All I know is that with buildup and two +ACC SO enhancements in my AS, I sure miss more than 5% of the time...
And they still haven't responded to my question:
With 2 SO's slotted, that puts my accuracy at 75x1.66 = 124.5
With an accuracy cap of 95%, I should always be at that vs even con minons- which the streakbreaker chart says I should have a maximum of 1 miss in a row, yet I can easily demonstrate otherwise. Why?
[ QUOTE ]
bah, edited. I give up to your persistance. You win. :-) I go back to enjoying the game instead. 
[/ QUOTE ]
If a discussion of the mechanics of the game hurts your enjoyment of the game, then you *should* stop discussing them: that goes for anyone including me.
Fortunately, it doesn't for me. When I go out to eat, I often ask myself "How was this prepared? Is that paprika I taste? Could I make it myself? Could I make it *better*?" It doesn't in any way prevent me from enjoying the meal.
Probably even more relevant, I'm something of a cinema buff, so when I go to see a movie, I'm often thinking about how it was shot, how the sound was added, the cuts, the editing, the dialog, the casting, the set design, and the cinematography.
But not when I'm actually watching it. I do that on the drive home. When I'm watching it, I'm just watching it, or at least I try to. I replay it in my head when I'm analyzing it. I look at CoH in exactly the same way.
[Guide to Defense] [Scrapper Secondaries Comparison] [Archetype Popularity Analysis]
In one little corner of the universe, there's nothing more irritating than a misfile...
(Please support the best webcomic about a cosmic universal realignment by impaired angelic interference resulting in identity crisis angst. Or I release the pigmy water thieves.)
[ QUOTE ]
With 2 SO's slotted, that puts my accuracy at 75x1.66 = 124.5
With an accuracy cap of 95%, I should always be at that vs even con minons- which the streakbreaker chart says I should have a maximum of 1 miss in a row, yet I can easily demonstrate otherwise. Why?
[/ QUOTE ]
Clearly, there is a bug, or a set of bugs.
Clearly, they refuse to admit even the possiblity of it, attributing any and all assertions of anomally or odd-seeming behavior to "observer error" (i.e. you notice missses more, and so on) or a lack of true "understanding" of the streak breaker code.
I write computer simulation models for a living. I also have written computer-based gaming engines (very simple ones to be sure). I have spent most of my life paying attention to random behavior and looking for anomalous behavior in a supposedly random milieu. After all that, I look at the behavior of their game, and how "streaky" hitting and missing seems to be, and I can say, "That doesn't look right."
What I lack are the tools to actually generate the data necessary to statistically document it not "looking right" -- but there's no question at all in my mind that if I were to do the data analysis, it would demonstrate beyond question that they have a bug, or a set of bugs, somewhere in their code.
Denial ain't a river in Egypt, fellas. 
F
[ QUOTE ]
With 2 SO's slotted, that puts my accuracy at 75x1.66 = 124.5
With an accuracy cap of 95%, I should always be at that vs even con minons- which the streakbreaker chart says I should have a maximum of 1 miss in a row, yet I can easily demonstrate otherwise. Why?
[/ QUOTE ]
1. Your assumption of being at the accuracy cap is only true if you're fighting certain mob types. If you fight enemies with debuffs, defense powers, high enough level, etc. then you may no longer be capped.
2. The correct question is not "Why haven't they responded", it's "Why haven't you provided the proof?" Since you say you can easily demonstrate otherwise, do it. Show us a demorecord, Fraps, HeroStats log - something that bears out what you're saying. Make sure you note the exact time, server, and toon name so the Devs can look it up on their side too.
Skip
My Char. List and Market Transactions
HeroStats Developer
Legion of Valor
Iron Eagles
[ QUOTE ]
Couple clarifications.
....
I did locate a small bug, which is now fixed internally. Essentially the streak breaker would not become active until you landed your first hit (so you have to do it yourself, no freebies). Pretty minor in the grand scheme of all things accuracy-related, but it's fixed now regardless. Note that this bug affected players and non-players alike.
[/ QUOTE ]
This explains some of what I've been noticing. I think this bug typically affects me the most right after I've added an enhancement in my power and use it as my first attack. It seems like it will miss most of the time (instead of hitting most of the time). Then I just keep missing for a few attacks. 
[ QUOTE ]
And they still haven't responded to my question:
With 2 SO's slotted, that puts my accuracy at 75x1.66 = 124.5
With an accuracy cap of 95%, I should always be at that vs even con minons- which the streakbreaker chart says I should have a maximum of 1 miss in a row, yet I can easily demonstrate otherwise. Why?
[/ QUOTE ]
If you can answer three questions definitively and unambiguously, then you have a genuine (potential) streakbreaker bug.
1. WB already noted a special case bug: you have to hit at least once before the streak breaker engages. So any miss streaks that occur immediately upon login (or zoning, or entering missions, perhaps) don't count. Can you point to multiple miss streaks outside these cases?
2. The streakbreaker has a very, very long worst case scenario that it can lock into. If you attack *anything*, even *once* with an attack with net 20% or less chance to hit (something with very high defense, say a drone), the streakbreaker is essentially turned off for 100 swings. Can you point to a streak of misses where you can say with certainty that for 100 swings prior to the miss streak, you definitely did not swing at any high defense targets at all?
[note1: accuracy enhancements are proportional to net tohit after defense, so if you attack something with 30% defense, your net tohit with 2 acc SOs is actually (75%-30%) * 1.66 = 74.7%. There are things that are packing really high defense: i.e. rikti drones. whatever their defense is, its enough to lower your tohit well below 50% even with a lot of acc slotting.]
[note2: at some point in the near past, *everything* started using more defense. a lot of reports of "accuracy" being off I believe are tracable to villains being granted more defensive powers. Consider: prior to I4, CoT scientists were not suddenly growing rock armor and making themselves difficult to hit.]
3. Are you absolutely sure there isn't anything else that the game might be counting as a "hit?" For example, fire tanks with blazing aura turned on might be overlooking all the "hit" ticks of BA and only noticing several melee attack swings in a row. According to WB, only *autohitting* attacks are ignored by the streakbreaker; things like damage auras might throw off what someone apparently observes.
One definitive test of the streakbreaker that would suggests a bug is to find something you *know* you have high tohit against, something you essentially never miss at, and then watch to see if you can find that double-miss against that target alone, without shooting at any other type of target at all.
I can say that in every case where I've seen accuracy jump the rails, its happened at the end of very long attack tests. I theorized way back then that perhaps the streakbreaker was overflowing a counter or something if you actually attacked or allowed to be attacked by the same target exclusively for hours at a time: a corner case the code didn't account for.
As to actual streakiness, as soon as I have a system capable of running CoH again, I'm going to start serious accuracy sequence tests to see if I can reproduce my old pre-I5 results, and very specifically to see if I can reproduce the streakbreaker table.
[Guide to Defense] [Scrapper Secondaries Comparison] [Archetype Popularity Analysis]
In one little corner of the universe, there's nothing more irritating than a misfile...
(Please support the best webcomic about a cosmic universal realignment by impaired angelic interference resulting in identity crisis angst. Or I release the pigmy water thieves.)
Give me a shout when you do, I'd like to coordinate our methodologies for these tests. (And steal/brutalize your demolog python parser scripts)
I will dispute this.
I was fighting even cons with an Ice/Energy blaster and missed 21 times in a row landed one and then missed 17 in a row.
I did not demo record it but I did start counting when I missed 6 in a row.
[ QUOTE ]
1. Your assumption of being at the accuracy cap is only true if you're fighting certain mob types. If you fight enemies with debuffs, defense powers, high enough level, etc. then you may no longer be capped.
[/ QUOTE ]
Yes, it is an assumption, for a number of reasons:
1) Unless ALL lvl 48 minions have signifigant (More than 34.5%)(124.5 -34.5 still puts me at the 90% to-hit range which dictates a maximum streak of 1 miss) defense bonuses or accuracy debuffs, It holds true. Maybe it's just me, but I think this is a somewhat reasonable assumption to start from.
2) Not knowing the ACTUAL accuracy numbers, I can ONLY speculate.
If I knew what my intended to-hit was against higher or lower level foes, I could test it against them as well.
If I knew for certian the accuracy bonuses from Martial Arts and Focused Accuracy were, I could figure them into the equation as well- though including them would only further my point, so I can see why you wouldn't want me to know those numbers.
[ QUOTE ]
2. The correct question is not "Why haven't they responded", it's "Why haven't you provided the proof?" Since you say you can easily demonstrate otherwise, do it. Show us a demorecord, Fraps, HeroStats log - something that bears out what you're saying. Make sure you note the exact time, server, and toon name so the Devs can look it up on their side too.
[/ QUOTE ]
Well, I'll admit, I haven't done so yet, but ill get to work on it as soon as the game stops crashing every time I try to zone. Being as it's something I could easily produce on many occasions, I didnt really think specific proof was neccesary, as I figured many other people observed this behavior as well.
Arcanaville, thanks for your more thought out reply, I'll endeavor to do more research bfore I post a more thorough response, but I can answer 2 of your questions right away:
#1: I can't specifically say for sure one way or another, but with the number of miss streaks Ive observed, it would be difficult to imagine they are all aresult of this bug.
#3: I'm Martial Arts/Regen, with Body Mastery as my EPP. I have no autohit attacks, and only 2 AOEs: Dragon Tail, and Energy Torrent. With the exception of Brawl, which I haven't used in about 10 levels, all of my attacks are slotted with 2 Lvl 50 Accuracy SOs.
Now, as far as #2 goes, that's something I'll have to look at more closely. This is the first time I've heard stated that Accuracy Debuffs/Defense are applied to my base to-hit chance before it being multiplied by my enhancement/ accuracy buffs- that's seems rather counterintuitive as it means with a high enough defense, I wouldn't matter how much accuracy I had, I'd never be able to hit.
Not to challenge you, but I'd be very interested to see that statement documented somewhere, as this is the first Ive heard of it.
[ QUOTE ]
1) Unless ALL lvl 48 minions have signifigant (More than 34.5%)(124.5 -34.5 still puts me at the 90% to-hit range which dictates a maximum streak of 1 miss) defense bonuses or accuracy debuffs, It holds true. Maybe it's just me, but I think this is a somewhat reasonable assumption to start from.
[/ QUOTE ]
34.5% DEF would actually give a net toHit (with your two ACC SOs) of
1.66*(0.75-0.345) = 67.23%.
In fact it takes a 17.7% Def value to drop you to 95% net ToHit. After that point it is a quick decrease of at a rate of -1.66% to hit for each % of Def.
As I stated in my answer to Arcanaville above, I was under the impression that it was
[Base Acc * [1+ Accuracy Buffs]] - [Defense +Accuracy Debuffs],
not [Base Accuracy - [Defense +Accuracy debuffs]] * [1+Accuracy Buffs].
While that seems like an odd way to construct this system, since the two of you seem to be in agreement on that equation, ill concede the point to you and Arcanville.
That still leaves the question of wether all lvl48 minions posess a [Defense + Accuracy Debuff] Value of -21 ((75-20=55*1.66= 91.3) remember, I have to be below 90% before a streak of 2 is acceptable), which I'd find a bit hard to believe.
[ QUOTE ]
I believe this is why your equation says that 75% will lead to 75.29% and the simulations show that it's more like 75.27%.
I just don't have the math background to point this out more eligantly.
[/ QUOTE ]
Also your simulations would need to run through 18.7 million iterations before you can say claim an sample uncertainty of 0.01% in your mean.
(0.75*0.25)/(0.0001)^2 = N samples.
I'm going to go ahead and trust the analytic solution that Pippy and I got two seperate ways.
[ QUOTE ]
Now, as far as #2 goes, that's something I'll have to look at more closely. This is the first time I've heard stated that Accuracy Debuffs/Defense are applied to my base to-hit chance before it being multiplied by my enhancement/ accuracy buffs- that's seems rather counterintuitive as it means with a high enough defense, I wouldn't matter how much accuracy I had, I'd never be able to hit.
Not to challenge you, but I'd be very interested to see that statement documented somewhere, as this is the first Ive heard of it.
[/ QUOTE ]
Its been carefully tested, and also confirmed by geko when he (very broadly) described the tohit formula in his Q&A, in this thread (see the answer to #8).
Its repeated here, annotated:
[ QUOTE ]
First, we need 2 things: The attackers ToHit (this includes all ToHit Buffs and debuffs),
[/ QUOTE ]
We infer from this that "Attackers ToHit" is the attackers base tohit, plus all tohit buffs, minus all tohit debuffs, although this is not explicitly stated.
[ QUOTE ]
the targets Defense (this also include all Defense buffs and debuffs). We then do the following:
(Total_To_Hit - Total_Defense)
[/ QUOTE ]
In other words, the intermediate sum of ToHit - Defense is calculated *first*.
[ QUOTE ]
That value is clamped between 5% and 95%. That is, you can never have more than a 5% chance to hit or miss a target. So if a target's Defense is greater than the attacker's ToHit, the attacker will still have at least a 5% chance to hit.
[/ QUOTE ]
Floors and ceilings.
[ QUOTE ]
We then multiply that number by the power's total Accuracy (total Accuracy = Accuracy + the power's Accuracy Enhancements).
[/ QUOTE ]
What Geko is calling here "Accuracy" is also referred to as "Base Accuracy" and "Inherent Accuracy" elsewhere, and is the inherent accuracy of the *attack* (note the phrase "the power's" above). Whenever the devs say a particular power has an accuracy bonus (i.e. snipes) they are referring to this number. For all normal attacks, "Base Accuracy" = 1.0.
Accuracy Enhancements are, of course, accuracy enhancements.
[ QUOTE ]
We again clamp the values between 5% and 95%.
[/ QUOTE ]
Floors and ceilings again.
[ QUOTE ]
So to sum up:
((Total_To_Hit - Total_Defense), 0.05,0.95 MinMax) x ACC = Chance To Hit.
[/ QUOTE ]
The final formula, which I will expand out for geko as:
(InherentPowerAccuracy + Accuracy Enhancements) x (Base ToHit + ToHitBuffs - ToHitDebuffs - Defense)
Or if you prefer:
(1.0 + InherentAccuracyBonus + AccuracyEnhancements) x [BaseToHit + ToHitBuffs - ToHitDebuffs - (Defense + DefenseBuffs - DefenseDebuffs)]
As a result:
[ QUOTE ]
1) Unless ALL lvl 48 minions have signifigant (More than 34.5%)(124.5 -34.5 still puts me at the 90% to-hit range which dictates a maximum streak of 1 miss) defense bonuses or accuracy debuffs, It holds true. Maybe it's just me, but I think this is a somewhat reasonable assumption to start from.
[/ QUOTE ]
It only takes about 22% defense to lower you to below 90% net tohit. (0.75 - 0.22) * 1.67 = .8851 = 88.51% (21% defense would be 90.18%). It doesn't have to be all of them, just enough of them to keep dropping you back down into the 2-miss range often enough to notice the streaks.
[Guide to Defense] [Scrapper Secondaries Comparison] [Archetype Popularity Analysis]
In one little corner of the universe, there's nothing more irritating than a misfile...
(Please support the best webcomic about a cosmic universal realignment by impaired angelic interference resulting in identity crisis angst. Or I release the pigmy water thieves.)
[ QUOTE ]
That's also interesting: either they couldn't have fully checked accuracy as they have been suggesting, or they knew defense was being hit harder than normal because of the streakbreaker. Or: the left hand analyzing defense didn't know what the right hand checking accuracy knew.
[/ QUOTE ]
Probably wasn't very in-depth analysis of streaks.
I made a 2 gig video of fighting in breakout. the largest miss sstreak i found was multiple 3 miss streakis, but no 4s (where the streak breaker kicks in to convert to a hit).
I also came across this.
[ QUOTE ]
Can someone look to see if when an Energy Melee punch disorients a foe... is it giving them a defense bonus of some sort? I am missing every swing until the disoriented foe becomes un-disoriented and I have my Accuracy slots with 1 acc +15 for a lvl 14
Support reply sent back was:
Thank you for contacting CoH support team. This is a known issue. We are working on it and hope to have it resolved soon. Please refer to the patch notes for any updates.
[/ QUOTE ]
My Current Belief is that Accuracy (and the streak breaker) is working in the correct ranges but that ...
(a) The rand function is very streaky and should be replaced
(b) Various bugs are effecting some encounters
(i) Streak breaker not kicking in until you hit once
(ii) Effective Auto-miss on disorientated foes
(iii) Effective Auto-miss on cone attacks
FYI: I asked WeirdBeard what else might cause a streak to be reset, other than a hit. He replied "Streak tracking resets every time you zone."
[ QUOTE ]
2. The streakbreaker has a very, very long worst case scenario that it can lock into. If you attack *anything*, even *once* with an attack with net 20% or less chance to hit (something with very high defense, say a drone), the streakbreaker is essentially turned off for 100 swings. Can you point to a streak of misses where you can say with certainty that for 100 swings prior to the miss streak, you definitely did not swing at any high defense targets at all?
[/ QUOTE ]
From WeirdBeard's post:[ QUOTE ]
Something to note, for those inclined to note such things. Because we do the lookup based on your worst to-hit in the series, the streak breaker is a bit less aggressive about breaking streaks than it might initially appear from the table. If you miss an ill-advised attack with a final to-hit of 0.15, you would in fact be allowed to continue the miss series for another 99 attacks, even if all the followup attacks are of capped to-hit, should you be unlucky enough.
[/ QUOTE ]
It seems like you've misunderstood WB's explanation of figuring worst to-hit. From a logical standpoint, it would make sense to keep track of the worst to-hit only when you start a miss streak. The streakbreaker resets once a hit is registered, so shouldn't the worst to-hit too? Note that WB said that the streak breaker is "a bit less agressive about breaking streaks." I'd consider turning off the breaker for 100 swings to be overagressive. Heck, from a coding standpoint, it's certainly harder to keep track of worst to-hit within the last 100 swings rather than simply resetting everything once a hit occurs.
[ QUOTE ]
The final formula, which I will expand out for geko as:
(InherentPowerAccuracy + Accuracy Enhancements) x (Base ToHit + ToHitBuffs - ToHitDebuffs - Defense)
Or if you prefer:
(1.0 + InherentAccuracyBonus + AccuracyEnhancements) x [BaseToHit + ToHitBuffs - ToHitDebuffs - (Defense + DefenseBuffs - DefenseDebuffs)]
[/ QUOTE ]
Just some questions to clear things up for me:
1) an AoE with 20% ACC penalty would be -0.2 for InherentAccuracyBonus? Burn would be around +1.0 InherentAccuracyBonus?
2) Where would the level difference bonus/penalty be applied? IAB or multiplied around the entire thing?
3) ACC inspiration is a ToHitBuff? Likewise, DEF insp. counts as a DefenseBuff?
Thanks!
[ QUOTE ]
[ QUOTE ]
2. The streakbreaker has a very, very long worst case scenario that it can lock into. If you attack *anything*, even *once* with an attack with net 20% or less chance to hit (something with very high defense, say a drone), the streakbreaker is essentially turned off for 100 swings. Can you point to a streak of misses where you can say with certainty that for 100 swings prior to the miss streak, you definitely did not swing at any high defense targets at all?
[/ QUOTE ]
From WeirdBeard's post:[ QUOTE ]
Something to note, for those inclined to note such things. Because we do the lookup based on your worst to-hit in the series, the streak breaker is a bit less aggressive about breaking streaks than it might initially appear from the table. If you miss an ill-advised attack with a final to-hit of 0.15, you would in fact be allowed to continue the miss series for another 99 attacks, even if all the followup attacks are of capped to-hit, should you be unlucky enough.
[/ QUOTE ]
It seems like you've misunderstood WB's explanation of figuring worst to-hit. From a logical standpoint, it would make sense to keep track of the worst to-hit only when you start a miss streak. The streakbreaker resets once a hit is registered, so shouldn't the worst to-hit too? Note that WB said that the streak breaker is "a bit less agressive about breaking streaks." I'd consider turning off the breaker for 100 swings to be overagressive. Heck, from a coding standpoint, it's certainly harder to keep track of worst to-hit within the last 100 swings rather than simply resetting everything once a hit occurs.
[/ QUOTE ]
Out of context, my statement poorly reflects the behavior of the streakbreaker. The intent was to point to situations were a report of the streakbreaker failing was unimpeachable, therefore I stated something much stronger than necessary.
If, however, I stated "can you be sure that you did not swing at any low tohit targets during any long terms miss streaks" that is a tricky thing to keep track of and observe, and therefore subject to being questioned. If, on the other hand, you can say "at no time at all in the last 100 swings did I attack any low tohit targets" then there cannot be *any* reason for the streakbreaker to be in the low mode at all.
What I should have stated is "the streakbreaker is essentially turned off for (up to) 100 misses."
Its a stronger statement than necessary, but its also a much easier thing to set up a test to verify: simply never attack high defense targets at all during your test.
For the record, I interpreted WBs post the same way you did: that the streakbreaker looks at the worst tohit in "the series" - meaning the current series of misses.
[Guide to Defense] [Scrapper Secondaries Comparison] [Archetype Popularity Analysis]
In one little corner of the universe, there's nothing more irritating than a misfile...
(Please support the best webcomic about a cosmic universal realignment by impaired angelic interference resulting in identity crisis angst. Or I release the pigmy water thieves.)
[ QUOTE ]
[ QUOTE ]
The final formula, which I will expand out for geko as:
(InherentPowerAccuracy + Accuracy Enhancements) x (Base ToHit + ToHitBuffs - ToHitDebuffs - Defense)
Or if you prefer:
(1.0 + InherentAccuracyBonus + AccuracyEnhancements) x [BaseToHit + ToHitBuffs - ToHitDebuffs - (Defense + DefenseBuffs - DefenseDebuffs)]
[/ QUOTE ]
Just some questions to clear things up for me:
1) an AoE with 20% ACC penalty would be -0.2 for InherentAccuracyBonus? Burn would be around +1.0 InherentAccuracyBonus?
[/ QUOTE ]
If I understand the devs correctly, I believe whenever a specific power has an inherent bonus or penalty to accuracy, its an "accuracy" bonus/penalty, in the sense of the equation. So yes.
[ QUOTE ]
2) Where would the level difference bonus/penalty be applied? IAB or multiplied around the entire thing?
[/ QUOTE ]
Level differences are said to be modifications to "tohit" and therefore they alter the BaseToHit in the equation. You can think of the level modifications as defacto tohit buffs or debuffs, or alternatively you can think of the BaseToHit as being itself something that is "looked up" per target and entered into the equation. Thus, player base tohit against even level minions is 75%, whereas base tohit against +1 minions is 65% (or you can think of it as a -10% tohit penalty against +1s).
[ QUOTE ]
3) ACC inspiration is a ToHitBuff? Likewise, DEF insp. counts as a DefenseBuff?
[/ QUOTE ]
I'm certain that defense inspirationos count as a defense buff - they raise defense, and during discussions of defense stacking they were said to raise "base defense" which for the purposes of that discussion implied they offered defense against everything (all ranges, all damage types) and stacked with everything.
I can't say with 100% certainty what acc inspirations do. I *believe* they boost Accuracy, like accuracy enhancements, but I've seen contradictory tests that people have done that seem to suggest they work one way or the other.
Assuming that there isn't a third way they can work, there is an easy way to determine which of the two they are: tohit or accuracy. Drag a low level character (say in the 20s) to PI, and chomp four acc insps, then shoot at something. If acc insps work like Accuracy buffs, then you aren't likely to hit anything. If they work like tohit buffs (vis-a-vis Aim) then you're going to hit everything almost all the time (and get sent to the hospital mighty fast - you probably wanna run this test on the test server).
For reasons I'm about to explain, I'm going to be (re)doing a lot of accuracy testing, so this is on my todo list, actually.
[Guide to Defense] [Scrapper Secondaries Comparison] [Archetype Popularity Analysis]
In one little corner of the universe, there's nothing more irritating than a misfile...
(Please support the best webcomic about a cosmic universal realignment by impaired angelic interference resulting in identity crisis angst. Or I release the pigmy water thieves.)
Weirdbeard has been graciously looking at my reported "triple-miss" problem, and based on the information he's provided to me, I have to say that without access to my files, I have to conclude there's some systematic error on my end, either in the demorecord itself, or my analysis of it. Its weird since the same scripts and techniques worked very well in other circumstances to measure accuracy and the streakbreaker reasonably well (it did confirm that the streakbreaker was working for villains, for example), but in the specific instance where I was seeing odd triplicate behavior, based on the info provided by WB, there's simply no way for what I was seeing to actually be happening. I'm prepared to conclude that, subject to getting a second look at the raw files, the triple-streak behavior has to be an error on my end.
Normally, I'm very meticulous about measurements, and try to triple check everything before posting, and in this case, a systematic error seems to have survived *five* separate checks on my part, but an error is still an error, and I apologize to anyone relying on it to form an opinion about the tohit calculator.
Because of that, I'm not going to say anything more about specific accuracy tests until I've had a chance to redo them all in a much more airtight fashion, which may take some time because this time I'm going to beat the numbers completely to death before I say anything at all.
I should mention what this does for my position on the random number generator and the tohit calculator, including the streakbreaker:
1. I've never suggested the random number generator itself was inherently streaky to the degree where it would create far out of bounds streaks on its own. Rand() has problems, but consistent streakiness is not one of them. It *can* create problems, including streakiness in certain circumstances, but they would be highly intermittent, and highly situational, and very dependent on how the random numbers were used. I don't and never did have direct evidence that the random number source was itself directly causing problems.
2. I *did* believe that there were corner cases where the tohit calculator behaved oddly. However, my primary example of that behavor now appears to be very likely flawed, which means I no longer have unambiguous evidence of corner case oddities.
3. I *did* have evidence that the streak breaker works, at least most of the time. I *don't* and never did have direct evidence that it failed to operate entirely, or operated consistently incorrectly in any particular situation.
Basically, I only had evidence of corner-case oddities, and that evidence can't be trusted anymore, so I no longer have direct evidence of tohit irregularities.
However, now that we know precisely how the streakbreaker is *supposed* to work, and can confirm whether it works correctly or not, two questions remain:
1. Does the streakbreaker *always* function as described?
2. If the streakbreaker functions exactly as described, does it introduce unwanted behavior, particularly to defense sets?
I suspect the answer to #1 is "almost always" but its always possible for there to be minor bugs in it, although because we now know how it works, we can construct precise tests to test its behavior.
I suspect the answer to #2 is that the SB, even when its working "properly" does do some non-intuitive things to defense sets at the low end (~ 20% net tohit) and in group situations (multiple attackers), but I can't say for sure if those effects are strong enough to care about.
Again, it looks like I screwed up this particular test; if I figure out precisely how a simple hit counter could fail this badly on a particular demorecord, I'll report back. And kudos to Weirdbeard, who took the time to research this carefully and give me the data to demonstrate that an error of some kind must have taken place (almost certainly on my end).
Yes, I do in fact feel like an idiot. The fact that my current RL assignment relies heavily on a lot of scripts processing a lot of data with absolutely zero defects is not helping my mood any. In any case, I'm going back and redoing all of my accuracy tests from scratch, including all base accuracy tests, all accuracy function validation tests, and all streak/streakbreaker-related tests, just as soon as I have a functioning system and the time. This time around, I'm releasing *all* of the raw data and scripts related to processing, so someone else can catch any stupid mistakes I might make. And I'm specifically going to go through my list of unanswered questions on how accuracy works and test them all, one at a time, until I have accurate answers for all of them.
[Guide to Defense] [Scrapper Secondaries Comparison] [Archetype Popularity Analysis]
In one little corner of the universe, there's nothing more irritating than a misfile...
(Please support the best webcomic about a cosmic universal realignment by impaired angelic interference resulting in identity crisis angst. Or I release the pigmy water thieves.)
[ QUOTE ]
...
What I should have stated is "the streakbreaker is essentially turned off for (up to) 100 misses."
Its a stronger statement than necessary, but its also a much easier thing to set up a test to verify: simply never attack high defense targets at all during your test.
For the record, I interpreted WBs post the same way you did: that the streakbreaker looks at the worst tohit in "the series" - meaning the current series of misses.
[/ QUOTE ]
Ah glad to have that sorted out. =) The original question about swings before a streak might really have seeded confusion.
Don't be too hard on yourself Arcana. The work you do is appreciated by many people here. Everyone makes mistakes.
For what it's worth, I tried to do some theoretical analysis on question #2. I'm certainly not skilled at this sort of thing, so I may have goofed something up.
The scenario is having one PC being attacked by 10 NPCs. For simplicity, I'm assuming the NPCs are all using the same attack, and attacking simultaneously, i.e. round 1 - all NPCs attack once, round 2 - they all attack again, etc. All attacks have an accuracy of 20%. There's a miss-streak limit of 8. It's also assuming the new streak-breaker fix is in place, i.e. so that it doesn't take a successful hit to start the streak tracking.
The question is: What are the chances that at least half of the 10 NPCs will miss 8 times in a row, creating a situation where 5 of them will then be auto-hitting on round 9? Note: this should result in at least 6 "simultaneous" hits taking place on round 9, because either more than 5 will have missed, or if it was exactly 5, one of the remaining 5 should hit (5 chances at 20%).
Please correct me if I'm doing something wrong here:
A 20% accuracy is a 80% chance to miss. The chance that a single NPC will miss 8 times in a row is:
(80/100)^8 = 0.16777216, or about 17%.
The chance that at least 5 of the 10 NPCs will miss 8 times in a row is (meaning at least 6 hits will occur on the ninth round):
COMBIN(10,5)*(0.16777216)^5*(1-0.16777216)^(10-5) = 0.013372457302% ~ 1.34%
*Note: the COMBIN() function is an Excel function. I got the formula from the following site and modified it slightly.
For comparison the chance that at least 6 of the 10 NPCs will all hit on the same round (streak-breaker aside) is:
COMBIN(10,6)*(.2)^6*(1-.2)^(10-6) = 0.005505024000% ~ 0.55%
So, the streak-breaker can increase the chances of clustered hits, but I guess the chances aren't all that great. Real experiments are probably still warranted, however.
I'm always boggled by people who talk about 6+ miss streaks.
If my Lvl 33 Kin/Elec Def were to miss 6 times, one of those would be Transference and I'd be dead already. I don't believe I've missed 3 times in a row in recent memory.
(Then again, I do have one or two SO ACC in every power, plus the Leadership toggle, and usually stick to Yellows & Oranges.)