Don't trust the "likely outcome" combat predictor

[Callonia]

CaptainPatch, there was a moment in warhammer 40,000.

I got this angry really evil chao squad launching 300+ attacks upon my 10man guardsmen squad with carapace armor.

150+ hits managed to land, the guardsmen dodged the rest.

My opponent was like, do you surrender now? I was like NOPE.

And rolled 150+ dices. Saving roll is 4+ per hit because carapace armor is 4+.

Only three guardsmen died because I failed three rolls.

Guardsmen squad held.

Now things is looking very awkward for my friend who wanted to play chaos for a battle. For his entire army is now within my charge range because he misjudged the charge distance which failed as the result. That one squad was only thing on chaos's charge that reached me.

After how well that squad held.

I yelled, For the emperor and did a massive counter charge that ended in a complete rout as my puny guardsmen outfought the chaos in melee combat that happened in following turn.

Then my friend went, never again. I'm going back to space marines. Forget chaos.

Such is nature of RNG. You never shall know the results until the dice is thrown.

 

[CaptainPatch]

CaptainPatch, there was a moment in warhammer 40,000.
...

Such is nature of RNG. You never shall know the results until the dice is thrown.

Recalling Lance The Lucky:

We were having a Bar-Lev tournament (a John Hill game, the guy that created Squad Leader for those that remember back that far). Most of those games followed a pattern of early Arab success, but if he hasn't won by Turn 6-8, the Israelis WILL recover and roll back over the Arabs all the way to Cairo and Damascus. So I'm playing Lance being the Israeli player. The way the game is laid out, there are two maps: the Suez Canal area in the West, and the Golan Heights area in the East.

If at any time the Arab player manages to clear ALL of the Israeli pieces from either map, the Arabs win an Instant Victory.

So my opening was damn strong and by Turn 4 I've pushed the Israelis back into tiny pockets on both maps. I'm closing in for the kill on the last Israeli unit on the West map. I've brought a whole lot of hurt on that one unit. Roll anything but a 6 and it's an Arab Instant Victory. The die goes tumbling and it's a .... 6. Oh, well. Things were going just as well over in the East map. I've cornered the last Israeli unit there, and again, it boils down to roll anything but a 6 and it's an Arab Instant Victory. The die rolls across the board and up comes... a 6. Lance's turn and the Israeli forces come pouring onto the maps. They launch their counterattacks and desperately need to create some elbow room. So Lance is making a LOT of desperation attacks at 1-to-2 and 1-to-3 odds where he needed to roll 1s and nothing but 1s to kill Arab units. He starts rolling the die, and calling out the results: "One. One. One. Four; where did that come from? One. One. One." Arab units are flying off the board.

Containing my frustration at having come sooooo close, I start Turn 5. That early in the game,the Arab reinforcements were plentiful. I resume the offensive and AGAIN I've pushed through in the West and cornered the last Israeli unit there. AGAIN it boiled down to "roll anything BUT a 6 and win!" The die rolls across the table and came up... 6. [Loudly grind teeth.] Switch to Syria and charge forward with a vengeance, slaughtering Israeli units with glee. AGAIN, I cornered the last Israeli and AGAIN it's roll anything BUT a 6. So naturally, AGAIN is came up 6. Lance's turn and AGAIN I'm hearing, "One. One. One. One."

Turn 6. It's now or never. The stream of Arab reinforcements is drying up. On the sidelines of the other side of the board, I see small mounds of Israelis just waiting for their opportunity to swarm onto the board. My forces press forward and AGAIN I've cornered the last Israeli unit in the West. And AGAIN, it's roll anything BUT a 6,.... Oh, who am I kidding? Of course it came up a 6, AGAIN. And AGAIN in the East, roll anything BUT a !@#$% 6 and... It was inevitable, wasn't it? If looks could crumble an inanimate object, that die would be dust. Then here comes Lance and the chorus of "One. One. One. One...."

As I said earlier, if the Arab hasn't won by Turn 6, he WILL lose. At the conclusion of my inevitable defeat, Lance was trying to be a gracious winner by pointing out flaws in my strategy by starting to say, "You know, where you went wrong was...."

I nearly knocked out his front teeth before he said another word. Nearly. He saw the fire in my glare and knew enough to shut up.

In sympathy, the rest of club members got me a plaque that read "World's Worst Dice-roller."

 

[Callonia]

HAHAHA Yeah! Captainpatch, that's what I'm talking about Its very likely that same thing is occuring ingame on Civ5.

Back then when Hp system was only 10 hp. Bushido was insanely good. Winning all kind of crazy charges. Now, its pretty good in making sure your units is in fighting peak all the time at 100 hp.

 

[adsin15]

Sometimes the combat calc is wrong for totally understandable reasons. Take this example:

Lets say I have a Longswordsman fighting an enemy longswordsman. All units have no promotions by the way and are fighting on plains. The enemy longsword is down to 22 health. Now lets say I have a Knight that is NOT adjacent to my Longsword but is within striking distance of the enemy Longsword.

If I select the Knight and look at the combat calc for attacking the Longsword it will tell me the enemy Longsword will survive the attack with a sliver of health...2 points...bc Knights do 20 damage and absent any bonuses for either unit 22 health - 20 damage = 2 health.

HOWEVER, as you well know, melee units fighting side by side are granted a 10% flanking bonus. Since the Knight ends up next to MY Longswordmen when it actually begins attacking, it gains the 10% bonus giving it 22 attack and then allowing it to destroy the enemy longsword.

Just an example of how the Battle Calc is affected by random stuff.

 

[TheMeInTeam]

Sometimes the combat calc is wrong for totally understandable reasons. Take this example:

Lets say I have a Longswordsman fighting an enemy longswordsman. All units have no promotions by the way and are fighting on plains. The enemy longsword is down to 22 health. Now lets say I have a Knight that is NOT adjacent to my Longsword but is within striking distance of the enemy Longsword.

If I select the Knight and look at the combat calc for attacking the Longsword it will tell me the enemy Longsword will survive the attack with a sliver of health...2 points...bc Knights do 20 damage and absent any bonuses for either unit 22 health - 20 damage = 2 health.

HOWEVER, as you well know, melee units fighting side by side are granted a 10% flanking bonus. Since the Knight ends up next to MY Longswordmen when it actually begins attacking, it gains the 10% bonus giving it 22 attack and then allowing it to destroy the enemy longsword.

Just an example of how the Battle Calc is affected by random stuff.

That is not an example of random stuff. That's not even a random factor.

It's an example of poor ui. If the game resolves the combat accurately, it can show accurate ui too.

But it didn't, because it is poor UI.

 

[NukeAJS]

My biggest problem with this system is that even if it does report the correct chances of winning -- you still can't see by how many HPs you might win.

This is a big problem when it's reported that you have a "decisive victory" but you only really have a 51% chance of you completely eliminating the enemy to zero HPs. There's still a 49% chance that they survive until the next turn and that your attacking unit could be in an unfavorable position. This is especially true of units that gain HP for kills -- Aztecs and Ottomans come to mind.

I actually have no problem with the above situation. There's a random element to that and it's fine. I just think it'd be better if there were negative HP predictions. If I went to attack xyz unit and the prediction was -1 HP (for them) I can safely assume there's a good chance that they'll survive. On the other hand, if I attack someone and the average damage I deal brings that unit to -30 HPs, I can safely assume that they will die and I'll survive.

 

[beetle]

But it didn't, because it is poor UI.

I actually think that it is a feature that the combat predictor is not much more accurate. It is more fun the way it is!

My biggest problem with this system is that even if it does report the correct chances of winning -- you still can't see by how many HPs you might win.

My only complaint is that the combat predictor understates overkill. With overwhelming odds the expected result is 100 points. But often fair odds is 100 points. I wish the overwhelming odds said 150 points (or whatever) so that the more even odds were in better contrast.

And what is up with city damage? I can see my city attacked for 100+ points of damage and it will still be in the green. What numerical scale are cities using? I just judge by the city health bar, but I feel like I am missing something!

 

[TheMeInTeam]

I actually think that it is a feature that the combat predictor is not much more accurate. It is more fun the way it is!



My only complaint is that the combat predictor understates overkill. With overwhelming odds the expected result is 100 points. But often fair odds is 100 points. I wish the overwhelming odds said 150 points (or whatever) so that the more even odds were in better contrast.

And what is up with city damage? I can see my city attacked for 100+ points of damage and it will still be in the green. What numerical scale are cities using? I just judge by the city health bar, but I feel like I am missing something!

A damage chance with variance is acceptable. Misleading the player by displaying a point off the center of a bell curve is not.

If the latter is happening, it is fake difficulty and has no place in any strategy game. Even if you're playing with rng, respectable mechanics do not lie about their nature.

If someone will get a flanking bonus on attack, the UI should not hide that arbitrarily.

 

[Primacide]

I respectfully disagree. So much in Civ is discreet and predictable that I like not having 100% accuracy. Sometimes it breaks my way, sometimes it doesn't. Let's face it - combat is predictable and not super challenging in civ to begin with. A little inaccuracy in a damage estimator is fine. I am cool with using it as a rough gauge instead of an absolute prediction.

@Beetle: are you playing in strategic view? The damage indicator for cities is bugged in strategic view and always shows the city as being at lower health than it actually is. If this is the case, switch to the normal view to check the city health.

 

[Callonia]

I actually think that it is a feature that the combat predictor is not much more accurate. It is more fun the way it is!



My only complaint is that the combat predictor understates overkill. With overwhelming odds the expected result is 100 points. But often fair odds is 100 points. I wish the overwhelming odds said 150 points (or whatever) so that the more even odds were in better contrast.

And what is up with city damage? I can see my city attacked for 100+ points of damage and it will still be in the green. What numerical scale are cities using? I just judge by the city health bar, but I feel like I am missing something!

Its not instantaneous, after all the enemy units is done attacking the city, health bar will drop to the correct hp level. And numerical scale depends on what kind of upgrades you have built for the city. Walls, castle, arsenal, military base..

 

[beetle]

A little inaccuracy in a damage estimator is fine. I am cool with using it as a rough gauge instead of an absolute prediction.

Glad to see I am not the only one that feels this way.

are you playing in strategic view?

Nope.

Its not instantaneous, after all the enemy units is done attacking the city, health bar will drop to the correct hp level.

The health bar during and after combat seems to be correct.

And numerical scale depends on what kind of upgrades you have built for the city. Walls, castle, arsenal, military base..

The numbers that go by during the enemy turn make no sense to me, not unless the base health is 1000 or something.

 

[TheMeInTeam]

I respectfully disagree. So much in Civ is discreet and predictable that I like not having 100% accuracy. Sometimes it breaks my way, sometimes it doesn't. Let's face it - combat is predictable and not super challenging in civ to begin with. A little inaccuracy in a damage estimator is fine. I am cool with using it as a rough gauge instead of an absolute prediction.

@Beetle: are you playing in strategic view? The damage indicator for cities is bugged in strategic view and always shows the city as being at lower health than it actually is. If this is the case, switch to the normal view to check the city health.

Do not confuse uncertainty with active misrepresentation. You have succeeded only in arguing for the former, while I am arguing against the latter. If you wish to refute what you quoted, you must make a case for active misrepresentation of the games rules. Say it directly: why is it a good thing for the user interface to lie outright?

Glad to see I am not the only one that feels this way.

I extend this argument to you as well. I'm not talking about "we don't know the outcome for certain", that's non-sequitur to this thread and arguing for that is a strawman. I'm talking about "the game is telling you the odds are one thing but they're actually another". My statement all along has been *if* the outcomes hold over a larger sampling, the user interface is lying, which is fake difficulty (trial and error gameplay), nothing but a noob trap trick ploy if intentional (more than likely, if it is happening it is not intentional).

Not displaying the flanking bonus when the flanking bonus is actually in effect (and would consistently show up if moving units painstakingly one hex at a time) is a UI failure. To make a rational case otherwise, you *must* demonstrate that trial and error gameplay is a viable element of a strategy title. I believe that you can't do this, but if you presented evidence to the contrary, showed that it presents a meaningful increase in the number of meaningful decisions per turn on average (as opposed to just memorizing the UI lie), then I will change my mind on the matter.

Absent that, if you take the "can't trust the display without trial and error is a good thing" argument to its logical conclusion, it would be irrational to reject an assertion that we should delete UI projections in general; wrong information is worse than useless. Just learn everything by trial and error alone.

Alternatively, if you just misunderstood my argument that's fine too .

Edit: Base health for cities can go well over 100.

 

[Primacide]

A damage chance with variance is acceptable. Misleading the player by displaying a point off the center of a bell curve is not.

If the latter is happening, it is fake difficulty and has no place in any strategy game. Even if you're playing with rng, respectable mechanics do not lie about their nature.

If someone will get a flanking bonus on attack, the UI should not hide that arbitrarily.

Well, to retort, you are assuming that active misrepresentation (lying to the player) is actually happening. The game isn't telling you "you will totally win this battle" and then your unit dies without inflicting damage.

When I attack and the predictor says I'll do 30 points of damage and take 20, that's usually roughly the case. Is it slighted a little to one side of the curve? I suppose. I wouldn't call it outright deception though. I just know it low-balls a bit. In your example, you talk about a knight coming in from a couple tiles out and how the damage predictor doesn't indicate the flanking bonus it will receive when it gets there. I would point out that in many situations, the code can't completely predict what tile you are actually going to be on when your attack lands. If you move the knight to the tile with the flanking bonus first, and then hover over the enemy to bring up the damage predictor, it will then include the flanking bonus.

Sorry to nitpick, but this seems like kind of a small aspect of the game.

 

[TheMeInTeam]

Well, to retort, you are assuming that active misrepresentation (lying to the player) is actually happening. The game isn't telling you "you will totally win this battle" and then your unit dies without inflicting damage.

No, I'm not assuming that. I said I *would* assume that if OP's outcomes hold over a large number of trials and/or vary by difficulty without the UI indicating so.

However, simply not displaying 10% or more in combat modifiers when they will apply for certain is, without question, a UI lie and likely a bug. If the knight is going to do flanking damage, and the combat rolls apply it, but the predictor doesn't show it, that's not working properly.

When I attack and the predictor says I'll do 30 points of damage and take 20, that's usually roughly the case. Is it slighted a little to one side of the curve? I suppose. I wouldn't call it outright deception though.

No. If it actually is "slightly one-sided", then it is either 1) bugged or 2) intentionally misleading (my bet in that case is on #1). If the outcomes don't hold over a large number of trials then it is neither and OP's test case was just unlucky.

You are describing a case where the predictor is measurably, reproducibly wrong, to the point that experienced players can just memorize the discrepancy. That is not strategy; once you know the lie you can adjust for it every time...but then why include it? I haven't always agreed with Firaxis' design, but I'm not willing to accuse them of making the UI lie on purpose, it's too far-fetched. But even if not WAI, in the scenario you describe it's still lying.

the code can't completely predict what tile you are actually going to be on when your attack lands.

That's...an interesting assertion. If that belief held, then the pathing the game generates for your attack route would sometimes be incorrect and your unit would wind up in a different place than the game tells you. Which would TOTALLY be acceptable right? After all, "some unpredictability is good for the game". Why not sometimes have the game move your units where you don't want them to go?

I haven't seen that since the "ranged attack = actually move forward" bug from vanilla, and that got patched out.

No? You don't want units moving randomly to places other than ordered? Then don't argue for a wrong display predictor, you can't rationally argue for one and not the other .

Again, I'm not talking about "50% of the time you do more or less damage and can't 100% predict the outcome" as the problem. I'm talking about the case where the projected damage doesn't hold over a large sampling, either because the predictor bar is just wrong or it's not applying modifiers that are actually used.

If you move the knight to the tile with the flanking bonus first, and then hover over the enemy to bring up the damage predictor, it will then include the flanking bonus.

Yes, you have demonstrated that this aspect of the UI would be bugged; two different average expected outcomes predicted on an identical attack in an identical situation (which is nonsensical), with nothing but trial and error to learn why the game lies to you in the less micro intensive case.

Sorry to nitpick, but this seems like kind of a small aspect of the game.

It's also the topic of the thread, which is why we're discussing it lol.

 

[beetle]

The combat predictor makes a one time mathematical estimate. Actual combat “rolls the dice”. The algorithms are not identical.

My statement all along has been *if* the outcomes hold over a larger sampling, the user interface is lying

Since the algorithms are different, it is a mathematical certainty that the larger the sample the more one would expect a demonstration of divergence. But “lying” is not a fair characterization of a UI element advertised as showing an estimation. Especially since we both agree that the differences are not intentionally misleading.

Not displaying the flanking bonus when the flanking bonus is actually in effect (and would consistently show up if moving units painstakingly one hex at a time) is a UI failure.

Wait, the combat preview does display information for flanking bonus. So are you are saying that does not actually figure into the estimated damage calculation? What about the GG bonus? I agree that is sloppy programming (a “bug” as you say). But the actual boosts from flanking and GG are marginal enough that it won’t change the success/failure outcome much of the time.

Which gets at the point, how useful is the combat predictor? I really only pay attention when I am trying to finished off a very weakened enemy unit with my own weakened units. It was also useful when I was learning how close my GG had to get to combat, and how the flanking bonuses stacked. So I think the flawed predictor, and maybe it is more flawed than I thought, is better than no predictor.

 

[TheMeInTeam]

The combat predictor makes a one time mathematical estimate. Actual combat “rolls the dice”. The algorithms are not identical.

Quoted statement is still missing the point I've made.

Since the algorithms are different, it is a mathematical certainty that the larger the sample the more one would expect a demonstration of divergence. But “lying” is not a fair characterization of a UI element advertised as showing an estimation. Especially since we both agree that the differences are not intentionally misleading.

I'm saying the UI is lying, not Firaxis. Firaxis did not make the UI lie intentionally, but the UI still lies.

Your statement doesn't logically follow. I will create an example with more extreme numbers that is otherwise identical, to illustrate:

Say you go back to Civ IV, and attempt 100 battles, each with 90% odds to win. However, when you fight these battles you win 5 times, rather than something around 90 or within 3 standard deviations of 90.

Are you willing to conclude in the above scenario that the UI is fine because the algorithm displaying 90% is different from the one being used to resolve the combat? If you are not, then you can't rationally be willing to use it to refute wrong displays at closer, harder to detect values.

Wait, the combat preview does display information for flanking bonus. So are you are saying that does not actually figure into the estimated damage calculation?

The post of mine you quoted was in response to a poster describing such a scenario where it was not used in the predictor, but still applied. I pointed out that such was not actually a "random" factor at all and would be a UI bug.

But the actual boosts from flanking and GG are marginal enough that it won’t change the success/failure outcome much of the time.

Smaller deviations from expected value are harder to detect and matter less frequently, but a wrong predictor is still a problem (wrong UI in general is a problem). Worse is that the only way to know when these bonuses are still being applied is through trials. If you start stacking these "hidden" bonuses, the damage differential adds up. An experienced player will have a more accurate expectation of the outcome, but if someone is trying to use the UI alone, they will constantly be making strategy estimations on wrong expected values and having to update.

Not being able to trust the UI is a slippery slope in the game experience. It's hard to blame someone for confirmation bias when they have actually seen the UI mislead them previously.

Which gets at the point, how useful is the combat predictor? I really only pay attention when I am trying to finished off a very weakened enemy unit with my own weakened units.

How useful is it? If it's inaccurate, it is decidedly less useful than it should be. At this point, you're a deity player who can rely on experience to constrain your anticipated outcomes. However, you're also (factoring all the people that never post here but play the game) in a very high percentile among players. People with less experience need to lean on the UI more than experience.

If a predictor is sufficiently flawed, it's harmful rather than helpful. If it's only a little flawed, then it's less helpful than it should be, but might not be a net negative. It's still a problem if players have good reason not to trust the game interface, however.

One of the reasons I dropped vanilla and stopped playing for years was because of crap like this. I tried to attack with a catapult and it moved instead. I would hit "end turn" with the city showing positive food and then the city would re-arrange the citizens working tiles and starve down a population between turns. I would give units orders and they would buffer inputs...only sometimes, leading to inane outcomes different from displayed. Coupled with long waits between turns and what was at the time a shallow experience, I quit.

Different people will quit at different thresholds, however a UI that isn't trustworthy is a serious barrier to games. The lingering issues in Civ V are comparatively minor to what they once were (IE game-altering in many cases), but it says a lot if we have to rely on experience instead of the UI to know how the game rules will be applied, even now.

 

[beetle]

The post of mine you quoted was in response to a poster describing such a scenario where it was not used in the predictor, but still applied.

To the best of your knowledge, is the predictor using flanking bonuses in its calculations or not?

Different people will quit at different thresholds, however a UI that isn't trustworthy is a serious barrier to games.

I still occasionally throw a garrisoned unit into the water instead of moving the workboat that just spawned. I mostly blame the UI. I am not unsympathetic to the larger point you are making.

As far back as I can remember, civ has always offered some kind of combat estimate. Can we not just agree to disagree as to if the current version is lying and broken?

 

[TheMeInTeam]

As far back as I can remember, civ has always offered some kind of combat estimate. Can we not just agree to disagree as to if the current version is lying and broken?

I don't think two rational agents discussing the same knowns can reasonably agree to disagree.

What I will agree to is that we don't have enough evidence here to demonstrate the UI is lying in the context the OP describes, but we do have enough evidence (given track record in particular) to at least suspect it.

As for UI, the biggest pet peeve for me right now is selecting/cycling units and input buffering. That is a non trivial issue. Unit cycling doesn't work when stressed in turn timer environments. The whole "desync might mean game over because at least one person can't move when rehosting" aspect of MP is pretty bad too of course.

Civ V is still in a state where TBS from over 20 years ago are objectively superior in UI accuracy, input buffering, and unit cycling. That's pretty weak for a title as popular as this.

 

[beetle]

I don't think two rational agents discussing the same knowns can reasonably agree to disagree.

Fair enough. But one of those rational agents is using unnecessarily pejorative terms and both rational agents have failed to answer simple yes/no queries. So my preference would be to go onto something else...

 

[TheMeInTeam]

Fair enough. But one of those rational agents is using unnecessarily pejorative terms and both rational agents have failed to answer simple yes/no queries. So my preference would be to go onto something else...

Unnecessary perjorative? We're talking about the UI element here, it's not going to hurt its feelings .

Don't get too bent out of shape about the word choice. Even if I concede that the development team did not make the UI wrong on purpose (and I have), UI wrong display will still meet at least some English definitions of "lying". It doesn't have to meet *every* definition for me to use it w/o it being "unnecessary perjorative". It's not like we disagree on the facts or are unclear on position here, no need to get upset over word choice alone.

I gave a conditional yes. My memory is imperfect, and I don't have access to means of settling that uncertainty at the moment. Giving a more definitive yes absent evidence to back my (shaky and conditional) convictions would be disingenuous. On the other hand, the next time I encounter the UI not displaying some factors and save evidence, I'd be willing to bet on my position outright. For example:

To the best of your knowledge, is the predictor using flanking bonuses in its calculations or not?

Going by what previous poster said, the answer is "it depends". I don't have perfect recall as I've trained myself to not rely on the predictor, so I don't want to say that his example is a fact in-game (unless we get screenshots showing the difference of course).

The OP's test case is even more murky due to the sampling. I think it is wrong to hold a belief one way or another based on that because the evidence is too weak, and thus structured my argument contingent on future evidence.