Remembered one of the things I hate about CoH... its RNG.

[Luminara]

5 minutes ago, tidge said:

I'd never advocate lowering the final ceiling of 'to hit' to 90%

 

Nor would I, but the streak breaker forces a hit on the next attack at 90.01%, rather than 90.00%, which makes everything above 90.01% utterly irrelevant.  It doesn't make any difference if you're at 90.01% or exactly 95.00% or 141.59%, if you miss and your next attack has a hit chance higher than 90.00%, it's a forced hit.  Anything above 90% is functionally identical because of the streak breaker.

[tidge]

13 minutes ago, Luminara said:

 

Nor would I, but the streak breaker forces a hit on the next attack at 90.01%, rather than 90.00%, which makes everything above 90.01% utterly irrelevant.  It doesn't make any difference if you're at 90.01% or exactly 95.00% or 141.59%, if you miss and your next attack has a hit chance higher than 90.00%, it's a forced hit.  Anything above 90% is functionally identical because of the streak breaker.

It's irrelevant for streakbreaker, but having a 95% final chance to hit delays (over a 90% final chance to hit) getting into the streakbreaker situation at all.

[Luminara]

2 minutes ago, tidge said:

It's irrelevant for streakbreaker, but having a 95% final chance to hit delays (over a 90% final chance to hit) getting into the streakbreaker situation at all.

 

The point is, as long as the streak breaker exists, 95% is irrelevant period.  Whether the forced hit was at 90.01% or 95%, it's still a forced hit, and unless streakiness is addressed in some other way, or the streak breaker changed to respect successful hit rolls rather than kicking in even if a hit roll would've succeeded, you're always going to be beholden to the streak breaker's actions.  The streak breaker is what determines whether we have enough +Acc or +ToHit, not the hit chance clamp.

[DSorrow]

On 11/11/2020 at 3:21 PM, Luminara said:

Perform the same test with a 10,000-sided die, with failure defined as anything between 1 and 500 (or 9501 through 10,000).  The probability of streaks of failures increases because you have a significantly wider range of potential failures.  A 1 is a failure, and so is a 2, and a 3, and so on, up to 500.  Every roll has 500 potential failures, rather than 1, and with 500 failure points instead of 1, streaks are more likely to occur.

I'm not sure I follow the logic of why streaks are more likely to follow just because granularity is increased.

 

On a 20 sided die, the probability of rolling range [20, 20] for a miss is 1/20 (i.e. "miss" occupies 5% of the outcome range). Assuming rolls are independent (and no streakbreaker), the probability of a n attacks missing in a row should be (1/20)^n. If you have a 10000 sided die, the probability of rolling range [9501,10000] for a miss is 500/10000 (i.e. "miss" occupies 5% of the outcome range). The probability of n attacks missing in a row should be (500/10000)^n = (1/20)^n.

 

Increasing granularity should never increase streakiness because you're checking for if you hit a certain range of outcomes. The only way I can think of how it would increase is if you'd roll once for every potential failure (i.e. roll for 9501, check if it happens and if not, move on to 9052, etc.) until you exhaust all failures at which point you declare success, then increasing granularity would have a huge impact, but that's with an abomination of RNG process. If you roll once, then check whether you're in the range of failures or not, it doesn't matter how granular the range is. Basically, the "significantly wider range of potential failures" is completely offset by the significantly wider range of potential successes, or as in your example every "added chance of fail" is balanced by 19 "added chances of success".

 

Or in other words, there isn't a significantly wider range of potential failures because you're still going to land in the fail probability mass with equally likely whether it's made of 1 or 1000 discrete parts comprising the same fraction of all possible outcomes.

 

As a thought experiment, would the chances of streaks continue to increase up to infinity (if not infinity, what is the limit?) when we'd be able to generate infinitely granular numbers on a number line? Would the chances increase or decrease if the length of the number line and the part of it representing the outcome "miss" were scaled up or down by the same factor? 

Edited November 17, 2020 by DSorrow

[Luminara]

12 minutes ago, DSorrow said:

Assuming rolls are independent (and no streakbreaker), the probability of a n attacks missing in a row should be (1/20)^n. If you have a 10000 sided die, the probability of rolling range [9501,10000] for a miss is 500/10000 (i.e. "miss" occupies 5% of the outcome range). The probability of n attacks missing in a row should be (500/10000)^n = (1/20)^n.

 

Increasing granularity should never increase streakiness because you're checking for if you hit a certain range of outcomes. The only way I can think of how it would increase is if you'd roll once for every potential failure (i.e. roll for 9501, check if it happens and if not, move on to 9052, etc.) until you exhaust all failures at which point you declare success, then increasing granularity would have a huge impact, but that's with an abomination of RNG process. If you roll once, then check whether you're in the range of failures or not, it doesn't matter how granular the range is. Basically, the "significantly wider range of potential failures" is completely offset by the significantly wider range of potential successes, or as in your example every "added chance of fail" is balanced by 499 "added chances of success".

 

Or in other words, there isn't a significantly wider range of potential failures because you're still going to land in the fail probability mass with equally likely whether it's made of 1 or 1000 discrete parts comprising the same fraction of all possible outcomes.

 

As a thought experiment, would the chances of streaks continue to increase up to infinity (if not infinity, what is the limit?) until there was no granularity at all and we'd be able to generate real numbers on a number line? Would the chances increase or decrease if the length of the number line and the part of it representing the outcome "miss" were scaled up or down by the same factor? 

 

Thank you.  That was exactly what I didn't know and needed to understand.  I was wrong.  👍

[DSorrow]

14 hours ago, Luminara said:

 

Thank you.  That was exactly what I didn't know and needed to understand.  I was wrong.  👍

Cool if I could clarify, probabilities are often so unintuitive I usually have to double or triple check before I can have any confidence about how they should work...

[Eclipse.]

On 11/3/2020 at 2:23 PM, Doomguide2005 said:

My understanding is it will check all the foes "struck" against the streakbreaker and will do so no matter who you actually target with say Fireball.  That is even if you throw the fireball at the boss but catch all the rest of the minions etc. and if you are at 95% final hit chance you could conceivably hit only half the mob ... but the odds of being that unlucky are kind of like rolling 12 1's on a 20 in a row and having streakbreaker force 6 to be hit as a result.  At least that's how I interpreted the italicized section in the Attack Mechanics page of the Wiki dealing with the streakbreaker

I got pinged into this thread, so I certainly haven't read the whole thing.

I do want to highlight something important though. COMPUTERS ARE BAD AT GENERATING TRUELY RANDOM DATA. With that in mind, the odds might not be what you expect as if you were to roll a fair die, for example.

I'm not going to dig into what specifically the tohit check is using for random number generation, and even if I did the seed for the Homecoming servers might be different (ie. lower entropy) than on a server with more uptime or something. That said, CoH is largely C based, so I'm going to assume it uses the standard rand() function, which is actually a psudo random number generator. The numbers it creates are entirely predictable, given the same seed, unless it is seeded with a higher level of entropy from srand(). Even if that part is Python, 1.) Python is also C based 2.) Python rand() is also psudorandom

IDK about CoH, but I am aware of at least 1 other games that had "Unlucky Characters" that were actually caused by how the RAND was seeded for characters. (There was a huge article on it, but I can't find it at the moment). As well as a few games where you could eek out patterns and increase your odds by spamming /roll until you saw a favorable pattern...

It's possible that what ever is seeding the random data does make to-hit checks more likely to miss even if, in theory, you have a 95% hit chance. Assuming seeds are changed regularly you should still even out. 

If you want to setup some log visualization for City of Heroes it might help with that. It is worth noting that some things DO NOT LOG their hits, and can still eat up streakbreaker possibilities (Auras, and a handful of AoEs).
@valcryst81 did a great job setting that up initially... I meant to make it easier to stand up via docker, but got distracted with my real work/life for the last few months.
 

[Doomguide2005]

@EclipseI suggest reading Arcanaville's post on page 3 of the thread.

 

If this was a new set or AT and not one from issue 0 (a popular one at that) and there was something wonky with the rng generator and not just operator perception I'd expect a lot more stink after 9+ years and something reproducible.  None of which exists.  So if nothing else far more evidence is needed before deciding it's something other than a perception bias.  I've at least 2 characters with that set, including a Corruptor and never remotely felt anything was "off".

[Eclipse.]

3 hours ago, Doomguide2005 said:

@EclipseI suggest reading Arcanaville's post on page 3 of the thread.

 

 

Thanks for that!
Yeah, I have no reason to doubt @arcanaville's work.
I was just highlighting that, while it is extremely unlikely, you shouldn't expect that in any SMALL sample size (few hundred) that the random generation won't end up streaky.  Just pulled it up, looks like an average non-farming play session for me is ~500 attacks out logged,  and ~1800 attacks in logged. With highs closer to 15k attacks out and 20k in

I'd be curious how the random is seeded, but ultimately I've also never noticed anything especially egregious in my logging (I have a few hundred hours of Combat logs in Grafana, both my own and others spanning years of live play). The few things I have found surprising, I've been able to find obvious answers for looking at the source code (IE. Auras, and some other AoEs only logging misses).

Still, if it's seeded at shard creation it's always possible that you get 1 unlucky seed, but the chances of it persisting for more than 1 mission is just really really really unlikely. 99.99% of the time I'd still bet on it just the player's negativity bias.

Edited November 18, 2020 by Eclipse.

[FoxyPrime]

On 11/5/2020 at 4:13 PM, macskull said:

This is why the beginner's luck mechanic exists. There's also a huge difference between "feels like" and "is like," especially when the ability exists to prove one or the other. When people say they have issues like that ("I always pop Aim/BU/a tray of yellows and miss every attack!") I will ask for combat logs and inevitably find one of a few things is happening:

• The player is unable to produce the combat logs, making their claim unable to be tested
• The player produces combat logs, and after further examination everything checks out - this one can be tricky because there are some powers that do not show every hit roll in the combat logs so it can seem like you are missing twice in a row with 95% chance to hit but that is not what is actually happening
• The player produces combat logs that show an odd scenario which I'm unable to replicate
• The player has found a bug

The first two options happen far more often, the third option happens extremely rarely (I think maybe once in the last year and a half), and the fourth option almost never happens. There was a thread several months back about how the game's hit roll RNG was biasing numbers >95 for players only and not NPCs, but the end result of that thread was "many powers only show a hit roll on misses and not hits so it would naturally bias numbers >95 if you're only looking at combat logs."

 

EDIT: Another couple things go into this discussion. Perception bias is probably the biggest one, because you don't notice those 95% of times you hit a target, but you definitely notice the 5% of times you don't. The next one is simple probability and accepting that just because something is extremely unlikely doesn't mean it won't happen. For example, there's an 0.25% chance I will miss two attacks with 95% hit rolls in a row (not counting the one in between where streakbreaker forced a hit) which means it is unlikely to happen but not impossible.

None of this really addresses my point: that possibility of that bad scenario is, in of itself, just a waste of the player's time. It being statistically rare is completely irrelevant. 

 

What is the purpose of low level missing? How does it enhance the player experience? How does missing on a dice roll teach the player anything about playing the game? Is it fun for a mission to be dragged out even more because you arbitrarily lost a fight that you will later win with no problem and then repeated ad infinitum? The fact this discussion happened at all suggests it is not, in fact, universally fun. Since enhancements and such exists specifically to try and help solve this problem, then it stands to reason that itself is evidence the problem exists by design to give a sense of purpose to progression, but I think that sense of progression can be maintained without needing to make the low levels pointlessly irritating. Just tweak it so it's not so damaging for the player to arbitrarily miss or buff the inspirations.

[Luminara]

46 minutes ago, FoxyPrime said:

What is the purpose of low level missing? How does it enhance the player experience? How does missing on a dice roll teach the player anything about playing the game?

 

NPCs don't have brains.  NPCs don't have players controlling them.  NPCs have scripts, and those scripts don't contain strategic scenarios with conditional actions based on tactical analysis.  A critter can't evaluate a player's actions and determine the best course of action.  A critter can't respond to a situation by making an assessment and proceeding based on that assessment.  A critter can move, use powers, and respect specific engine triggers (status effects, Avoid, pathing).  The only thing that makes them even remotely challenging or threatening is having that small chance to defeat your character before you defeat them.

 

It's not punitive.  It's strictly so NPCs aren't completely, laughably non-threatening.  You, the player, have every advantage in every fight.  You have all of the attributes and capabilities that an NPC controlled by a scripted AI sequence can't replicate.  You have two to ten times as many powers, so you're significantly more likely to have another attack available.  You can adjust the way your powers function by slotting different types of enhancements.  NPCs are fixed.  They never level up.  They never slot anything.  They never improve their powers, they never develop better tactics.  They can't even use inspirations.  The odds are always in your favor.  Allowing misses to occur changes the odds, just a little, to add meaning, context and value to combat.  Not to torture the player, but to give the player some semblance of a fight, instead of an endless parade of piñatas.  With intellectual capacities just slightly higher than rocks, critters need help, like occasional missed attacks from the player, to be interesting, engaging and potentially capable of defeating the player, because they have none of the benefits that come from sitting on your side of the keyboard.

 

The AI in this game was the product of the greatest minds of a generation working together with the express purpose of building the dumbest moron who ever lived.  It's not remotely hard to beat, and a few misses are the only thing giving it any hope of surviving against your superior intellect and adaptability.

 

That's the purpose of missing, at every level.  Just to give NPCs a chance to not be completely overwhelmed when you roll your face back and forth on the keyboard.  It enhances the experience by adding the remote possibility of defeat in any combat situation.  And missing teaches players to pay attention to their hit chances and adjust their approach by slotting for +Acc or using powers which grant +ToHit, so they're less likely to miss in the future.  Not guaranteed, but less likely, because regardless of level, critters are just AI and scripts, and dumber than a post.

 

[Lunchmoney]

1 hour ago, FoxyPrime said:

None of this really addresses my point: that possibility of that bad scenario is, in of itself, just a waste of the player's time. It being statistically rare is completely irrelevant. 

Would you prefer 100% hits 100% of the time? 

[DoctorDitko]

Fascinating glimpse into player perceptions of probability. Thanks for this thread.

 

Here's a possibly relevant footnote:

https://en.wikipedia.org/wiki/Littlewood's_law

[TimesSeven]

I have woefully given up on allowing myself to become upset over a constant string of misses. One thing that everyone seems to agree on is that the streaks are streaky and anything under a (91? 92?) certain percentile can allow multiple misses to go unchecked. I have resigned to the fate that I am among the supernaturally unlucky.

 

What I find fascinating now is that a 95% chance to hit that misses, doesn't automatically generate a streakbreaker on the next shot. Three times in a five minute period, I have missed with a 95% chance to hit and the next shot didn't involve streakbreaker -- but the one after it did. Is this because of the update? I don't remember seeing anything that related to the streakbreaker, but then again, I have a life, a wife and a fulltime job. :P

[tidge]

1 hour ago, TimesSeven said:

What I find fascinating now is that a 95% chance to hit that misses, doesn't automatically generate a streakbreaker on the next shot. Three times in a five minute period, I have missed with a 95% chance to hit and the next shot didn't involve streakbreaker -- but the one after it did. Is this because of the update? I don't remember seeing anything that related to the streakbreaker, but then again, I have a life, a wife and a fulltime job. 😛

Is this happening with single-target attacks or AoE attacks?

 

Is this happening on a character with an aura that makes "to hit" checks?

[TimesSeven]

On ‎12‎/‎6‎/‎2020 at 4:24 PM, tidge said:

Is this happening with single-target attacks or AoE attacks?

 

Is this happening on a character with an aura that makes "to hit" checks?

Nope, this is happening on single-target attacks only. No auras, PBoE or AoE. I took five or six more screenshots just to make certain.

 

 

Edited December 8, 2020 by TimesSeven
Added photo

[Luminara]

The first attack, Throwing Knives, was at the hit rate cap, but any roll in the 95.01-100 range is a miss and the roll was 98.74, so it missed.

 

The second attack, Blaze, had a lower hit chance, 87.00%, and it also missed when the roll came up higher than that number, 94.52.

 

The streak breaker kicked in properly when you missed two attacks in a row and the lowest hit chance of the two was within the two miss allowance range, 80.01-90.00.

 

Had that second attack had a 90.01% or higher hit chance, the streak breaker would've forced a hit, bypassing any result of the hit check.

 

More information on the streak breaker and hit chance thresholds.

[Snarky]

lol  my current build for my main is a Dark/Dark PBAoE passive nightmare walking.  I use focused accuracy and sets that give accuracy bonuses.  +11% to hit, +68% Accuracy.  I always assume RNG is out to kill me.  Why I avoid Def sets.  I think that is why I like Dark as well, Res heavy and debuff opponents to-hit, as well as fear and confuse effects.  F Def, go offense!

[macskull]

On 12/6/2020 at 3:55 PM, TimesSeven said:

What I find fascinating now is that a 95% chance to hit that misses, doesn't automatically generate a streakbreaker on the next shot. Three times in a five minute period, I have missed with a 95% chance to hit and the next shot didn't involve streakbreaker -- but the one after it did. Is this because of the update? I don't remember seeing anything that related to the streakbreaker, but then again, I have a life, a wife and a fulltime job. 😛

Streakbreaker isn't determined when you miss with a power, it's determined when you activate a power. In your case you missed with a 95% hit roll which would make it seem like the next attack should be a guaranteed hit, but because the attack you activated after missing had less than a 90% chance to hit it moved into the lower bracket where two misses forces a hit. If your followup attack had >90% chance you would have seen streakbreaker force a hit.

[TimesSeven]

Oh, okay. That makes sense. Was wondering what was going on with that. I thought a miss with a 95% chance to hit guaranteed a streakbreaker as the next result. You can see how this would be confusing as a lot of the posts here made that declaration without mention of the next. 
 

All good now. Thanks. 

[TimesSeven]

1 hour ago, Snarky said:

lol  my current build for my main is a Dark/Dark PBAoE passive nightmare walking.  I use focused accuracy and sets that give accuracy bonuses.  +11% to hit, +68% Accuracy.  I always assume RNG is out to kill me.  Why I avoid Def sets.  I think that is why I like Dark as well, Res heavy and debuff opponents to-hit, as well as fear and confuse effects.  F Def, go offense!

Yeah, I'm starting to agree. My fav builds now all involve pets. Unless you can throw 91% tohit blasts every single time against every single opponent, you're going to get creamed in large groups OR you play with small groups at lower levels. And defense is a joke. RNG is a beast. 
 

I mean, storywise you're supposed to be an Incarnate. The baddest of the bad. Getting flattened by a group of Nemesis because of bad rolls is depressing. 

[Luminara]

6 hours ago, Snarky said:

F Def, go offense!

 

You do realize that -ToHit is almost identical to +Def (All (including Toxic, which has no Defense typing)), with the difference being the purple patch's effects on debuffs, right?  By using Dark/Dark's -ToHit as mitigation, you're not actually avoiding Defense, you're just using a different version of it.

[TimesSeven]

2 hours ago, Luminara said:

 

You do realize that -ToHit is almost identical to +Def (All (including Toxic, which has no Defense typing)), with the difference being the purple patch's effects on debuffs, right?  By using Dark/Dark's -ToHit as mitigation, you're not actually avoiding Defense, you're just using a different version of it.

I absolutely agree with this statement except there are a LOT of ways an NPC can nuke your defense:  swords, psionics, guns, radiation, water.. almost every powerset has at least one -defense in it somewhere. And some sets have a TON of -def. However, not a lot of NPC's have -tohit resistance (and if I'm wrong, please let me know). So in this regard, -tohit seems a lot more reliable than +def.

 

I learned that lesson the hard way after soft-capping my fire/dev blaster on ranged defense and then running into a group of eight psychics. A literal fifteen seconds later and I was face-planted on the floor. Wondering if a dark/dev blaster would have been better.  Maybe I should just scrap blasters and controllers and just be a freakin' tank. Less fun solo, but at least I don't die as much.

[Grindingsucks]

RNG's have always felt wonky to me, regardless of the game I'm playing, though this manifests in different ways depending on the game.

 

In CoH, I notice quite frequently the results are probably accurate, but weighted oddly, such that when I have a 95% chance to hit, instead of missing about once every 20 attacks, I seem to run a hot hit streak for like 30 minutes, then I'll roll over 95% three or four times in a row.  That happens an astonishing amount of the time.  I've also noticed both my attacks and the enemies attacks being incredibly accurate in sequence (rolling several instances of 5% or less, sequentially).

 

If I went back and did the calculations, it would likely average out properly to a 95% hit rate, but  the way it manifests during combat feels weird.  😁

[Luminara]

16 minutes ago, TimesSeven said:

However, not a lot of NPC's have -tohit resistance (and if I'm wrong, please let me know).

 

Everything above +0 resists all varieties of debuffs.  The greater the level differential, the stronger the resistance.  At +4, critters resist debuffs by more than half.  So yes, and no.  There aren't very many foes which specifically resist -ToHit, primarily AVs/scaled down EBs, but fighting enemies above your current level causes your debuffs to have less effect.

 

More information on debuff resistance due to level differential.

 

Also of note, a few power sets have the capability to debuff -ToHit resistance (this was added to Trick Arrows in the most recent update), but all lieutenants, bosses and AVs/scaled down EBs have limits to how much their resistance to -ToHit can be debuffed. Consequently, debuffing resistance to -ToHit is of limited value outside of AV fights.  One can debuff minion/underling -ToHit resistance significantly, but the time spent doing so is rarely worthwhile when it's simpler and quicker to just beat the snot out of them.

 

29 minutes ago, TimesSeven said:

-tohit seems a lot more reliable than +def.

 

Well, +Def is never comprehensive, covering all types and positions.  Most Defense is mitigation to either positional attacks or attacks with specific damage types.  There are +Def (All) powers, which do exactly that, provide Defense to all positions and damage types, but even +Def (All) doesn't provide any Toxic Defense because there isn't any Toxic Defense in the game.  -ToHit, on the other hand, affects every attack and every damage type used by the debuffed foe, which makes it far more effective as a mitigation tool.  With -ToHit, there is no Psi hole, or Toxic hole, or any hole.  -ToHit bypasses all of that, granting mitigation to every type and position equally.  If it weren't for the purple patch, -ToHit would be absolutely superior to +Def in all situations.  Even with the purple patch, -ToHit is excellent mitigation, but most players prefer passive mitigation in the form of toggles/set bonuses, rather than active mitigation which potentially interferes with their dee pee ess.

 

I can't address the value of -ToHit or +Def in Hami raids.  I don't attend them and I haven't researched Hami's mechanics since Paragon updated it.  I'll get around to it eventually.

 

53 minutes ago, TimesSeven said:

Wondering if a dark/dev blaster would have been better.

 

Dark Blast is an excellent set, for defenders and corruptors.  Blasters lose access to Night Fall, which significantly reduces the value of the set, in my opinion.  Being limited to mostly single-target attacks makes it much more difficult to stack -ToHit on multiple foes, which reduces the mitigation it could provide.  But if you're comfortable with limited AoE options, it can provide plenty of -ToHit to layer with +Def, even accounting for blaster debuff scalar values.  With Umbral Torrent to back it all up with reliable KB (or KD with a KB->KD IO), it should turn the tables on the Psi critters for you.

 

Fire, though... you really have to pair it with something which provides more mitigation than /Devices, or go all out and take every high damage power you can so you can obliterate spawns before they can hit you.  Fire's only bonus effect is MOAR DAMAGES NAO PLZKTHXLOL.  Damage is mitigation, but it has to be quickly delivered and in sufficient amounts to defeat everything before you're defeated, or it doesn't do the job.

 

For controllers, try /Trick Arrows.  I rather enjoyed my Fire/TA controller on the original servers, and with the improvements to TA in the latest update, it's undoubtedly an even better combination than it was then.  Between Smoke and Flash Arrow, you can build up over 30% -ToHit when slotted, and that's enough for you to soft cap versus +0 critters with only 15% Defense (All).  6% of that can come from the two 3% +Def (All) unique IOs.  Maneuvers offers 4% slotted.  You can find more +Def in a variety of powers and set bonuses, and one of the Interface options in the Incarnate system also adds -ToHit to every attack (including your pets').

 

Oh, and /TA also has two powers which debuff Defense, so you'll see fewer miss streaks in the long run.