# Outcome manipulation for autobattler games

### How to manipulate randomness in games, and how it may already be happening.

TL;DR: games which can accurately simulate results of parts of the game may already not be honest about RNG use, possibly to your benefit, possibly against it.

#### Poker example:

There is no more player input and the outcome depends on random variables. A pair of Tens wins ~54% of the time vs AK. In normal poker, the EV of each is proportional but with high variance. In EV Poker, actual returns match expected returns (since you reward based on simulated actual equity).

But in this mode, you reward based on the median of the simulations. So TT vs AK would be simulated as:

• +1
• +1
• +1 ... (54 of these)
• 0 (<1 of these)
• -1
• -1
• -1 ... (46 of these)

Then you would take the median (+1) and reward all the money to TT 100% of the time. EV Poker rewards you based on actual equity, but this is something different. Does it even still reward quick play? The meta for going all-in would definitely change.

#### Storybook Brawl example

This is a game where you build an army and then the game has them fight vs another army. Skill in building the army matters, but the order that your guys attack, and who they target, is randomly chosen, and it really matters. Watching the outcomes of fights, it's infuriating how random it all seems. Better players still win more often, but there is a huge amount of variance. Or is there?

This game also has the property that there is a clear results, and you can determine the win margin numerically. The "score" for each simulation doesn't capture 100% of the meaningful variation - i.e. you can "win by X" in good or bad ways; but the score captures most of the meaningful variation.

What if this game is already doing this - simulating N battles, and then picking the one to show the players based on some algo? They may not just be choosing the median - there are all kinds of interesting choice systems you could use.

### How would games choose which random outcome to give the player?

• median - likely boring
• most amazing - in either direction
• based on user stage in the funnel
• based on a personality profile of the user

All of these are doable with and without preservation of statistical qualities, fairness, and/or detectability.

#### Giving amazing experiences

The game can quantify how "amazing" or "cool-looking" an outcome is! i.e. if it sims a fight which you lose 99% of the time, wouldn't it be amazing to display the 1% battle as the "actual outcome"? If you're watching you'll have a sense of dread before the battle - but as you watch your army continuously get lucky, you'll be thrilled! If I were an evil PM, I'd intentionally set up new players to experience something like this in their first play session.

What I'd actually do is create a list of amazing and cool mechanics about the game, then just keep picking sims which expose them to the user. Most users wouldn't pick up on it, and would have consistently amazing experiences.

In poker this would result in cracking aces more often than probability would result in - or you could go the other way and choose the median and punish speculative play.

### This is the randomness axis

It's typically thought that more randomness is friendly to casual players, and hardcore players like reduced randomness.

### This may already be happening everywhere!

• We know that PvM games already manipulate randomness / enemy levels / rewards to keep players hooked
• But for PvP games in an environment, if you can simulate player behavior, you can also just directly measure the simulated outcomes of battles as you vary the input params you control (the rng seed, rewards, environment etc.) and give a result that you calculate will optimize your company's value function.
• i.e. not losing a whale who's showing signs of flaking out.
• or burning the experience of someone who's hard core addicted already

### Compare this to a game like Slay the Spire

This game has additional properties that make it pure:

• single player
• single random seed determines all outcomes
• choices presented to the user have little / no feedback based on user success ** i.e. if you are about to die the game won't go easier. Your hit points or power level rarely is used as input for game decisions.

#### But a Slay the Spire version of this may still be interesting.

• Imagine you had a reasonable AI player for STS
• Before every fight, the engine might simulate 1000 battles, then pick the seed which had the median outcome.
• this seed mainly controls card shuffle order, and also enemy behavior (which has scoped randomness)

It's interesting to think about how this would effect STS. Picking the seed from the extremes would definitely make you die more. So picking the median would tend to completely eliminate any enemy behavior which has <50% probability. So I can't claim that this is "best-play preserving" which is unfortunate.

Overall the goal is to find a way to truly and quickly evaluate skill in games which have a lot of randomness in them.