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

[Hedgefund]

Lowbies should get the Offensive Amplifier from p2w if you're not happy with your hit rate.  10% to-hit bonus (among other bonuses).  Cheap for any budget up the low 20s.  

 

And I have to lol at complaints about this game's RNG.  Play some XCOM if you want to see brutal consequences of missing a 98% shot.

[Twisted Toon]

2 hours ago, Hedgefund said:

Lowbies should get the Offensive Amplifier from p2w if you're not happy with your hit rate.  10% to-hit bonus (among other bonuses).  Cheap for any budget up the low 20s.  

 

And I have to lol at complaints about this game's RNG.  Play some XCOM if you want to see brutal consequences of missing a 98% shot.

That would be why "That's X-Com" is a thing.

 

[Hedgefund]

"That's X-Com baby" to be more precise.

[Sovera]

On 11/3/2020 at 10:09 PM, TimesSeven said:

Okay, I understand and I'll drop this topic, but honestly, it's less fun to play this game at lower levels because of these reasons. Assuming I have a defense slot to put Kismet, I'd need a few million infamy just to purchase it. As a noob, I don't have that kind of money (if I were an actually newbie to the game).

I'm not sure how much of a newbie you actually are, so take this in the spirit of helping, but the guide in my signature helps with money making. At level 8 you can earn around 3 million by doing a Taskforce that takes 25-40 minutes to do. It is a bit less involved than Luminara's example of making money, but I prefer it since it involves running around playing the game instead of playing the AH (ultimately playing the AH is much better in terms of cash, but it's not what I log into City of Heroes to do). The other link in my signature has a post on page 3 which documents my starting with zero inf on a brand new character and after one day of playing the game (not playing the AH) I had something like 27 million with multiple IOs costing in the 10 mill range. This can be repeated by anyone who cares since it only involved me leveling.

 

I have not seen it mentioned here but on top of Beginner's Luck lowbies benefit greatly from another mechanic: Amplifiers from the P2W vendor. They cost a pittance at level 1 (24k for eight hours of all three Amplifiers) which turns lowbies into mini-Wolverines.

 

Finally I will add that people should try to be flexible. If they cannot find enough people red side to run a DFB in the low levels just for the reward, then hop to blue side, get it done, then hop back to red side and continue playing.

 

Edited November 6, 2020 by Sovera

[Hedgefund]

12 minutes ago, Sovera said:

 

I have not seen it mentioned here but on top of Beginner's Luck lowbies benefit greatly from another mechanic: Amplifiers from the P2W vendor. They cost a pittance at level 1 (24k for eight hours of all three Amplifiers) which turns lowbies into mini-Wolverines.

Um, like 3 posts above yours?

[Sovera]

Just now, Hedgefund said:

Um, like 3 posts above yours?

My apologies then, I missed it.

[Luminara]

While I was cutting, splitting and shouting at firewood, it occurred to me that there may be, in fact, a problem with the hit chance formula.  Granularity.

 

We're not rolling an icosahedron (20-sided die) when we make hit checks, we're comparing an XX.XX% number to a YY.YY% number.  That's 10,000 possible rolls.  10,000.  Ten thousand.  We refer to "capped" hit chance as having a 1/20 chance to miss, or, alternatively, a 5/100 chance.  But it isn't.  It's a 500/10,000 chance.  And while all of those are mathematically identical, functionally that leaves 500 chances of failure, not 1 or 5.  Granted, that still means 9500 of the rolls are going to be hits, and over time, it should average out that way, but 500 possible miss rolls on every check is... nuts.  Precision down to the hundredth of anything is the kind of thing you need for piston ring tolerances and rapid production machinery gearing, not a video game hit check.

 

The formula seems too precise.  Too granular.  Too many possible rolls for players to miss.  That's clearly why they had to code the streak breaker and apply it to hit rolls.  It's not there to throw players a bone, it's there to compensate for the overly-complex nature of the hit formula.

 

And the 95% cap would actually be creating that problem by permitting those 500 potential rolls to exist.  It's ensuring that no matter what, the player would have 500 opportunities to miss.  The margin for failure is much wider when viewed from the perspective of the actual math, rather than simply considering it to be equivalent to 1/20.

 

Furthermore, the streak breaker itself lack the same granularity.  It's too broad in scope, with huge margins differentiating between thresholds of forced hits.  Consequently, players are royally screwed by that mechanic if they're 00.01% below a threshold, or they're forced to slot extra Accuracy or ToHit to go above a threshold, thereby sacrificing slotting for something else, and even slots in something else.  There's no parity between the cap, the formula and the streak breaker, they're all using different approaches in an attempt to reach the same goal.

 

Having thought about it from that perspective, I think the HC team needs to spend a little time poking around in these mechanics, maybe trying a few alterations.  The streak breaker should match the granularity of the hit check formula, the hit check formula just doesn't need that level of granularity with the cap in place, and the cap is counter-productive to preventing streaks or working within the confines of the streak breaker thresholds (in fact, the cap is completely pointless at 95%, it should be at 90.01% since that's where the streak breaker kicks on with a forced hit after one miss.  functionally, they're stepping on each others' toes).

 

Frankly, I'm surprised the engine hasn't shit itself into oblivion, trying to juggle all of that.  @Jimmy, thoughts?

[tidge]

Some minor editing down to the bits I want to comment on:

1 hour ago, Luminara said:

And the 95% cap would actually be creating that problem by permitting those 500 potential rolls to exist.  It's ensuring that no matter what, the player would have 500 opportunities to miss.  The margin for failure is much wider when viewed from the perspective of the actual math, rather than simply considering it to be equivalent to 1/20.

 

Having thought about it from that perspective, I think the HC team needs to spend a little time poking around in these mechanics, maybe trying a few alterations.  The streak breaker should match the granularity of the hit check formula, the hit check formula just doesn't need that level of granularity with the cap in place, and the cap is counter-productive to preventing streaks or working within the confines of the streak breaker thresholds (in fact, the cap is completely pointless at 95%, it should be at 90.01% since that's where the streak breaker kicks on with a forced hit after one miss.  functionally, they're stepping on each others' toes).

 

The one piece of the Streakbreaker that doesn't appeal to me is: the Streakbreaker doesn't trigger on next roll that would miss, it triggers on the next roll because "it is time for you to have a forced hit". These are very different things! When I'm at 'finl chance >= 95%', I'd prefer it if the streak was broken when I got the second miss, and not on the followup roll no matter what.

 

The RNG appears to be a pseudo-random "space filling" algorithm design to guarantee a flat spectrum of results. To put this in the context of a ">=95% "final chance to hit", for such a pseudo-RNG (for historical implementations that look to 'evenly populate' the space of potential results) the chance of getting a second "1-in-20" result after a first "1-in-20" is less than 1-in-20.

 

I'd like to measure how big this effect is, but my data collection skill are not up to the challenge.

[Peacemoon]

I think assault rifles burst and full auto are examples where the mechanics can make missing infuriating.

 

Full auto only has one hit roll. So you either land every shot or miss every shot.

 

Imagine what your character must be doing to miss every shot 😄

 

and I always think some aoe attacks should always hit if the enemy is caught within. Patches don’t miss so..
And maybe have diminishing returns for accuracy over 95% - especially if you debuff a targets defence

 

some of this would smooth the edges without making any big changes to the core mechanic. I think it would be good myself. 

[Blastit]

32 minutes ago, Peacemoon said:

Imagine what your character must be doing to miss every shot 😄

 

The deal with "spray and pray" is that your prayers aren't always answered. Maybe there's a saint of firearms you can ask for a favour.

[jubakumbi]

2 minutes ago, Blastit said:

The deal with "spray and pray" is that your prayers aren't always answered. Maybe there's a saint of firearms you can ask for a favour.

Great.

Now I want to make "The Holy Handgun" ...

[Blastit]

It's good that there's no "holy" powerset in the game because it means people have to do their own interpretation, which is a lot more interesting and a lot more appropriate for the subject.

[CrudeVileTerror]

I'm pretty sure there are some "holy" weapons, though, as Temp Powers.

https://hcwiki.cityofheroes.dev/wiki/Holy_Shotgun_Shells

[macskull]

No, no, "holy" in that context just means it puts lots of holes in the enemy.

[Twisted Toon]

Basically, for the "Pray and Spray" Assault Rifle power, you'd want to use the Rain of Arrows mechanic, but less arcing of the projectiles.

To target that needs to be picked, just the location.

It would work.

Now, which power in the AR power-set would you be willing to toss out?

Personally, I could see replacing Full Auto with this. You could even keep the same name.

[RogerWilco]

On 11/6/2020 at 8:37 PM, Luminara said:

While I was cutting, splitting and shouting at firewood, it occurred to me that there may be, in fact, a problem with the hit chance formula.  Granularity.

 

We're not rolling an icosahedron (20-sided die) when we make hit checks, we're comparing an XX.XX% number to a YY.YY% number.  That's 10,000 possible rolls.  10,000.  Ten thousand.  We refer to "capped" hit chance as having a 1/20 chance to miss, or, alternatively, a 5/100 chance.  But it isn't.  It's a 500/10,000 chance.  And while all of those are mathematically identical, functionally that leaves 500 chances of failure, not 1 or 5.  Granted, that still means 9500 of the rolls are going to be hits, and over time, it should average out that way, but 500 possible miss rolls on every check is... nuts.  Precision down to the hundredth of anything is the kind of thing you need for piston ring tolerances and rapid production machinery gearing, not a video game hit check.

 

The formula seems too precise.  Too granular.  Too many possible rolls for players to miss.  That's clearly why they had to code the streak breaker and apply it to hit rolls.  It's not there to throw players a bone, it's there to compensate for the overly-complex nature of the hit formula.

 

And the 95% cap would actually be creating that problem by permitting those 500 potential rolls to exist.  It's ensuring that no matter what, the player would have 500 opportunities to miss.  The margin for failure is much wider when viewed from the perspective of the actual math, rather than simply considering it to be equivalent to 1/20.

 

Furthermore, the streak breaker itself lack the same granularity.  It's too broad in scope, with huge margins differentiating between thresholds of forced hits.  Consequently, players are royally screwed by that mechanic if they're 00.01% below a threshold, or they're forced to slot extra Accuracy or ToHit to go above a threshold, thereby sacrificing slotting for something else, and even slots in something else.  There's no parity between the cap, the formula and the streak breaker, they're all using different approaches in an attempt to reach the same goal.

 

Having thought about it from that perspective, I think the HC team needs to spend a little time poking around in these mechanics, maybe trying a few alterations.  The streak breaker should match the granularity of the hit check formula, the hit check formula just doesn't need that level of granularity with the cap in place, and the cap is counter-productive to preventing streaks or working within the confines of the streak breaker thresholds (in fact, the cap is completely pointless at 95%, it should be at 90.01% since that's where the streak breaker kicks on with a forced hit after one miss.  functionally, they're stepping on each others' toes).

 

Frankly, I'm surprised the engine hasn't shit itself into oblivion, trying to juggle all of that.  @Jimmy, thoughts?

 

Sorry, but this is just mathematically incorrect. Your "perspective of the actual math" just is not.

Each "to hit" roll is a separate random event. 1/20 and 500/10,000 makes no difference.

Those same "500 potential rolls" exist with a humble 20-sided die, if you roll it 500 times.

The problem is that you are rolling those 500 times every day you play, maybe even more.

1/20 and 500/10,000 are equivalent, as long as the random events are independent.

The streak breaker messes with that a little, but in the player's favour.

 

The precision just helps with debugging and people parsing logs to optimize their build. It makes no difference to the math.

And it better shows what the computer is actually doing, the internal number it uses probably has even more precision, about 6-7 decimal digits if it is using 32 bits.

95% cannot even be represented precisely in binary, so it's probably using 94.9999988079071044921875 %

 

And a few months back there was a huge thread (which I think was already mentioned here), that looked at all the details of the combat and to hit mechanics. It is behaving as it is supposed to.

Edited November 10, 2020 by RogerWilco

[tidge]

56 minutes ago, RogerWilco said:

Sorry, but this is just mathematically incorrect. Your "perspective of the actual math" just is not.

Each "to hit" roll is a separate random event. 1/20 and 500/10,000 makes no difference.

Because of the way Streakbreaker forces a hit while ignoring the actual result (and then resets the 'streakbreaker counter') I don't believe it is 'mathematically correct' to treat the 'rolls of the d20' as independent. A true 'streakbreaker' wouldn't fire until the next result of a roll was an actual 'miss'; that's the point at which there is an actual streak of misses that require 'breaking'.

 

Spoiler

An odd variant of the game that we don't have would be if Streakbreaker counted rolls in the 'streakbreaker tiers' irregardless if those rolls were hits or misses, and then forced a hit after the counter exceeded the threshold. For example, half of the rolls in the '>.9' tier will be hits ignoring the counter, and the other half will be misses that increment the counter.

 

The tiers are (from the wiki):

 

Final to-hit : misses allowed

>.9 : 1
.8-.9 : 2
.6-.8 : 3
.4-.6 : 4
.3-.4 : 6
.2-.3 : 8
0 -.2 : 100

 

 

It is subtle, but the RNG generating flat results over the spectrum 0.0 < P < 1.0 (forgive me, but I don't know what the exact limits are) doesn't mean that a character at the ceiling of 95% final-chance-to-hit is actually hitting enemies 95% of the time. This is because Streakbreaker is artificially throwing out one of the results of RNG.

 

Spoiler

In the case of the 95% final-to-hit ceiling, for a flat RNG: 19 times out of 20 Streakbreaker isn't actually needed to break a streak.

 

If I have a true peeve about the RNG/To-Hit system, this is it! I can accept the mathematics of probability, it's these human choices to interfere with the maths that I find annoying!

 

To my mind, there are two different experiments that I'd do if I could. I'd do this at the ceiling of 95%, but only because the math is easier and it ignores the effects of final-to-hit chances being in different tiers.

 

1) With Streakbreaker off, record every attack cast against every target and see what the actual fraction of hits is. My expectation is that the fraction of hits is going to be 95%. Changing the code isn't really an option, but I assume that back-in-the-day some of this was done pre-streakbreaker.

 

2) There is an easy way to see if Streakbreaker is actually behaving as if breaking a streak of 'unfortunate' RNG rolls, provided that we can log the result of the 'to hit' rolls that are ignored by streakbreaker. The ceiling case of 'final to-hit chance' of 95% is easy to achieve without worrying about debuffs and the like... so run a level 50+ through something like the Sewer Network of Tunnels of the Trolls (to restrict the routes of enemies who run away) and see how often the forced hit by streakbreaker aligns with a to-hit roll that would have been a miss. Also record every attack cast against every target and see what the actual fraction of hits is.

 

My hope would be that only 5% of the time would Streakbreaker align with an actual miss roll, but that by comparing the actual hits with the expected hits (when streakbreaker is forcing hits) the final fraction of hit enemies will be less; something like 94.7%. The difference between 94.7% and 95% is pretty small, so a lot of data would have to be collected to see this difference... but if the results aren't close, that would show up with less data.

 

Now having blathered about all this, I do think there is a sort of experiment we can do without either 'turning off streakbreaker' or 'trying to capture data that may not appear in the logs (i.e. the ignored rolls), but it wouldn't be conclusive (*1) just informative:

 

3) We can look at occurrence rate of RNG rolls in the range P<  0.05, as well as the occurrence rate of consecutive RNG rolls in that same range. I would hope that these mimic the situation at the ceiling of 95%. Obviously streakbreaker is still going to get in the way or recording some of the rolls, but (assuming a character could survive) it may be possible to 'turn down' your own characters final to hit chance such that streakbreaker interferes with the data record less frequently. My suspicion is: that because everything points to the RNG populating the result space (0.0 < P <1.0) evenly/flatly that a followup result in the (same) range P <0.05 is happening less than 5% of the time.

 

(*1) It wouldn't be conclusive because while I would expect that the results of the RNG near the boundary 0.0 would be the same as near the boundary 1.0, I know enough about probability density functions and priors to know that there can be subtle effects at one boundary and not the other... and with the rather crude implementation of streakbreaker I wouldn't trust that a similar crude approach I can easily believe that some other odd bias would be present at one end and not the other.

 

I think this week's statistical Riddler from FiveThirtyEight is tangentially related to this question: (my wording) given that you know just how biased a 2-sided coin is (i.e. P <> 0.5), can you establish the limit of the algorithm to achieve a final result of P=0.5.

[RogerWilco]

43 minutes ago, tidge said:

Because of the way Streakbreaker forces a hit while ignoring the actual result (and then resets the 'streakbreaker counter') I don't believe it is 'mathematically correct' to treat the 'rolls of the d20' as independent. A true 'streakbreaker' wouldn't fire until the next result of a roll was an actual 'miss'; that's the point at which there is an actual streak of misses that require 'breaking'.

I wanted to avoid adding the effects of the streak breaker to the discussion as that messes with the math, and I am not entirely sure I fully understand what it does.

My comment was purely on 1/20 and 500/10,000 not being the same because of "the perspective of the actual math", which seemed to be the opposite of what I know about "actual math" , statistics and computer science.

 

There was a big tread about streak breaker a few months ago. People there verified the odds using a lot of data from combat logs. It does what it is supposed to do, which helps the player.

As I understand it, it prevents you from missing more than 50% of the time, if you have a high chance to hit, while otherwise that would be a thing because random numbers are random.

 

What a lot of people also do not understand, is that with a 95% chance to hit, you already have a more than 50% chance to miss when you target 14 enemies. 0.95¹⁴ = 0.4876 chance to hit all 14

 

I am to tired right now to have anything sensible to say about alternatives to streak breaker and the rest of your post.

Edited November 10, 2020 by RogerWilco

[tidge]

2 minutes ago, RogerWilco said:

There was a big tread about streak breaker a few months ago. People there verified the odds using a lot of data from combat logs. It does what it is supposed to do, which helps the player.

The recent thread I recall was simply a collection of the (as logged) to hit rolls. There was some concern that there was a 'plateau' of  to hit rolls in the >0.95 range but the discovery (IIRC) was that there were auras in play that were also making to-hit checks that weren't being logged (unless they miss). I don't specifically recall any analysis of streakbreaker except that the code was presented.

 

My own posts in that thread were written while not knowing that streakbreaker code 'throws away' the result of a roll when you are 'due for a hit'. All praise to @Eclipse.for including the source code snippets in that thread.

 

 

[Luminara]

3 hours ago, RogerWilco said:

Those same "500 potential rolls" exist with a humble 20-sided die, if you roll it 500 times.

 

500 checks in a 1-20 range with 1/20 being the clamped miss rate gives 500 potential misses.

 

500 checks in a 1-10,000 range with 500/10,000 being the clamped miss rate gives 250,000 potential misses.

 

Simply increasing the number of checks doesn't bring the two to equivalency.  If anything, it emphasizes the problem with the formula, rather than proving that there is no problem.  You can't treat a 500/10,000 range as a single number.  It isn't.  It's 500 possible results out of 10,000, not 1 possible result out of 20.  Every hit check has 500 chances to miss, not 1.  Statistically, they're the same, and they may both average out to the same over time without the streak breaker fudging the results, but this isn't about statistical averages, it's about probability, large numerical ranges, the hit rate clamp and how they're all interacting.  Statistics and probability are two different things.  I'm addressing the latter, you're addressing the former.

[RogerWilco]

19 hours ago, Luminara said:

 

500 checks in a 1-20 range with 1/20 being the clamped miss rate gives 500 potential misses.

 

500 checks in a 1-10,000 range with 500/10,000 being the clamped miss rate gives 250,000 potential misses.

 

Simply increasing the number of checks doesn't bring the two to equivalency.  If anything, it emphasizes the problem with the formula, rather than proving that there is no problem.  You can't treat a 500/10,000 range as a single number.  It isn't.  It's 500 possible results out of 10,000, not 1 possible result out of 20.  Every hit check has 500 chances to miss, not 1.  Statistically, they're the same, and they may both average out to the same over time without the streak breaker fudging the results, but this isn't about statistical averages, it's about probability, large numerical ranges, the hit rate clamp and how they're all interacting.  Statistics and probability are two different things.  I'm addressing the latter, you're addressing the former.

I think I don't understand what you are trying to say then.

 

Are you saying that if I would roll a 10-sided die, where 1 would be a miss, or a 20-sided die where 1 and 2 would be a miss, these have different probabilities?

I think both have a probability of 1/10  and so would 1-500 on a 10,000 sided die.

 

Maybe we can clarify things by talking about smaller differences/numbers, and then we extrapolate from there.

I am really puzzled by the point you're trying to make.

Edited November 11, 2020 by RogerWilco

[Luminara]

1 hour ago, RogerWilco said:

I think I don't understand what you are trying to say then.

 

Are you saying that if I would roll a 10-sided die, where 1 would be a miss, or a 20-sided die where 1 and 2 would be a miss, these have different probabilities?

I think both have a probability of 1/10  and so would 1-500 on a 10,000 sided die.

 

Maybe we can clarify things by talking about smaller differences, and then extrapolate.

 

No, what I'm saying is that the streaky behavior of hit rolls is a result of the increased opportunity for misses created by the hit chance clamp.

 

Using your previous 500 roll example, if you roll an icosahedron 500 times, you will have occasional streaks of your predefined 5% failure rate (miss streaks), but they'll be rare.  Mathematically, it's very unlikely to roll, for example, seven 1's in a row. The probability of that occurring is very low, and drops with each additional 1 rolled.  Streaks aren't a common occurrence when we're using small ranges of numbers and single-digit failure points.

 

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.

 

Statistically, the overall hit rate would still average out to 95% over time, but I'm not addressing the average over time, I'm looking at streakiness and why it's happening.  It appears to be happening because the formula creates the possibility for it to happen, through high granularity, and the clamp increases the probability of it happening by locking in a specific and comparatively wide range of sequentially numbered results which can be considered misses.  The precision allows it occur, the clamp makes it more likely.

[Shred Monkey]

On 11/6/2020 at 1:37 PM, Luminara said:

While I was cutting, splitting and shouting at firewood, it occurred to me that there may be, in fact, a problem with the hit chance formula.  Granularity.

 

We're not rolling an icosahedron (20-sided die) when we make hit checks, we're comparing an XX.XX% number to a YY.YY% number.  That's 10,000 possible rolls.  10,000.  Ten thousand.  We refer to "capped" hit chance as having a 1/20 chance to miss, or, alternatively, a 5/100 chance.  But it isn't.  It's a 500/10,000 chance.  And while all of those are mathematically identical, functionally that leaves 500 chances of failure, not 1 or 5.  Granted, that still means 9500 of the rolls are going to be hits, and over time, it should average out that way, but 500 possible miss rolls on every check is... nuts.  Precision down to the hundredth of anything is the kind of thing you need for piston ring tolerances and rapid production machinery gearing, not a video game hit check.

 

The formula seems too precise.  Too granular.  Too many possible rolls for players to miss.  That's clearly why they had to code the streak breaker and apply it to hit rolls.  It's not there to throw players a bone, it's there to compensate for the overly-complex nature of the hit formula.

 

And the 95% cap would actually be creating that problem by permitting those 500 potential rolls to exist.  It's ensuring that no matter what, the player would have 500 opportunities to miss.  The margin for failure is much wider when viewed from the perspective of the actual math, rather than simply considering it to be equivalent to 1/20.

 

Furthermore, the streak breaker itself lack the same granularity.  It's too broad in scope, with huge margins differentiating between thresholds of forced hits.  Consequently, players are royally screwed by that mechanic if they're 00.01% below a threshold, or they're forced to slot extra Accuracy or ToHit to go above a threshold, thereby sacrificing slotting for something else, and even slots in something else.  There's no parity between the cap, the formula and the streak breaker, they're all using different approaches in an attempt to reach the same goal.

 

Having thought about it from that perspective, I think the HC team needs to spend a little time poking around in these mechanics, maybe trying a few alterations.  The streak breaker should match the granularity of the hit check formula, the hit check formula just doesn't need that level of granularity with the cap in place, and the cap is counter-productive to preventing streaks or working within the confines of the streak breaker thresholds (in fact, the cap is completely pointless at 95%, it should be at 90.01% since that's where the streak breaker kicks on with a forced hit after one miss.  functionally, they're stepping on each others' toes).

 

Frankly, I'm surprised the engine hasn't shit itself into oblivion, trying to juggle all of that.  @Jimmy, thoughts?

I'm getting out of my depth, but, a computer is going to make a 32 bit (or 64 bit?) number anyway, right?  Reducing the granularity is just done by rounding something that starts off as more precise.

[Luminara]

51 minutes ago, Shred Monkey said:

I'm getting out of my depth, but, a computer is going to make a 32 bit (or 64 bit?) number anyway, right?  Reducing the granularity is just done by rounding something that starts off as more precise.

 

That's already done by the engine.  It either rounds up, rounds down or just drops everything after the hundredth when calculating hit chances.  Granularity is relative in this context.  In context, it's a question of the granularity of 1-20 versus 1-100 versus 1-10,000, and how that corresponds to the 5% clamp, and specifically, the granularity of 1 miss result versus 5 miss results versus 500 miss results within that 5% and the probability of streaking miss results.

[tidge]

On 11/6/2020 at 2:37 PM, Luminara said:

Having thought about it from that perspective, I think the HC team needs to spend a little time poking around in these mechanics, maybe trying a few alterations.  The streak breaker should match the granularity of the hit check formula, the hit check formula just doesn't need that level of granularity with the cap in place, and the cap is counter-productive to preventing streaks or working within the confines of the streak breaker thresholds (in fact, the cap is completely pointless at 95%, it should be at 90.01% since that's where the streak breaker kicks on with a forced hit after one miss.  functionally, they're stepping on each others' toes).

I'd never advocate lowering the final ceiling of 'to hit' to 90%, I do think it is mathematically silly that (with a 95% final chance to hit) after 1 miss there is now a forced hit (100% with no regards to the roll) when for a 'fair' RNG the actual chance of a hit is 95%.

 

If the RNG is 'unfair enough' that it is required to trigger an auto-hit after a single 1-in-20 miss, I don't know why we'd believe that we're still getting 'fair' 19-in-20 rolls after the streakbreaker count is reset.

Spoiler

I can certainly speculate on the reason behind the streakbreaker.

 

It appears to me (in the absence of actual results and analysis from the RNG) that the streakbreaker is really intended to be nothing more than a panacea implemented to soothe (potentially) hurt feelings of players who are missing with their attacks. This may actually work (as a panacea) when swinging away against enemies where the final to hit chance for all attacks in the chain is 20%: "just 7 more wild swings and I'll hit!", but at the high end of the to-hit probability spectrum (P= 95%) I find the 'forced to hit by streakbreaker' has an opposite effect for me, as I have a deep suspicion that the forced hit was actually wasted on a roll that would have hit otherwise and is much more likely to guarantee that one of my next ten rolls is going to miss rather than one of my next twenty rolls.

 

Rather than soothing my nerves, it is more like an allergic reaction!

 

I could work out the posterior expectations for the results (no allergy to applying probability theory) but without being able to collect data in a controlled way (*1) I feel like the best that could be done in the game is demonstrate that the rate of missing attacks (at the P=95% ceiling) is statistically significantly greater than 5%. I don't believe the RNG is (directly) responsible for this (see linked threads above), my feeling is that it streakbreaker is hurting the hit-to-miss ratio. I'd like to be able to disprove this, or at least measure the actual size of the (possible) effect.

 

(*1) Ideally, we'd be able to position a large group of indestructible targets for which (ceiling, P=95%) attacks could be made to test both single-target and AoE attacks.