NPC Streakbreaker

[Ultimo]

What the heck are you all still raging about?

 

I said I thought the streakbreaker for npcs was something new.

It was pointed out that this is not the case, so I said, OK, THERE'S NO ISSUE THEN.

 

And yet there's FIFTEEN more posts.

 

It's fine, I got the answer I needed, if you guys want to go on about it, knock yourselves out, just leave me out of it.

[Seed22]

11 hours ago, ScarySai said:

 

It kinda already is.

I don't have one, so I wouldn't actually know on that front. But this would make it disgustingly worse

[macskull]

4 hours ago, Ultimo said:

I said I thought the streakbreaker for npcs was something new.

It was pointed out that this is not the case, so I said, OK, THERE'S NO ISSUE THEN.

Are you upset because you thought NPCs getting access to the streakbreaker mechanic was a recent change, or are you upset because NPCs get streakbreaker at all? This reply makes it seem like the former but your OP definitely says otherwise.

[Ultimo]

42 minutes ago, macskull said:

Are you upset because you thought NPCs getting access to the streakbreaker mechanic was a recent change, or are you upset because NPCs get streakbreaker at all? This reply makes it seem like the former but your OP definitely says otherwise.

Ok, since I need to clarify.  I'm not upset about anything.  I just encountered a mechanic I thought was new, that NPCs get the benefit of the streakbreaker.  I thought that was a bad idea, because it nullifies the defense of not getting hit, no matter how much you do to prevent it (I kind of still do).  However, it was pointed out that this mechanic has been in place all along.  Since I didn't notice any problem before, that suggests there IS no problem.  I just never noticed it in the combat log before.

 

So, it's NOT an issue.  My original complaint is founded on a misconception, so it can be ignored.

[tidge]

I have a different take: I pretty much think it should be tested if removing Streakbreaker would have any noticeable change on the game. Hear me out.

 

We know that the pseudo-RNG is incredibly flat in populating results between [0,1]; this is by design. This has been repeatedly demonstrated by multiple players. This means that there really isn't anything like "streaks" (lucky or unlucky).

 

The streakbreaker was added, as near as I can tell, to offer a "good feels" to players who don't understand what the RNG is doing, and/or players who feel like "I should be due for a hit, gorramit." It was sold as a "help players with floored ToHit" but it is almost certainly hurting results where players have a ceiling ToHit.

 

I am convinced that Streakbreaker doesn't prevent a call of RNG, just that it ignores RNG results (for some 'rolls'... ToHit checks), so that Streakbreaker is actually muddying the waters. The ur-example is if a player (or critter) has a 19-in-20 chance of hitting, but misses, Streakbreaker ignores the next roll and makes it an auto-hit. However, because of the flat RNG, there was (at worst) only a 5% chance that the followup roll was going to miss... so 95% of the time streakbreaker is just tossing out rolls.

[Hedgefund]

I'm 100% in favor of removing streak breaker.  I too suspect it was thrown in to lessen the "ACCURACY WAS NERFED" complaints when someone missed a 95% 4 times in a row, as can happen with a real RNG.

[tidge]

7 minutes ago, Hedgefund said:

I'm 100% in favor of removing streak breaker.  I too suspect it was thrown in to lessen the "ACCURACY WAS NERFED" complaints when someone missed a 95% 4 times in a row, as can happen with a real RNG.

 

/agreed.

 

Missing "four times in a row" at the ToHit ceiling would be highly improbable with a truly random set of rolls, but it is even less likely with the (flatly populating) pseudo-RNG. This is because the (pseudo)RNG  explicitly tries to "fill" the possible result space evenly. I'll repeat: "Streakbreaking" is inherent to this sort of pseudo-RNG. At the low end (a 1-in-20 chance ToHit) streakbreaker only kicks in on the 101st roll, so it would also be weird (given the flat nature of the results) that RNG simply wouldn't have given a result in [0, 0.05] by then.

 

The ONLY non-feels reason I can think of to kluge in Streakbreaker that I can think of is if there really is something wrong with the RNG under specific circumstances. We can't see ALL the calls to RNG (for example, for drop rewards or %proc rolls) but the ones we do see have completely flat results in the [0, 1] probability space.

[Rudra]

3 hours ago, tidge said:

I have a different take: I pretty much think it should be tested if removing Streakbreaker would have any noticeable change on the game. Hear me out.

 

We know that the pseudo-RNG is incredibly flat in populating results between [0,1]; this is by design. This has been repeatedly demonstrated by multiple players. This means that there really isn't anything like "streaks" (lucky or unlucky).

 

The streakbreaker was added, as near as I can tell, to offer a "good feels" to players who don't understand what the RNG is doing, and/or players who feel like "I should be due for a hit, gorramit." It was sold as a "help players with floored ToHit" but it is almost certainly hurting results where players have a ceiling ToHit.

 

I am convinced that Streakbreaker doesn't prevent a call of RNG, just that it ignores RNG results (for some 'rolls'... ToHit checks), so that Streakbreaker is actually muddying the waters. The ur-example is if a player (or critter) has a 19-in-20 chance of hitting, but misses, Streakbreaker ignores the next roll and makes it an auto-hit. However, because of the flat RNG, there was (at worst) only a 5% chance that the followup roll was going to miss... so 95% of the time streakbreaker is just tossing out rolls.

 

2 hours ago, Hedgefund said:

I'm 100% in favor of removing streak breaker.  I too suspect it was thrown in to lessen the "ACCURACY WAS NERFED" complaints when someone missed a 95% 4 times in a row, as can happen with a real RNG.

I would personally prefer we retain streak breaker. As someone who with a 95% chance to hit was defeated by equal con enemies at default difficulty (on multiple occasions) back on Live because absolutely none of my attacks ever landed in the fight, I can attest to streak breaker having an impact. Not a constant impact, but an impact.

[tidge]

1 minute ago, Rudra said:

 

I would personally prefer we retain streak breaker. As someone who with a 95% chance to hit was defeated by equal con enemies at default difficulty (on multiple occasions) back on Live because absolutely none of my attacks ever landed in the fight, I can attest to streak breaker having an impact. Not a constant impact, but an impact.

 

This is why it bears testing, because anecdote is not the singular form of data.

 

I know that (on Live) it was very common for players to not quite understand the debuffs from something like common Circle of Thorns.

[Stormwalker]

On 2/8/2025 at 11:41 AM, macskull said:

There is no case in which the streakbreaker mechanic will force hits at a higher rate than what the attacking character should already be hitting. To use OP’s example:

1. A Super Reflexes Scrapper with its auto and toggle powers active has 16.65% defense, assuming no other sources of defense - and no slotted enhancements in any of those powers.
2. If that Scrapper is being attacked by an even-con minion, that minion will have a (50-16.65=33.35%) chance to hit, or roughly 1 in 3.
3. A 33.35% chance to hit puts that critter in the streakbreaker bracket that requires six misses before forcing a hit. The chance of an enemy missing six times in a row with that hit chance is less than 9%.
4. In other words, in this specific scenario it’s more than three times as likely for your character to simply get hit because of a successful hit roll than it is for your character to get hit because of streakbreaker.

 

Actually, below level 10 the melee auto power isn't available yet.  The only melee defense available is the toggle.

 

Now, for ranged defense, you'd be correct.

 

I put together a "way too many defense slots" build while writing my post, and it came out to roughly 21%/29%/3% melee/ranged/aoe.  Of course, it was a ludicrous build.  Having seven defense SO's slotted at level 9 is absurd.

[Shin Magmus]

5 hours ago, Stormwalker said:

 

Actually, below level 10 the melee auto power isn't available yet.  The only melee defense available is the toggle.

 

Now, for ranged defense, you'd be correct.

 

I put together a "way too many defense slots" build while writing my post, and it came out to roughly 21%/29%/3% melee/ranged/aoe.  Of course, it was a ludicrous build.  Having seven defense SO's slotted at level 9 is absurd.

     Well yeah, as an experienced player you know that slotting like that at low levels actually makes the game harder.  If you aren't going to use any attuned unique IOs like Panacea proc in Health (available at low levels), then your character has no sustain.  Experienced players know that spending those slots on Acc/Dmg in attacks will end fights faster, meaning you spend less End on attacks and spend less Health because enemies didn't get to launch as many attacks. 

 

     Especially on Melee ATs, optimizing damage is a significantly higher priority than bulk if you are soloing.  Nobody else is there to deal damage for your, so you can't build like a 0 DPS Meatshield which would only be viable on teams.  Whether or not Streakbreaker is actually an issue, Ultimo was the architect of his own demise with a bad build and poor insp management.

[Stormwalker]

2 hours ago, Shin Magmus said:

     Well yeah, as an experienced player you know that slotting like that at low levels actually makes the game harder.  If you aren't going to use any attuned unique IOs like Panacea proc in Health (available at low levels), then your character has no sustain.  Experienced players know that spending those slots on Acc/Dmg in attacks will end fights faster, meaning you spend less End on attacks and spend less Health because enemies didn't get to launch as many attacks. 

 

     Especially on Melee ATs, optimizing damage is a significantly higher priority than bulk if you are soloing.  Nobody else is there to deal damage for your, so you can't build like a 0 DPS Meatshield which would only be viable on teams.  Whether or not Streakbreaker is actually an issue, Ultimo was the architect of his own demise with a bad build and poor insp management.

 

Yeah.  The reason I created that build was to get a sense for "Ok, the OP said their defenses were 'pretty high'.  How high could they possibly be, and what kind of odds would that give them?"  So that when I wrote up my response I could base it on a roughly-best-possible-case scenario.  Even though that's really the worst case scenario in terms of an actual usable build.

 

At low level, the best defense is a good offense.  On an SR scrapper, I usually don't even take Focused Senses until level 12, because at very low levels you're better off spending the endurance on attacks than you are spending it on trying to run two toggles.

[macskull]

On 2/9/2025 at 6:33 AM, tidge said:

It was sold as a "help players with floored ToHit" but it is almost certainly hurting results where players have a ceiling ToHit.

Not really, the effect of streakbreaker actually increases the the number of attacks a player with max hit chance will land. Given enough attacks at 95% chance to hit, you’d expect your overall hit percentage to be 95% but streakbreaker makes it 95.24%. If the player has a 90% chance to hit, streakbreaker turns that into 90.91%, which means at the low end of the top bracket streakbreaker is actually helping a player hit ten percent more often.
 

10 hours ago, Stormwalker said:

Actually, below level 10 the melee auto power isn't available yet.  The only melee defense available is the toggle.

 

Now, for ranged defense, you'd be correct.

 

I put together a "way too many defense slots" build while writing my post, and it came out to roughly 21%/29%/3% melee/ranged/aoe.  Of course, it was a ludicrous build.  Having seven defense SO's slotted at level 9 is absurd.

Yeah, that was meant to be a quick back-of-the-napkin thought experiment. I could do it again with only the toggle power but the results will stay the same - the point of the math was to show that streakbreaker does not force hits at a rate higher than the attacker will already be rolling.

[tidge]

27 minutes ago, macskull said:

Not really, the effect of streakbreaker actually increases the the number of attacks a player with max hit chance will land. Given enough attacks at 95% chance to hit, you’d expect your overall hit percentage to be 95% but streakbreaker makes it 95.24%. If the player has a 90% chance to hit, streakbreaker turns that into 90.91%, which means at the low end of the top bracket streakbreaker is actually helping a player hit ten percent more often.

 

I think you have this backwards, because of the way the pRNG flatly populates the space [0, 1] (or if you prefer [0.95, 1] vs. [0, 0.95] A reasonably straightfoward way to test is to use the same AoE attack against a large number of con grey for which all the enemies are in the AoE. The number of multiple misses in those spawns (pick a low level hazard zone) doesn't jive with a 95.24% hit rate. Again, for hypothesis testing it doesn't take much data for high confidence, but it does take a lot for high power (to discriminate between 95% and 95.24%).

[macskull]

2 hours ago, tidge said:

 

I think you have this backwards, because of the way the pRNG flatly populates the space [0, 1] (or if you prefer [0.95, 1] vs. [0, 0.95] A reasonably straightfoward way to test is to use the same AoE attack against a large number of con grey for which all the enemies are in the AoE. The number of multiple misses in those spawns (pick a low level hazard zone) doesn't jive with a 95.24% hit rate. Again, for hypothesis testing it doesn't take much data for high confidence, but it does take a lot for high power (to discriminate between 95% and 95.24%).

So you are saying the streakbreaker mechanic makes you miss more? That doesn’t make any sense either, and the math is fairly straightforward.

 

If I attack 100 targets (it doesn’t matter if it’s the same target 100 times, or 100 targets once, or 10 targets 10 times, since streakbreaker only cares about who is attacking and what their last hit chance was) with maximum chance to hit I should hit 95 of those 100 attacks. Each of the five attacks that miss will guarantee a hit on the follow-up attack, which means that I will essentially land 100 hits with 105 attacks, which is higher than 95%. Yes, you’d need a lot of data to accurately test that, but that’s true for any RNG testing.

[tidge]

3 hours ago, macskull said:

So you are saying the streakbreaker mechanic makes you miss more? That doesn’t make any sense either, and the math is fairly straightforward.

 

If I attack 100 targets (it doesn’t matter if it’s the same target 100 times, or 100 targets once, or 10 targets 10 times, since streakbreaker only cares about who is attacking and what their last hit chance was) with maximum chance to hit I should hit 95 of those 100 attacks. Each of the five attacks that miss will guarantee a hit on the follow-up attack, which means that I will essentially land 100 hits with 105 attacks, which is higher than 95%. Yes, you’d need a lot of data to accurately test that, but that’s true for any RNG testing.

 

pRNG is not truly random, it flatly populates the result space [0, 1]

 

So let's say without Streakbreaker the ToHit rolls for a character ToHit > 0.90 are: 0.06, 0.16, 0.26, 0.36, 0.46, 0.56, 0.66, 0.76, 0.86, 0.96

(yes this is weird, but it is an example of a flat distribution on pRNG result space chosen to show the 'decades' of possibly results)

 

The final result (0.96) is a miss, so we have a 9-of-10 hits. The next (11th) is ignored [pRNG doesn't "not roll"] and treated as a hit no matter the actual result. *IF* the exact same 'decade' set of results occured, then (and only then) would streakbreaker be 'honoring' the 9-of-10 hits. If at any time the pRNG returns a 'missed' result EARLIER than the 1(skipped)+9(more) then the player will be missing more often than 9-in-10 times.

 

In the case of the ToHit ceiling: Having a streakbreaker for 0.95 is basically assuming that after a roll >0.95 that there was a 100% chance of the next roll missing... but really the chance was only 5%. The flat pRNG means that clumping in the regime [0.95, 1] is discouraged (but not impossible). We know that pRNG prefers to anti-clump from actual data.

Edited February 10 by tidge
role -> roll

[macskull]

1 hour ago, tidge said:

things

I have done long-term testing with hit rolls and proc chances in the past (think leaving a character in an AE mission overnight with a single attack on auto), on the order of thousands of attacks. I’ll have to see if I can dig up the results of those tests.

 

I’m also not sure if streakbreaker simply overrides the hit roll calculation for that attack, or if it’s entirely separate thing - I would need to look at the code to confirm.

[tidge]

5 hours ago, macskull said:

things

 

 

This was from a long-ago thread where IIRC a player was not accounting for something that wasn't logged.

[macskull]

12 hours ago, tidge said:

 

 

This was from a long-ago thread where IIRC a player was not accounting for something that wasn't logged.

Yeah, the numbers in that thread skewed higher above 95% since there are a lot of powers that don’t display a hit roll in the combat log unless they miss.

 

Unfortunately I still have not been able to dig up my results from back when I was testing the Theft of Essence proc. I may still have the raw chat logs floating around somewhere, but whatever spreadsheets I used to do the math have been lost to time.

[tidge]

On 2/10/2025 at 5:54 PM, macskull said:

If I attack 100 targets (it doesn’t matter if it’s the same target 100 times, or 100 targets once, or 10 targets 10 times, since streakbreaker only cares about who is attacking and what their last hit chance was) with maximum chance to hit I should hit 95 of those 100 attacks. Each of the five attacks that miss will guarantee a hit on the follow-up attack, which means that I will essentially land 100 hits with 105 attacks, which is higher than 95%. Yes, you’d need a lot of data to accurately test that, but that’s true for any RNG testing.

 

I don't have access to minitab or JASP, but I can back-of-the-envelope offer these numbers for sample sizes, to give you an idea of how much data you would have to collect to tell the difference between 95% ToHits (H0) and an alternative hypothesis of 95.2% ToHits (H1). For 80% Confidence and 50% Power, you would have to see something like 7927 hits out of 8329 attempts. For something like 95% Confidence and 80% power, I think the necessary number of swings would have to be on the order of 145.5 million.

 

This of course isn't the test I would do: I would try to count "misses", where the null hypothesis (H0) is 0.05 and the alternative hypothesis is something like 0.055 (H1, honestly this is just a guess). I think the numbers for 95% confidence and 80% power are only 12251 "swings", and if 653 (or more) "misses" occur that would reject the 0.050 miss rate in favor of the 0.055 miss rate.

 

I don't know that I believe the miss rate to be 0.055% (or higher!) but for the frequency at which I see a MISS floaty when targeting con-grey spawn sizes smaller than 10 with AoEs, makes me feel that the miss rate is greater than 0.050. For example, for 80% Confidence and 50% Power, it would only take a minimum of 78 misses from 1412 swings to reject 5% in favor of 5.5%. "Science" doesn't like such low Confidence and abhors such low Power... but the low power is a type II error about accepting the 5.5% miss rate as true, not an assessment about that the 5% miss rate. I feel that 80/50 is fine for "gut checks" that something isn't right about the null hypothesis.

 

Edited February 11 by tidge
removed a stray "%"

[macskull]

On 2/11/2025 at 9:05 AM, tidge said:

This of course isn't the test I would do: I would try to count "misses", where the null hypothesis (H0) is 0.05 and the alternative hypothesis is something like 0.055 (H1, honestly this is just a guess). I think the numbers for 95% confidence and 80% power are only 12251 "swings", and if 653 (or more) "misses" occur that would reject the 0.050 miss rate in favor of the 0.055 miss rate.

One of the reasons I like AE is because it makes doing long-duration testing like this a bit more practical. In this case, I dropped a character into a mission against NPCs with no attacks, put Fire Blast (with a 95% chance to hit) on auto, and walked away. The instance mapserved after a little over 16 hours but I got some numbers well in excess of what you were looking for:

 

Stat	1st run	2nd run	3rd run	Totals
Attacks	4248	2980	14533	21761
Misses	174	133	695	1002
Forced hits	174	133	695	1002
Total hits	4074	2847	13838	20759
%acc	95.90%	95.54%	95.22%	95.40%
%missed	4.10%	4.46%	4.78%	4.60%

 

(The "runs" are just how I had the chat logs broken up. The first instance was proof of concept, and the second instance ran through the night so the logs automatically generate a new day's file at midnight.)

Edited February 14 by macskull

[tidge]

Looks good! Keep in mind the numbers I gave were for a specific hypothesis test with lowish confidence (only 80%) and 50% power.

[macskull]

47 minutes ago, tidge said:

Looks good! Keep in mind the numbers I gave were for a specific hypothesis test with lowish confidence (only 80%) and 50% power.

I was basing these off your suggestions that I quoted in the previous post - from what I can tell the number of attacks in this data set results in 95% confidence and around 76% power to confirm the null hypothesis of 0.05 and reject the alternate hypothesis of 0.055. I think I would need another 1000 or so attacks to break past 80% if the calculator I was using is correct. I will fully admit stats is not my strong point so this is mostly me guessing at things.

 

It does track that the overall hit rate once streakbreaker is factored in is actually higher than 95%, though.

Edited February 14 by macskull

[tidge]

8 hours ago, macskull said:

I was basing these off your suggestions that I quoted in the previous post - from what I can tell the number of attacks in this data set results in 95% confidence and around 76% power to confirm the null hypothesis of 0.05 and reject the alternate hypothesis of 0.055. I think I would need another 1000 or so attacks to break past 80% if the calculator I was using is correct. I will fully admit stats is not my strong point so this is mostly me guessing at things.

 

It does track that the overall hit rate once streakbreaker is factored in is actually higher than 95%, though.

 

Now if only we could test it without Streakbreaker! Is it possible to add some attacks to the enemies to make sure that there any confounding issue? We know it is supposed to be seperate, but it would be nice to add checks to make the test more like common use to see what tracks.

[Luminara]

6 hours ago, tidge said:

Now if only we could test it without Streakbreaker! Is it possible to add some attacks to the enemies to make sure that there any confounding issue? We know it is supposed to be seperate, but it would be nice to add checks to make the test more like common use to see what tracks.

 

It is separate.  The streak breaker is not shared between any two sources.  It doesn't have to be tested or verified because pets and pseudo-pets would skew the streak breaker by an enormous amount, and we have a large body of evidence to indicating that it doesn't occur simply by noting the absence of mass complaints about missing in the controller and dominator forums.

 

There's no need to test NPCs separately, for the same reason.  There hasn't been one report of missing twice in a row with 95% hit chances that wasn't attributable to other known causes.  Ever.  In the entire operational history of the game.  ~15 years of empirical evidence has proven that there's no interaction between player character streak breaker forced hits and NPC streak breaker forced hits.