Gems of Wars RNG is currently broken

Tons of pet gnomes lately for me, its good when rng is good in return hehe

1 Like

If its manipulated for new players, I am just fitting it into my narrative assuming it is manipulated for veteran players as well.

My assumption is such a stretch, I know.

2 Likes

lol, nothing wrong with that assumption. I don’t fault the idea there could be something going wrong. Just the fact that the devs responded directly to the question and if that still wasn’t satisfyingly, you’d have to now prove your assumption. If there was even a way to reliably do that, I’d volunteer for a data analysis.

Now we could point to the Portal Shard drop rate as a situation where the developers said one thing and the code showed another. This ended up being proven by the players to be wrong after deep statistical analysis. The burden of proof was on the player since the devs claimed otherwise. It won that case. If you want to do it again, you’d have to prove it again. That’s all I’m saying.

4 Likes

Sorry, I don’t have access to legendary tasks. We are a casual guild that does complete regular tasks and a few low tier epic tasks. However, if somebody can provide a dataset, I can try to analyze the data. Keep in mind that the outcome of legendary tasks is determined by the server and not by actual in-game pRNG.

That is an example of heavily conditional output, which cannot be tested reliably. There are too many things at play here. Ideally, this type of testing should be limited to challenges to lower variability and associated conditional randomness.

It is somewhat conditional, so the only thing that can be tested is randomness and independence at the same time. This is fairly easy and can be tested by anybody. The only thing that is required is record full sequence of the events (dodge or no dodge) as 1 or 0. It has to be full sequence without missing numbers exactly as it happened. The dataset also needs dodge percentage of each specific card. ideally, it should be done against a consistent team (to simplify the calculations) containing a dodge card on top by using a skull-generating team. For 25% dodge, a sequence of approximately 150-200 events is needed to estimate randomness and independence (both together) with 95-99% probability.

That is also determined be the server. Also, you are talking about something that happened in the past, so unless somebody already has a dataset, this is simply impossible.

Keep in mind that anything that happens on the server is a complete black box because it is heavily conditional and the confounding factors are completely unknown.

1 Like

Sounds like for the most part I was originally correct
… Because it’s RNG based… It’s “impossible” to prove/test then. :man_shrugging:

That, and also, anything that’s happening on server happens for thousands of players. If there’s an RNG generator involed on servers you can get the impression that some things roll same way for you and/or your clan, while other people dont.

Results of such experiments could also be skewed in some way. From all that i’ve noticed in other topics RNG is giving results for various things → gem colors that are being dropped, tratis triggering, number of damage done by some troop spells, troop that’s gets affected by "apply X status effect to random ally/enemy) - all these things happen in between the sequence of dodges… or IK trait poping up… it might look streaky, but again - it might not be… given a simple example:

Skulls hit IK, IK dies, RNG rolls for it’s trait → respawns in 1st slot(with a 25% chance). RNG also rolls for each gem that’s dropped to be replaced, by the game for each gem removed while applying skull damage. To simplify thing, let’s say it were 7 gems (one for match4, and other for match3). So there were 7 pseudorandom numbers taken from generator. Next move → you match another 3 skulls, which than line up for match 4 of another color, and than another extra match3 of skulls goes off. RNG, needs to roll for all those 10 gems that were removed. IK, getts hit and dies again. And respawns again.

it’s looks streaky, because 2x 25% changes resulted in 2x succes rate, but between those 2 “rolls” there were 7 (or more realistic → 10, as new gems to replace the board are rolled earlier than on death effects) RNG rolls. If you want to analyse if RNG works fine, you’d have to get all those 12 numbers that were used to determine, if it gives random numbers or not… and that would still require a much bigger poll of data

Final note: Most RNG return number between (0,1). 25% chance for XX is usually scripted as:
if rng_number < 0.25 then success

Last week had a battle vs thief, where i killed 11 enemies. 7 bandits spawned, and at least 3 of them spawned in a row. Had a bonestorm running, and it just looked like:

me: matched skulls
game: you killed troop#1
game: bandit spawns
game: new skulls matched
game: you killed troop#1
game: bandit spawns
game: new skulls matched
game: you killed troop#1
game: your turn
me: matches skulls
game: you killed troop#1
game: new skulls matched
game: you killed troop#2
game: bandit spawned

So there were lots of skulls that spawned, and lots of bandits that spawned.

I had few rare things happen to me: TPK converting same troop 3 times, with a single cast (now this requires 3x 20% in a row).
Raising Shadows procing twice in a single cast → two 7% chances in a row*

  • other on death effects (like Lust charm) might be calculated between checks for Raising Shadows in this situation

No, the test as suggested will produce reliable estimate of chance that the output is (random and independent) or is not (random and independent). It will not tell if the output is (nonrandom and independent) or (random and not independent) as these two cases are equally considered not (random and independent). The test is not based on “impression”, “bias”, “perception”, etc as there is nothing subjective about it, only accurate statistical estimation.

You can continue to think as you wish or instead try to educate yourself and learn something new in the process.

1 Like

You admitted you can’t test much of the stuff I assumed couldn’t be tested in the first place. But you kept insisting and pushing. So I gave in and went into pointless detail about some of the streakiness. And now for wasting my time. You add insult to the bonus. Thanks. :+1:

You are getting very touchy out of a sudden. I was trying to wake up some exploratory positive spirit of yours without knowing that it apparently does not exist. :wink:

Anyhow, I was thinking about conditional probabilities and how convoluted code strings might actually create a mess out of random, and a streaky mess at that. What I wanted to say that it is quite possible that heavily conditional and poorly optimized code with respect to handling random values may indeed produce excessive streakiness and skewed output. While it looks random to a developer, as I said already, it is not actually even remotely random in some specifically convoluted cases.

Developer thinks looking at the code that everything is nice and random while the code is actually sufficiently complex and convoluted to link conditional values that seem completely irrelevant, like what you mentioned about mimic devouring in the arena. Theoretically, and probably practically as well, this is not unreasonable completely although it does sound absurd. The more conditions are entangled in a certain output, the harder it gets to actually determine what is going on for the developer and output tester as well (player). As the game became more and more complex over the years, apparently, pieces of old code got included in the conditional resolutions with new code to create current mess, which GoW has gradually become. Most likely, it will stay that way unless somebody decides to clean up the whole random-related code to remove all major cases of conditional output. IMHO, seems very unlikely. So, while there is no conspiracy, the game is becoming more and more streaky as the complexity grows, which is apparently a natural process.

Some scientists, for example, analyzed streakiness of modeling of golfer or football player behaviours in a simulation similar to model used in computer games. What they found is that increasing complexity and adding multiple factors actually increases streakiness to the absurd values. It does achieve more realistic simulation though. So, seems like anything sufficiently complex and based on random will be streaky unless the model includes special measures to prevent that. These measures include so-called “adaptive” sampling to generate output, which essentially means separate unconditional coding (modeling) of every single random number as a separate string without combining nested or linked conditions.

The whole theory of these things is gradually becoming some relatively popular subject in mathematics. If somebody is interested, I can provide the references. Sorry, I understand it is a bit tldr and too convoluted for the forum audience.

So, tldr: Complexity of the game and suboptimal coding of random output may indeed cause absurd streaks. And no, it cannot be fixed without specific optimization of the code.

2 Likes

Yesterday, while fighting some PvP matches, I stumbled upon a team with 3 TDSs… I ended up killing all 3 of them in the same turn with my QB. Guess what? I’ve got myself 3 empowered baby dragons; that’s a 12.5% chance, fine. Just two matches later, I happened to fight the same team, again I killed all 3 TDSs in the same turn. Guess what? None of them transformed this time, that’s another 12.5%… Ok, fine, I’ll bite.

I believe in random number generators, but I also believe in bugs that may cause the same generated value to be applied multiple times (an event which is even more likely to happen when you start purposely messing with the RNG to favor players at easier difficulties).

2 Likes

LOL We ALL know that the RNG is to purposely make us lose sometimes.

If you’ve ever gotten 2 mythics on a 50 chest guild seal pull (which has happened in our guild), you just saw something that has a .11% probability of happening, or once every 900 times.

Getting 1 mythic in 200 gem key pulls has 16.4% chance of happening.

20% dodge happening 4 times in a row .16%.

Missing an irongut devour with 67% chance 4 times in a row 1.2%.

I have experienced all of these things. Just adding “data” to this pool.

1 Like

@Nemesis_Online the RNG being skewed to players at lower levels happens with board drops, not with percentage chances on traits or troops. It stops the AI from getting a lot of cascades.

Also, at no point does the RNG purposefully make a player lose. There is no purpose to this, and wouldn’t benefit us or our players. It doesn’t make any logical sense.

1 Like

I’m a new player, about 3 weeks in, but I’ve been playing a lot. I’ve posted on another thread about my observations about the Agile trait (20% to evade Skull attacks). I have seen many times where a troop with Agile would evade 3 times in a row. The replies have been that it’s RNG and I only notice streakiness when it’s not in my favor. For me, 20% means, on average, Agile should proc 1 out of 5 times. But that’s not what I have seen. Before a fight, I look over the enemy’s cards to see what abilities they have. Now when I see the Agile trait on an enemy, I laugh because I know that it will evade my first Skull attack and more probably, the next one or two, also.

Now, I can’t say if the RNG is working in my favor because I don’t yet have any troops/hero with a percentage to do something.

I also have seen the AI get 4 to 5 4-gem matches much more often than I have (normal play, not when AI enemy casts a spell to change the board colors). I consider it lucky if I find 1 or 2 on my turn. So yeah, I think there is something amiss with the numbers and percentages in the game.

1 Like

Yes and no.
You have a mechanic called combo breaker that 2-3 years ago when first implemented was fair to both sides. But now that the player is going against the AI that has far superior stats (level 20 versus level 100+) that combo breaker (aka AI controlled “rng”) can quickly decide the match in favor of the AI. If one skull combo is the difference between winning and losing then unintentionally the AI makes the player lose… But it 100% intentionally makes them lose their turn by interfering what would of been a 4 or 5 skull match. Trust me I know, Fang Moor has a skull spawner that can be a great example of it. And when the combo breaker doesn’t know it’s triggering too many times because the RNG streakiness is either not being capped or is broken. It can lead to great frustrations.

And I’m sorry, I don’t want this to be “this happens because the devs don’t play the game enough”. But back when Sirrian played often, I never saw this type of behavior. Probably because he saw it before I could and kept me and others from ever seeing it. So if Sirrian isn’t available to guard us from broken RNG… Who is?

2 Likes

Btw…
Careful making this argument when your company is telling you your current official response to Pet Gnomes and the arena is to ghost the matter for now.

4 Likes

For the sake of the argument, most game modes have an energy system. You need to buy sigils to play battles, winning enough of them unlocks rewards. There are already mechanisms in place to coax players into buying extra shop tiers, like the Thursday class event where you almost always run out of sigils two battles short of the final reward. It stands to reason you would benefit from nudging paid-for battles towards a loss.

However, I’d be very surprised if there is anything fishy going on, reported streaks sound well within expected values. It’s possible to analyze this in detail, without knowing what is going on behind the scenes, it just doesn’t seem to be worth the huge effort it would take to reach a high enough confidence rating. Besides, there are far easier ways to influence battle results, like putting Lust in the last slot of teams that need to be killed with skull damage.

2 Likes

Its RNG; we all know that; its just having the willpower and fortitude to move on and ignore the repeated in game situations where you are tempted to add sigils which are actually real money, never ever forget that; purchasing beyond gems earnt in game is never ever worth it, its RNG.

1 Like

This is a common trap to fall into. Yes 20% is 1 in five. But when you are talking about probability, like a rng, it’s not a static thing. Think of it like flipping a coin. When you do that, you would say you have a 50% chance on each flip of it being heads, and 50% tails. So 1 in two. But that chance is per flip. If you flipped it and got heads, it doesn’t mean the next flip is going to be tails. Getting heads 5 times in a row, doesn’t mean the coin is broken. I hope I’m making sense. It’s late!

1 Like