Why Players think the AI cheats


#123

Basically my position is:

  • We have Sirrian confirming in this thread there is some luck manipulation.
  • The only intended luck manipulation should decrease CPU luck.
  • Sirrian insists several devs have scoured that code to see if it inadvertently fails because they, too, have really bad days.

That doesn’t mean I trust it, but it’s a developer confirmation that there is some luck manipulation and it’s been heavily inspected. I don’t want to get into why I don’t always trust dev scrutiny, it feels mean.

What I want to get into is we do have to acknowledge this game encourages us to create bad situations. When you cast a converter like Ishbaala, the board goes from 7 colors to 5. If the board’s already clogged enough you can theoretically reduce it down to 3. Statistically speaking, that’s a very cascade-rich environment. You might get some moves out of it, but it’s also likely you can lose momentum and hand a good board to your opponent after gaining some advantage.

What happened in OP is a little different, but worth considering. Skyfall is risk.

Let’s consider around the final move, near the 0:22 mark. AWR has a few options on his plate and I think each has a strategic observation. Maybe I missed one but whatever:

  1. Cast TPK.
  2. There’s a red 3-match higher up and blue 3-match next to it.
  3. A right-column purple 3-match.
  4. A lower 3-match at H4-G4 that would cascade into 2 yellow 3-matches.
  5. Retreat.

We also have to consider the board state. Both players can win with a single skull match. None are very obvious aright now, and no clear way to engineer one is present. Both players are ready to cast. Marilith’s cast is going to destroy 21 random gems if it happens. (The HP totals on the board make the other ability effects irrelevant for this situation.)

Destroying 21 random gems is the worst possible outcome. That is 32% of the board and means it’s exceptionally hard to predict the final outcome. But it’s also notable that the skull configuration on-board can fall into a configuration that allows a skull match several ways:

  • Columns 2, 3, and 5 could horizontally match. It requires a lot of gems to explode, but we know that is possible.
  • If column 8 falls 3 rows relative to column 7, a match is set up.
  • We expect 20+ gems to fall from skyfall, so there is high risk of lucky cascades that we cannot analyze at this moment.

If these three situations happen, AWR loses. So it’s clear that it is very dangerous to let Marilith cast. Ideally, AWR wants to prevent her turn. But he can’t.

I say that makes “cast TPK” his best move in this scenario. It has a 20% chance to transform Marilith, eliminating the risk of her gem explosion. If TPK fails to transform, AWR is no better off than he was before. The alternatives are:

  • (2)'s moves represent allowing a Marilith cast with no improved chance of winning.
  • (3) is suicide. That purple match leaves the skull exactly where Marilith needs it to win.
  • (4) has unknown probability because it causes a lot of churn. It makes the skull matches along 2, 3, and 5 a little more likely, but it’s also possible after destroying 3 gems in c4 and 1 gem across c{2, 3, 5, 6} the board might change enough to either cause new matches or thwart Marilith’s impending doom. I don’t believe in this move. More churn usually means more risk.
  • (5) doesn’t make sense because there is some expectation that a lucky win can happen and it is clear the game should end soon.

So that’s it, AWR’s smartest moves were (1) or (4). But we know (1) has only a 20% probability of removing an immediate threat, and (4) is entirely dependent on luck. Even if (1) succeeds, AWR has to face a new troop from a very weakened position: he could still lose.

TL;DR: AWR had no guarantee of winning this game, and was in a position that was very likely to lose.

He made one of (2)'s moves. That introduced a new skull at A3, which means there are now many ways for c{2, 3, 5} to create a match. Then Marilith destroyed her gems. The columns drop by { 3, 3, 3, 0, 4, 2, 3, 3 }, which seems a fair way to distribute 21 gems across 8 columns. What happens after 32% of the board is shuffled?

  • The skull in c5 is destroyed, so it is no longer a risk.
  • The skulls in c7 are also molested, but there is still a risk of a c7/c8 skull match with enough matches.
  • A 5-gem L match is set up at C6. Marilith gets a free turn and 3 columns with skulls change.
    • Note the problem here: C7 and D8 are in a position where if a skull falls into C6 or D6 a match can be made.
  • A 3-gem blue match happens at B6. This opens C6 to skyfall.
  • Skulls fall into C6 and B6.
  • Marilith gets her skull match, but not without slow-rolling a 4-match via F3-F2. I hate when the CPU does this.

So OK, what I see here is the problem is “Marilith exploded 30% of the board”. Once that happens, it’s as likely for a skull match to happen as any other cascade. There were hundreds of ways for AWR to lose this match. The only clear way to not lose was to prevent Marilith’s cast. He only had a 20% chance to do so.

So I think he should’ve cast TPK, but even if he did he’s looking at an 80% chance of loss. His next best move would’ve been that green match cascading to yellows, but since Marilith’s outcome is very random there’s no telling if that was better than 20%.

In short: this is “bad beats”. AWR had to make a decision, but didn’t have enough information to judge the outcome. It’d be worse bad beats if he hadn’t made one of the worst moves, but it took me 3 minutes to make my analysis and I doubt anyone really analyzes the board that closely outside of specific GW matchups.

This was a bad situation. No player likes losing when the entire circumstance is “I only win if I get lucky.”


#124

Had the troop been transformed. I would of been dealing with a level 30ish troop with full life and armor. I only needed one skull match to win the match.
There was literally nothing I could do but eventually lose. Period.


#125

I think it’s fair to say you were standing on top of a land mine and had a gun pointed at you. So “cast TPK” could defuse whichever of those you feel like, but would’ve left you dealing with the other.

I’m not saying it as “ha ha, AWR made the wrong move”. I’d have made the green move, personally, and not considered TPK the best option. Truth be told, you were looking at probably an 80% chance of loss best-case-scenario. Given how skyfall works, the green match was just as likely to end poorly. You do not want to be in a scenario like this with the computer primed to destroy 30% of the board.

It’s OK to make a move that wasn’t 100% optimal when the stakes are so bad. But it’s also important to realize there were other decisions. The result of the green match is super unpredictable, in hindsight I value it more than the TPK cast personally because “I don’t know wtf this does but I can win” is better than “20% it deals with one of my two lethal problems.”


#126

Replacing a troop that has 40 life with a troop that has 200+ overall HP is not the best option…


#127

Yes, I was editing the post when you made this reply but I do want to appendix: “Your reply to that paragraph is why I tried to edit it, I wasn’t properly aware of the full context of the result of the TPK cast and evaluated it poorly.”

I like the green move. At least it’s a prayer.


#128

Between last night and today. I managed to get a win streak of 85 wins. Then…
Screenshot_20190131-173430
Infernal King respawned, Thief spawned after, thief again, in the same slot, then thief for Mercy. Then I gave the AI a 5 yellow match since it wouldn’t get any Mana from it… Well Cascades made it so Doom Bow filled anyway, casted, summon a Sister of Shadows into the first spot… And ultimately I lost. Oh and Lust transformed one of my Troops from one cast.
In my head all I could think was “it’s just RNG, it’s just RNG”. 🤷


#129

Actually, I do wonder if TPK is one of those cases where it can multihit a single troop if it is the only one left like the ship cannon or knight coronet.


#130

You can’t. It only hits once per troop on the board. And it will hit, then transform if able to.


#131

Wording is always tricky in spell descriptions such as skeleton key and kc having the same damage targets, but function entirely differently. The 2 key words that get me are “separate” and “enemy”. In cases like infernus, druid, and ship cannon, with one target left, the spell will hit multiple times. The separate could easily refer to (and probably is) the transform effect on multiple enemies so if one transforms, the other one or 2 don’t.

Hey @Cyrup, could you confirm if The possessed king’s transform portion of his spell hit the same target again such as infernus or ship cannon if there are no additional new targets?


#132

From what I’ve witnessed. TPK can transform 3 troops at one time.
So it targets 3 and tries to transform those 3 only.
If less troops are on defense. Then it will only have 2 or 1 Troop to try and transform.
As far as that chance stacking. You’d need to ask a dev yeah.


#133

I’m at the point where if I see Lust, I check to see if it’s a Thief hero and just nope away.

I was thinking about the numbers for hitting Stealthy slot 3 earlier and figured out about 2 bandits is what to “expect” here, with 3 being more likely than I want.

I also figure my party can take about 2 charm hits before I’m in serious danger of losing a troop, and 3 before I’m pretty sure I’ve lost one.

So it’s just not worth it. I start with about a 50% chance I’m on the back foot. That’s not my kind of game.


#134

@KYLENATOR001 https://youtu.be/fpaQpyU_QiM I always thought it did stack against a single remaining troop. Off-topic, but I’d defs like to know the answer.


Although not specifically mentioned by @awryan, I actually always thought this thread surmounted to a request for RNG to be more controlled/evenly distributed, as per:

This could be for both the player and the AI, such that it’s incredibly unlikely (impossible, if you wanted) for a 4- (or even a 3-) match to drop, unless there’s a storm running.

When @awryan referred to the ‘Luck Factor’, I generally interpreted it as (or focused on) the possibility for these unfortunate scenarios to occur, rather than something that deliberately affects the coding and difficulty at particular points in the game.

I’d be super curious to try out the game with a more controlled/evenly-distributed RNG.

@Lyya, you mentioned this:

Has there been much of a discussion around the possibility of increasing the combo-breaker for both sides before, or a post with the developer’s thoughts on it? Not asking you to be my personal forum historian/librarian :stuck_out_tongue_closed_eyes: just if you remember it, somewhere, otherwise I’ll go have a peek (the threads on AI luck/RNG/cheating are pretty long and windy).


@Slypenslyde, I was the same, but I actually now run gleefully towards Thief/Lust defences with a Khorvash, Valkyrie, Doomed Blade (Sunspear), Tai Pan team (I forget the banner, probably Mist of Scales). Order can be changed to facilitate better looping (and thus more stuns) (i.e. putting Khorvash down lower). I’ve been told this would take too long, or not be consistent enough, by some? But I’ve generally found it pretty easy and quick. I can usually get away with only stunning one Lust, too :stuck_out_tongue: if I’m feeling lazy.

Also, I know a ‘this team works’ isn’t generally a valid argument against broken teams, just thought I’d share in case it made life more fun for you :slightly_smiling_face:.


#135

No, I don’t think the devs have ever chimed in on such a discussion. I’m not opposed to the idea of this, though it’s hard to imagine that the devs would simultaneously buff enemy stats to 500ish and also lower their chance to hit with a percentage-based spell.

However, adding logic to alter percentages based on “luck” or “am I the CPU” opens up the permanent possibility that the logic is flawed. Any time anything goes south from then on, a player can point to this “tampering” (even if it’s in their favor) and go, “see? They made a bug! It’s worse than before!” And the ammunition of “it’s just a random.Next() call” goes right out the window.

Likely this isn’t worth the headache, but I don’t dev for IP2 and can only editorialize.


#136

I was imagining it to be indiscriminate of player/CPU, although I can definitely agree that it would open up

E.g. I know you’ve talked about how the starting board checks for ‘Are there any alignments of 3 or more of the same gem’ before proceeding, etc. - the suggestion would be to do something similar for any skydrop (unless there’s a storm).


#137

This is what I tend to ask for with a lot of words.

Even when I get a really long series of cascades, I roll my eyes. It’s miserable when the CPU gets it. The game would be better if you just couldn’t take 5 turns in a row. Ever. I’ve proposed everything from “more combo breaker” to “what if it capped at 3 turns” to “what if there were no free turns”.

And RE: whether the skyfall could be checked, I’ve played Match 3 games that did. It intentionally dropped its icons in a sort of checkerboard pattern that wasn’t quite predictable, but really good players had a good feel for what the top row should look like to maximize the chance of cascades. Combos longer than 4 or 5 were exceptionally rare in that game outside of specific puzzle setups.


#138

I meant to mention that it wouldn’t exclude skyfalls interacting with gems already on the board, but from what you said,

And you found this to be a good thing, correct?


Me too :stuck_out_tongue_closed_eyes:.


#139

Anyone who thinks this game is 100% RNG based is deluded. There is definitely other factors at play that determine how good the AI will perform AND how lucky it will be.

When was the last time you lost 2 pvp battles in a row?


#140

Last night. Then I won 85 in a row. 🤷


#141

I always love it when the CPU matches three gems and gets a free turn. And no extra turn flash on screen


#142

The increased luck factor for the losing team in the end of a match is the most annoying feature in this regard. In some modes it can really destroy you as has already been mentioned. When AI has 3-5x attack value compared to you, even a slightest favoring factor for them is murder.

Treasure hunt is the only game mode where I feel comfortable and get the feeling I’m playing a logical game.

I remember playing the total random shuffler Elspeth-Sacrifice-Crimson Bat which based 100% on luck to win ~66% of the matches against any team. Winning and losing streaks involved when playing that were interesting to follow. There I understood what’s said here about PvP win-streak killing mechanics. I should have recorded that, since there is no equal team to test today on PvP. So many new mechanics and the gem creation for sacrifice was changed too. Though those are a positive upgrade.