D&D 5E (2024) How useful/interesting is something like this?

[TigerDude]

yo need to explain what you're doing to make something interesting

 

[Deset Gled]

Personally, I would prefer to have access to the data/tables/formulas more than the graph. I like to see the actual numbers. Part of this is just the way I think/learn. Part of this is that I've become too used to seeing bad charts and infographics being used for bad arguments.

That being said, I also understand that the visual matters, and the chart can be a good summary. I still prefer it at the end rather than the beginning. But putting it up front and center will absolutely do more to get engagement. Such is the medium we are on.

 

[FrogReaver]

There were quite a few votes for 'what is this'. So maybe it's a good time to elaborate a bit on what it is.

Currently I pick some builds, compute the dpr and effective hp. I then plot those. What the graphical position of the points tells you is that points higher have better DPR's than those lower than them, and points further to the right have better eHp than those to the left of them. Practically what this means is that anything lower and to the left of a point is objectively worse for these 2 metrics. Obviously these 2 metrics alone aren't the full picture, but for most combat scenarios they likely have the biggest impact for numerically comparable features, which is why I focus on them.

Ideally the end state would allow filtering for high mobility or strong save tags applied to each build to compare builds with just those tags (these tags would be a bit more subjective due to all the different mobility and save style enhancements there are).

What is the Frontier line? That's the line going through the 2 outermost points. It's useful to show the optimal threshold. Essentially it shows what a linear tradeoff between damage and eHp would ideally look like based on the best damage and best eHp builds. Practically this means any point close to that line is a strong performer, any point far away is a weak performer, at least for those 2 important metrics.

Additionally this allow us to rank builds based on their closest distance to that optimal frontier line (that distance not displayed on the graph, but easily computable). That distance could then be used as a combined offensive and defensive score (or even better normalized into a value between 0 and 1 which is then used as the score). One of my major motivations for this was wanting to be able to better compare defensive builds to offensive builds to balanced builds and I think this will be a strong framework for that.

Also once builds are tagged, and all the numbers are computed, machine learning algorithms can be applied to 1) cluster builds together and 2) determine which features have the most impact on the final rating.

 

[ECMO3]

Not very useful for two reasons.

First, and most important, these are calculated values which come with a host of assumptions that will rarely be valid in play (for example how does the Rogue Bonus action hide factor into this as it both boosts his DPR and his effective hit points). A better method would be to perfomr statistical analysis of data collected in play. I recognize that data is probably not available, but an alternative if you think your assumptions are broadly applicable is to calculate the values then conduct an experiment (i.e. a campaign) to refute or substantiate the hypothesis. Pick 5 characters and play a campaign recording all damage rolled against them and how much they actually took, normalize this against their total hit points. Then record ow much damage they actually deal on each turn (after resistances are applied and not accounting for overkill). Then compare this over a campaign to the calculated values to see if they are accurate. If you have 5 PCs and they all finish close to where they should then your assumptions are reasonable. If one or more is way off then they are not.

Second it is DPR and hit point focused which equates to one part of one pillar of play.

 

[FrogReaver]

Not very useful for two reasons.

First, and most important, these are calculated values which come with a host of assumptions that will rarely be valid in play (for example how does the Rogue Bonus action hide factor into this as it both boosts his DPR and his effective hit points). A better method would be to perfomr statistical analysis of data collected in play. I recognize that data is probably not available, but an alternative if you think your assumptions are broadly applicable is to calculate the values then conduct an experiment (i.e. a campaign) to refute or substantiate the hypothesis. Pick 5 characters and play a campaign recording all damage rolled against them and how much they actually took, normalize this against their total hit points. Then record ow much damage they actually deal on each turn (after resistances are applied and not accounting for overkill). Then compare this over a campaign to the calculated values to see if they are accurate. If you have 5 PCs and they all finish close to where they should then your assumptions are reasonable. If one or more is way off then they are not.

First of all thanks for the explanation. Not many voting that way have commented on their vote.

My answers, in no particular order:
1. As you already acknowledged such data doesn't exist. I don't believe some theoretical but unachievable methodology can make any analysis not very useful.
2. I don't believe performing a statistical analysis of data collected in play would actually be better, at least without some robust methodology of ensuring the data was actually representative of the D&D playing population. Without such a methodology Data collected in play has similar issues of only being valid for tables that play like the majority of those you collected from and are placed in similar circumstances in their particular campaign. It gives the illusion of a more generalized truth than making assumptions and computing values without actually being one.
3. Running a single, or even small handful of experiments reveals nothing because of the variance inherent in D&D from dice rolls. You'd have to run enough experiments to at least be statistically significant. Usually in the best case that's 40, but can easily reach into hundreds, maybe 1,000s.

Second it is DPR and hit point focused which equates to one part of one pillar of play.

I never made a claim that this was for anything outside the combat pillar. I'm curious, what numerical values would you prefer to use to compare martial classes in the combat pillar? I can't think of any better 2, maybe you can?

 

[aco175]

Which is better?

Some might be based on what effective HP is and how to determine that. I'm guessing things like rage taking half damage or rogue's ability to half damage at level 5 comes into play or fighters second wind and how many times you short rest to get things back. I'm guessing this might be determined in some sort of optimized way.

Damage Per Round (DPR) might be better since the other variables are something players cannot control. They do not build the encounters or monsters or monster actions so taking HP damage might be more swingy. Again if you optimize things for combat only this makes more sense to me.

I do like how the chart shows damage and who can optimize it better.

 

[Kinematics]

First question is, what level is this supposed to be at?

Since there are a couple entries that approach 500 effective HP, I would by default assume we're at level 20. However the rogue is at about 70 HP, and even with 0 Con bonus and just average HP values per level, the rogue should be over 100. So that suggest this is not level 20. Except this is for effective HP, so maybe you're penalizing the character for something? I have no idea.

Then there's DPR. You have a max of 35 for a Zealot barbarian. I know for certain than 35 DPR is fairly low in the overall scheme of things, if taken to level 20. Barbarians and fighters can easily pass 80 DPR. 35 DPR might show up around levels 5-10, depending on build. The rogue with 12 DPR might be around level 3? Actually, several builds mention feats, so at least level 4.

The second question is about the "Efficient Frontier".

You've anchored that metric to arbitrary points that have limited correlation, and haven't established why those endpoints are meaningful, or whether that should be a linear formula at all.

For example, there can be a minimum number of effective HP below which it's meaningless to have a character of a given level. And the more HP you have, the less that additional HP adds to the character's utility. That suggests there should be a minimum asymptote, and probably a power curve.

The third issue is about the correlation between the Efficient Frontier and your character build results.

Since the EF line goes through two of the result points, it suggests you simply set the limit of the curve at those two points. For a meaningful "value" system, it would be more appropriate for the EF line and the data points to be independent. That is, establish the parameters for what you feel the EF line should be, and then compare the data to that, rather than basing the EF line on the data you calculated. In the latter case, all you're saying is that the EF line is the upper limit of what I've made, not that it has any meaning in and of itself.

 

[FrogReaver]

Since the EF line goes through two of the result points, it suggests you simply set the limit of the curve at those two points. For a meaningful "value" system, it would be more appropriate for the EF line and the data points to be independent. That is, establish the parameters for what you feel the EF line should be, and then compare the data to that, rather than basing the EF line on the data you calculated. In the latter case, all you're saying is that the EF line is the upper limit of what I've made, not that it has any meaning in and of itself.

This is actually not how efficient frontiers work at all. There is no what you feel it should be, it is a data driven line.

 

[FrogReaver]

First question is, what level is this supposed to be at?

Since there are a couple entries that approach 500 effective HP, I would by default assume we're at level 20. However the rogue is at about 70 HP, and even with 0 Con bonus and just average HP values per level, the rogue should be over 100. So that suggest this is not level 20. Except this is for effective HP, so maybe you're penalizing the character for something? I have no idea.

Level 5.

The current graph is just an example with non-finalized numbers. I did attempt to get accurate numbers for a first pass, but I didn't go back and proof check everything yet. That's why my question was whether something like this would be useful. The question is more about whether it's worth putting in the work.

Then there's DPR. You have a max of 35 for a Zealot barbarian. I know for certain than 35 DPR is fairly low in the overall scheme of things, if taken to level 20. Barbarians and fighters can easily pass 80 DPR. 35 DPR might show up around levels 5-10, depending on build. The rogue with 12 DPR might be around level 3? Actually, several builds mention feats, so at least level 4.

Yea, a finalized rouge using a rapier should probably be a bit higher. Looks like I left off advantage for that version. But again, my current focus is more on whether this view is even worth creating, not whether all the numbers in the proof of concept version are accurate.

For example, there can be a minimum number of effective HP below which it's meaningless to have a character of a given level. And the more HP you have, the less that additional HP adds to the character's utility. That suggests there should be a minimum asymptote, and probably a power curve.

If your saying that against a specific static challenge there's a point where adding more hp won't matter, i agree. But if you view the set of possible challenges as being unbounded then there is no threshold where more hp doesn't matter. IMO, for optimization comparisons you really need to assume unbounded challenges or every axis ultimately hits whatever arbitrary threshold you designed as the cutoff.