D&D 5E Silvery Barbs, how would you fix it? Does it need fixing?

[Stalker0]

Redo the analysis mathematically with the assumption that this is a binomial population (success or failure) and the median roll is a "success" with a standard deviation of 0 and you will see what I am talking about.

For a binomial distribution, standard deviation: σ= √(npq)

Setting the deviation to 0,
0=√(npq)
0 = (npq)
0 = pq (we assume n = at least 1, since we have seen a success, aka a non-zero value).
q = 1- p (by definition)

0 = p(1-p)
0 = p - p2
p = 1 or 0 (aka 100% or 0% chance of success)

That is not a useful result, as we know monster saves fall between this.


As to your note above, I am not sure how you would take a median of 1 and make a calculation without some knowledge of p, I am not aware of any binomial distribution median formula that doesn't involve p in some way shape or form

 

[EzekielRaiden]

Doesn't any distribution become trivial when you assume SD 0? With SD 0, you have 0 variance, which means there is no spread of results--all results are the same.

 

[NotAYakk]

Just throwing out ideas so they may not be the most well thought out changes, but oh well.
I’d personally change it to 2nd level and remove the part of giving advantage to someone. affecting a saving throw as a reaction is still plenty good, and i feel people would still jump for this even with these alterations.

The giving advantage has a larger impact on how good it feels than how good it is.

"Reroll.. did nothing" feels sucky. "Reroll... oh well, someone got advantage" feels better.

The actual power budget of advantage isn't the largest part of the ability.

 

[NotAYakk]

For a binomial distribution, standard deviation: σ= √(npq)

Setting the deviation to 0,
0=√(npq)
0 = (npq)
0 = pq (we assume n = at least 1, since we have seen a success, aka a non-zero value).
q = 1- p (by definition)

0 = p(1-p)
0 = p - p2
p = 1 or 0 (aka 100% or 0% chance of success)

That is not a useful result, as we know monster saves fall between this.


As to your note above, I am not sure how you would take a median of 1 and make a calculation without some knowledge of p, I am not aware of any binomial distribution median formula that doesn't involve p in some way shape or form

Well, we know Variance is under 0.25 (it is maximal at p=0.5). I find using a Variance of 0.25 is a good second-order napkin math step.

Between 0.3 and 0.7 p, the variance is 0.21 to 0.25 (save on a 7+ to save on a 15+).

Then we end up taking the square root, which flattens it, and most of the cases we care about are "near" p=0.5, so the 0.25 variance estimate isn't far off.

 

[jgsugden]

We've whiteboarded this thing to death. Now it is time to see what it is actually like in play.

I have a PC I am currently running - a redux of another PC I have played in the past. This is a Glasya Tiefling archer with a religious bent. She is a 5th level Gloomstalker that took a level of Order Cleric and is now taking Sorcerer levels (Divine Soul). The next time she advances, she will be retraining Shield to Silvery Barbs. My expectation is that it will be a bit less exciting than most of you think - but I'll be recording the results.

 

[NotAYakk]

We've whiteboarded this thing to death. Now it is time to see what it is actually like in play.

I have a PC I am currently running - a redux of another PC I have played in the past. This is a Glasya Tiefling archer with a religious bent. She is a 5th level Gloomstalker that took a level of Order Cleric and is now taking Sorcerer levels (Divine Soul). The next time she advances, she will be retraining Shield to Silvery Barbs. My expectation is that it will be a bit less exciting than most of you think - but I'll be recording the results.

Silvery Barbs isn't a great replacement for Shield. Shield lasts until your next turn, and when you use it it is guaranteed to make an attack miss (barring further interrupts), and you can use it on 25% of incoming attacks (so pretty often), and follow up attacks become less likely to hit.

Silvery Barbs can be used on every hit, but only has a fair chance of making it miss.

The big deal of Silvery Barbs is on a PC with lots of low level slots (more slots than they can use in the adventuring day). It converts slots into power pretty efficiently and with good action economy.

To be strong offensively, you need excess low level slots (1), and lots of cases where enemies have a chance to make saves that are important that they fail (2).

Defensively, it beats shield on crits or when someone besides you is attacked. Against a hit it isn't as good as shield.

On someone who is a 3rd level spellcaster, unless you have extremely short days you don't have excess slots. The higher efficiency of shield will make it a better defensive spell. And offensively, the big advantage will rely on someone else in the party landing good save-or-sucks, as that PC will usually be better off using their action to attack, not debuff.

 

[ECMO3]

For a binomial distribution, standard deviation: σ= √(npq)

Setting the deviation to 0,
0=√(npq)
0 = (npq)
0 = pq (we assume n = at least 1, since we have seen a success, aka a non-zero value).
q = 1- p (by definition)

0 = p(1-p)
0 = p - p2
p = 1 or 0 (aka 100% or 0% chance of success)

That is not a useful result, as we know monster saves fall between this.


As to your note above, I am not sure how you would take a median of 1 and make a calculation without some knowledge of p, I am not aware of any binomial distribution median formula that doesn't involve p in some way shape or form

If you have 1 sample with which to estimate a population the sample is the estimated population median.

I did not say it was extremely useful, I took exception to assumption that the failure rate is 50% when your only sample is a success.

This spell is peculiar in that you must know a save suceeded to use it, this biases the entire expected result towards "success". If you know nothing else about the population this means the base assumption is you will succeed which in this case means the assumption is the spell will be useless (in terms of the reroll).

When I mentioned a target 11 on an earlier post on a completely different point, showing that this is statistically inferior to disadvantage. The 11 was an example, you could use any number from 2 to 20 and get the same result.

 

[Stalker0]

If you have 1 sample with which to estimate a population the sample is the estimated population median.

I did not say it was extremely useful, I took exception to assumption that the failure rate is 50% when your only sample is a success.

I guess what I am not following is... so if I assume success is the median.....well than what is the % chance of success then? Aka how does this assumption help me make calculations?

Also keep in mind, my greater point beyond my math was not to nail down EXACTLY how good silvery barbs is compared to portents...but to show there is a rough equivalency, aka both abilities are potent and superior to the other depending on the kind of adventuring day you are having. So in a nutshell, we have a spell that is giving all casters a form of power that has till now been reserved for divination wizards....and given divination wizards even more uses of that ability.

Since portents is often cited as one of the strongest wizard subclass powers....that's a strong case for a potential OP status.

 

[Stalker0]

If you know nothing else about the population

I think this is a flawed assumption. Based on the discussions with shield, the player should know what the save result was.

Further, while players may not have exact knowledge of a specific monster, most monsters have consistencies that give us some meta information. If a massive 20 foot golem passed a dexterity save by two over my DC, I can make a reasonable assessment that the golem passed based on low probability result and not on some massive bonus.

So I disagree with the premise that we have no implicit understanding of the probabilties, while a player may not do the math at the table, they do have a reasonable intuition on whether a new save is going to have a "good chance" of success or failure.

 

[Paul Farquhar]

Silvery Barbs can be used on every hit, but only has a fair chance of making it miss.

No, it uses your reaction, so if you get hit more than once in a round - as you do if you get mulitiattacked - it only works on the first one.

Whether or not the chance of turning it into a miss is fair, good, or poor depends on what roll was needed to hit in the first place. It's most useful if your AC is already high and only have to worry about "lucky hits". It's not going to stop an AC 15 wizard getting hammered, but it's great for an AC 20+ Bladesinger.