Play Author: Mathdino

History

Original Publication: March 25th, 2018 by Mathdino

Original Thread.

Everyone On This Site Needs A Goddamn Reality Check

I'm not currently a Great Player.

As of writing this article, I can carry games in the Newbie Queue and the Open Queue, and there are a whole host of players I can read fairly well. I'd like to think I'm pretty good at reading newbies/lynchbait players. Part of why I don't play in the Newbie Queue that much is because I don't think me using newbie tells on everyone is that representative of the rest of the site.

But outside of those queues, if you, for example, told me to get a solid confident read on xRECKONERx, or RadiantCowbells, I'd look at you like "Are you crazy?" Same goes for most 2016/17 players.

Playing Town Is Hard

Playing town in forum mafia is a game of probabilities (credit to mastina). Your job is to process a MASSIVE amount of input, weigh the evidence, and come out with people more likely to be town and more likely to be scum. Charisma and getting lynches is a different skill. The first step is scumreading scum and townreading town.

But how often do you do that? Really. Actually. Go back through your game history on, say, D1s, and check your top scumreads and top townreads. How often are they correct? Here, I'll go through my game history post-hiatus and check for myself:

Three Of My Games

Micro 764: Marked For Death 2: I was, at different points in D1, scumreading Not_Mafia (off policy), chesskid's slot, and TheGoldenParadox. 2/3 of these were town. I was townreading northsidegal, JaydragonKing, and Gamma Emerald, all correct.

Open 704: Switch: Weird circumstance because I had a N0 cop guilty on Bins. Decided she was scum with Raya36 and havingfitz, both town. My top townreads were Almost50 and Lalendra, both not mafia (A50 was SK playing for town), with Elmo, mutantdevle, and Maxous as my next townleans. Elmo was scum.

Open 707: JK9++: My D1 scumreads were RedFlavor, JaydragonKing, Assemblerotws, and Creature. Creature was scum, so that doesn't count, so we'll say RedFlavor scum means one of those was correct. My townreads were momo, Almost50, Hawk, and UnaBombaH, all correct.

Results

These were strange circumstances at times, but we'll run with it. 3/8 correct scumreads (not counting cop guilty or Creature) = 37.5% 11/12 correct townreads = 92%

Completely random D1 reads gives 25% accurate scumreads and 75% accurate townreads. I did reasonably better than that. Should I be totally certain of my scumreads on D1? Of course not, I've been wrong before and have mislynched town on D1 a bunch. But what if everyone were 37.5% accurate?

I wrote a program that plurality lynches the town's top scumread on D1 with mountainous 10:3 (super scumsided setup). Scum's reads are random. My mode algorithm is slightly scumsided, so the program is about 1% off. Anyway.

If town is 37.5% accurate, scum is lynched D1 40% of the time. That's VERY good.

If town is random (25% accurate), scum is lynched only 21% of the time. Why? Scum won't lynch themselves, so they're likely to be wrong!

The point is, while 37.5% accurate isn't the best thing ever, if we all followed our 37.5% accurate leads, town would easily win most games.

If your own reads are around random, that's okay. Mafia is hard, scum is good on this site, and the nature of forum mafia lends itself to bad reads. But you need to recognise when they are essentially random. What can you do instead?

Play To Your Strengths

When I started out, my scumhunting was godawful (like all 3 scumreads are town). Over time, I realised that while I'm not amazing at catching scum, what I can do is townhunt players with a >90% accuracy rate. I'm a towntell reliant player; in a 13p game there are almost always 3 or more people that bleed town.

What if you can lock 3 people as town and be confident about it?

If you can get to a 92% accuracy rate for your strong townreads, there is a 78% chance you're right about all of them. And if you can make sure town never lynches any of them?

The open setup Friends and Enemies with 3 masons vs 3 mafia is roughly balanced. So if you can make damn sure town doesn't lynch your 3 townreads, you've turned a 10:3 mountainous game into a balanced setup, 78% of the time. That's the power of PoE.

Disclaimer: Of course, your townreads will probably also get shot, so it's not quite like masons, but you get the point.

There are a lot of strengths you can play to:

• Are you better at townhunting? Lynch anyone outside your 3 townreads and you'll lynch scum 33% of the time in minis.
• Do you recognise a player in game you know you can read? Sort just those players. Personal experience with players is invaluable; learn how to meta well and how not to meta badly.
• Does a player look easily readable? Read through their ISOs in previous games, and try to predict their alignments. If they're always town when you townread them, you can townhunt them. If they're always scum when you scumread them, you can scumhunt then.
• Got charisma but bad reads? Convince the rest of the playerlist to play to THEIR strengths. Sort people by using everyone else.

Townhunting and confident meta-reads do a crazy amount of work for you. If EVERYONE could accurately sort 3 or more players with a high confidence, scum would almost always get lynched D1 or D2 in every game.

Maybe you've done all you can, but you still can't solve enough.

Know Your Weaknesses

I can name dozens of players that I can't D1 read for shit. When you start out the game, almost every player will be near unreadable. But consider this scenario:

• You're in a game, and Almost50 claims "When I'm townreading northsidegal, she is town 95% of the time. She's town this game". He has good reasons. You can't read Almost50 or northsidegal.

This is a team game. Trust your fellow players' strengths. The most probable scenario is that Almost50, if he's town, is correct. This means either: A. They are scum together. B. Almost50 is scum who knows NSG is town. C. Almost50 is town who is 95% likely to be right on NSG.

Can they be scum together? Remember, it's hard for scum to confidently defend their buddies; they know all their reasoning is wrong! Ask around, see if others can rule out A50/NSG.

...so you found A50/NSG isn't the scumteam? Great. You should be townreading her with 95% confidence.

Sheeping players who you trust when they're town will do a lot to boost your accuracy. You just have to find out whether the player is lying scum (scum don't always lie, remember).

And when a player flips town, now it's a game of "how likely is it this dead townie was right?"

Here's another scenario:

• You're in a game, and Mathdino claims "I can read JaydragonKing with 90% accuracy, and he's scum here". But you found a towntell that makes JaydragonKing town with 90% confidence as well!

If and ONLY IF you have good self-appraisal skills, you might be able to say "there's only a 10% chance I'm wrong here". If Mathdino is town, there's a 18% chance you'll ever disagree on a 90% confident read. Not. Good.

Burden of Proficiency is a valid tell. If you KNOW you can read Jay, and you KNOW Mathdino should be able to read Jay, it's HIGHLY likely in a Bayesian way that Mathdino is lying scum. The correct move here is to lynch Mathdino.

Disclaimer: Remember to back up other players' skill! If I claim confidence in a read on someone that I am bad at reading, this shouldn't tell you much at all other than "Mathdino is overconfident". See if their reasons are any good.

Regardless, a great way to get solid reads on extremely good players is to find out how likely it is that they could be wrong. I've lynched otherwise incredible scum players just by pinpointing where they should know better.

Always Room For Improvement, For EVERYONE

Self-appraisal is an ongoing process. There are still tons of players I still can't read, and the players that I CAN read will evolve their scumgames to get townread.

After every game, figure out why you were wrong and why others were right. That is how to get better as town.

My Self-Appraisal

Micro 764: Marked For Death 2: I shouldn't have lynched TheGoldenParadox when my scumread (chesskid) replaced out and started pushing him. And when Paradox was mislynched, I should've trusted that slot to be right on Paradox as town, and realised they were lying scum.

Open 704: Switch: I shouldn't have assumed Elmo was town just because Bins tunneled her. Bins is a known distancer/busser, which others figured out. I also shouldn't have assumed Raya36 was scum by associations when she was really just playing her towngame.

Open 707: JK9++: I could've caught scum-Creature sooner, and I should've stuck to my RedFlavor scumread going into D2 even after he hammered his buddy. I should've realised that I was scumreading Assemblerotws and JaydragonKing for playstyle.

Tells

Maybe you've used a tell for years and realised it went horribly wrong in your last game. The site evolves, tells change. We've had an influx of good scum players in the past year or two. Learn new scumhunting tactics. Test your old ones. If a tell works with an accuracy equal to random chance, it's not a good tell.

Tells are about what ACTUALLY works, not what SHOULD work. Armchair scumhunting won't catch scum. Learn from every game what town ACTUALLY does, and what scum ACTUALLY does.

Conclusion

Honestly, this kind of Bayesian reasoning can apply to anything. Identify your strengths and how confident you can be in them. Identify your weaknesses and when you shouldn't be confident in something. Identify the strengths and weaknesses of others, compare, and teamwork becomes easy.

And always, always, improve, staying one step ahead of the rest. That is how to become a Great Player.