If you've played ADV for long enough, you've definitely been in a situation like this:
Zapdos hits Snorlax for 25%, leaving 26% HP left on Snorlax. You pray hard, but Zapdos does 26% the next turn and Snorlax is gone. You rationalize by saying "I got unlucky". But what if I told you that with just a tweak of your Snorlax EV spread, you could have sacrificed nothing, maybe boomed the next turn and won the interaction?
---
Snorlax EVs have always appeared random to just about anyone who has touched ADV. In this thread, I want to communicate that despite how strange they look, there are concrete ways to think about Snorlax EVs. At least, I will try to demystify what is subjective and what is not.
Recent posts in the Beginner's Lounge thread have highlighted a general lack of understanding of Snorlax EVs in the community, and I am party to that as well. However, with careful analysis, I think we can actually solve the problems of (1) understanding what can/cannot be optimized, (2) tailoring spreads for specific benchmarks, and (3) what these spreads generally look like.
I propose that
(A) Benchmarks are mostly what make Snorlax EVs subjective, but after they are determined
(B) Optimizing spreads is mostly an objective exercise.
(A) Benchmarks
These are clearest where it comes to Curselax, but less so offensive lax. For example, for Curselax, possible benchmarks are
(Atk) KOing Dugtrio with uninvested normal move
(Atk) 3HKOing Suicune under sand +1, or with a crit
(SpDef) 5HKO from Zapdos (to avoid 2 hits during Rest + 1 crit)
etc.
I do not want to dwell on this subjective part, but I want to put forth a general principle:
[A1] Probabilistic (rather than definite) benchmarks for Rest loops are most efficient
When it comes to benchmarking for Rest loops, it is much more efficient to EV for a small chance for a KO than definite survival. The reason is that variance is reduced over multiple hits. Getting max-rolled consecutively becomes exponentially unlikely with the number of hits. Take for example
252 SpA 30 IVs Zapdos Thunderbolt vs.
252 HP / 152+ SpD Snorlax: 110-130 (20.9 - 24.8%) -- guaranteed 5HKO after sandstorm damage and Leftovers recovery
252 HP / 140+ SpD Snorlax: 112-132 (21.3 - 25.1%) -- 0.1% chance to 4HKO after sandstorm damage and Leftovers recovery
252 HP / 128+ SpD Snorlax: 113-133 (21.5 - 25.3%) -- 0.2% chance to 4HKO after sandstorm damage and Leftovers recovery
252 HP / 116+ SpD Snorlax: 114-135 (21.7 - 25.7%) -- 1.7% chance to 4HKO after sandstorm damage and Leftovers recovery
252 HP / 100+ SpD Snorlax: 115-136 (21.9 - 25.9%) -- 3.9% chance to 4HKO after sandstorm damage and Leftovers recovery
With 100 SpD EVs, one already gets to a 96% chance of survival if Zapdos crits Snorlax during the rest loop, a pretty good number. But to fully survive, one has to invest a whopping 52 more EVs.
Now let's assume that you can easily find a spread that can satisfy at least one benchmark desired, that you can use in part B.
(B) Optimization
Some prelude:
It has been suggested that it is currently trendy not to invest HP in Snorlax. This is seen in many teams in the team sharing thread (Exhibit 1 & team) (Exhibit 2 & team) (Exhibit 3 & team).
I want to propose that having minimal HP is suboptimal. Now, I acknowledge that I may have overlooked something in my proceeding arguments and I may be mistaken in part. The point is not to call out players for their lack of understanding, but to highlight that even experienced players who have been thoroughly immersed in ADV can still overlook some small details, hopefully learn something from this thread, and turn the game around with a lax surviving at 1%. If there are reasons having low HP is better, I will be glad to hear them.
[B1] More HP is always better, given any two spreads that provide the same physical and special bulk.
Benefits: Switching Snorlax into Blissey's Seismic Toss is a very common interaction. Side points: higher damage on Counter, lower recoil on Double-Edge.
Costs: More Leech Seed healing from opponent. Less healing from same side Celebi.
For Curselax, the point is to last long enough to set up all Curses and OHKO everything, so trying to win any momentary damage trade through Leech Seed is not worth it. For offensive lax, switching into Blissey to take Seismic Tosses is usually more relevant than the small gains Celebi gets from increased HP. Every 4 HP EV counts towards reduced % Seismic Toss damage, but Leech Seed (12.5% HP transferred) only provides extra healing every 32 HP EVs. Just to get it out of the way, Pain Split is essentially never a point of contention as their use on even Misdreavus/Weezing/Dusclops is rare and insignificant.
[B2] An optimization based on "effective bulk" can always be performed. Unless (rarely) ignoring one defensive stat completely, the optimization will place EVs in at least two of the three defensive stats,
but almost never Def and SpD without HP.
Optimization Example 1: CurseLax Spread
Let us go back to the spread posted by the curious beginner in the Beginner's Lounge thread,
The damage formula suggests that the "effective" physical / special bulk is just HP*Def / HP*SpD (the stats). This means that as long as two spreads have the same HP*Def, they would receive roughly the same physical damage.
Optimization Steps
Notation: EVs assumed to be divided by 4 already
(1) Set the Atk benchmark. The remaining EVs such that Total EVs = HP EVs + Def EVs + SpD EVs. Then you can express the "effective" Def and SpD as such:
(2) Determine the desired benchmark on one defensive stat based on a random spread that works that you have found with the damage calc. It could be that through the damage calc, you have found some combination of defense and HP lets you survive say 3 Dugtrio EQs after 1 Curse while you're resting. So let's say it is the physical defense benchmark. Solve for the Def EVs required to hit that given benchmark Def0 = HP*Def. This is a quadratic equation. Choose the root that gives non-negative values of HP/Def/SpD.
(3) Express the other effective defense (SpD) in the Def EVs.
(4) Plot the other effective defense (SpD), inserting the Def benchmarks and total defensive EVs. Find the point of max SpD.
In this case, you can see that the effective SpD is maximized when SpD EVs are 0. However, this is not achievable because the HP would exceed 252 EVs. Still, it is important to know the trend that the smaller the number of SpD EVs, the higher the special bulk.
(5) Plot the HP EVs to find the optimal HP
To stay within the bounds of HP EVs < 252/4 = 63, we need about 19*4 = 76 SpD EVs
(6) Plot the Def EVs to check for sanity.
With 72 SpD EVs we should get 84 Def EVs.
The final spread is 252 HP / 96 Atk / 84 Def / 76 SpD (Careful).
Now compare the physical and special damage that Snorlax receives, say from Dugtrio and Zapdos:
Original
Optimized
252 Atk 30 IVs Choice Band Dugtrio Earthquake vs.
12 HP / 184 Def Snorlax: 197-232 (42.4 - 50%) -- 0.4% chance to 2HKO after sandstorm damage and Leftovers recovery
252 HP / 84 Def Snorlax: 222-262 (42.3 - 50%) -- 0.4% chance to 2HKO after sandstorm damage and Leftovers recovery
252 SpA 30 IVs Zapdos Thunderbolt vs.
12 HP / 216+ SpD Snorlax: 105-124 (22.6 - 26.7%) -- 29.4% chance to 4HKO after sandstorm damage and Leftovers recovery
252 HP / 76+ SpD Snorlax: 118-139 (22.5 - 26.5%) -- 20.4% chance to 4HKO after sandstorm damage and Leftovers recovery
You get more HP for similar/slightly more bulk after optimization!
Optimization Example 2: Offensive Snorlax Spread
Consider Exhibit 1 & team using the spread
We get an optimized spread of 228 HP / 252 Atk / 28 SpD (Adamant), but the calcs show a slight imbalance in the Def and SpDef damage. I suspect that these discrepancies are caused by flooring and the small additive part in the damage formula.
Anyway, we can retune the EVs by hand to provide more Def. One possible retuning would be 156 HP / 252 Atk / 52 Def / 48 SpD (There are HP values closer to 228 that work too but I wanted to show a spread that is convincingly better in both Def/SpD).
Original
Raw Optimal
Retuned
252 SpA 30 IVs Zapdos Thunderbolt vs.
0 HP / 144 SpD Snorlax: 123-145 (26.6 - 31.4%) -- 24.1% chance to 4HKO after Leftovers recovery
228 HP / 28 SpD Snorlax: 136-160 (26.2 - 30.8%) -- 6.1% chance to 4HKO after Leftovers recovery
156 HP / 48 SpD Snorlax: 133-157 (26.6 - 31.4%) -- 18.7% chance to 4HKO after Leftovers recovery
252 Atk 30 IVs Choice Band Dugtrio Earthquake vs.
0 HP / 112 Def Snorlax: 215-253 (46.6 - 54.8%) -- 10.9% chance to 2HKO after Leftovers recovery
228 HP / 0 Def Snorlax: 250-295 (48.2 - 56.9%) -- 41% chance to 2HKO after Leftovers recovery
156 HP / 52 Def Snorlax: 232-274 (46.4 - 54.8%) -- 10.5% chance to 2HKO after Leftovers recovery
TLDR: Compare Red and Green.
Again, we get more HP for similar/slightly more bulk after optimization!
[B3] Observation: optimized spreads tend to have a significant amount of HP EVs.
Following B1 and the examples in B2, we see that optimized spreads have a significant amount of HP EVs. This becomes increasingly true the more defensive a Snorlax spread is
Caveats
The graphs show that effective bulk of one stat tends to increase by about 3-5% if the other stat is kept constant after optimization from a spread without HP to one with HP, but it is unclear to me why the improvement is not as manifest in the final calculations. Perhaps the flooring and additive factor in the damage formula matters, but it is not easy to model without a brute force calculation.
Closing
I hope I've put forth some convincing arguments. I'm attaching a Mathematica notebook (if you're a college student, there's a good chance your college has a subscription). It's not necessary though - the plotting can be done in any numerical coding package. Happy to get feedback, whether on the benchmarking / optimizing part of this post!
Zapdos hits Snorlax for 25%, leaving 26% HP left on Snorlax. You pray hard, but Zapdos does 26% the next turn and Snorlax is gone. You rationalize by saying "I got unlucky". But what if I told you that with just a tweak of your Snorlax EV spread, you could have sacrificed nothing, maybe boomed the next turn and won the interaction?
---
Snorlax EVs have always appeared random to just about anyone who has touched ADV. In this thread, I want to communicate that despite how strange they look, there are concrete ways to think about Snorlax EVs. At least, I will try to demystify what is subjective and what is not.
Recent posts in the Beginner's Lounge thread have highlighted a general lack of understanding of Snorlax EVs in the community, and I am party to that as well. However, with careful analysis, I think we can actually solve the problems of (1) understanding what can/cannot be optimized, (2) tailoring spreads for specific benchmarks, and (3) what these spreads generally look like.
I propose that
(A) Benchmarks are mostly what make Snorlax EVs subjective, but after they are determined
(B) Optimizing spreads is mostly an objective exercise.
(A) Benchmarks
These are clearest where it comes to Curselax, but less so offensive lax. For example, for Curselax, possible benchmarks are
(Atk) KOing Dugtrio with uninvested normal move
(Atk) 3HKOing Suicune under sand +1, or with a crit
(SpDef) 5HKO from Zapdos (to avoid 2 hits during Rest + 1 crit)
etc.
I do not want to dwell on this subjective part, but I want to put forth a general principle:
[A1] Probabilistic (rather than definite) benchmarks for Rest loops are most efficient
When it comes to benchmarking for Rest loops, it is much more efficient to EV for a small chance for a KO than definite survival. The reason is that variance is reduced over multiple hits. Getting max-rolled consecutively becomes exponentially unlikely with the number of hits. Take for example
252 SpA 30 IVs Zapdos Thunderbolt vs.
252 HP / 152+ SpD Snorlax: 110-130 (20.9 - 24.8%) -- guaranteed 5HKO after sandstorm damage and Leftovers recovery
252 HP / 140+ SpD Snorlax: 112-132 (21.3 - 25.1%) -- 0.1% chance to 4HKO after sandstorm damage and Leftovers recovery
252 HP / 128+ SpD Snorlax: 113-133 (21.5 - 25.3%) -- 0.2% chance to 4HKO after sandstorm damage and Leftovers recovery
252 HP / 116+ SpD Snorlax: 114-135 (21.7 - 25.7%) -- 1.7% chance to 4HKO after sandstorm damage and Leftovers recovery
252 HP / 100+ SpD Snorlax: 115-136 (21.9 - 25.9%) -- 3.9% chance to 4HKO after sandstorm damage and Leftovers recovery
With 100 SpD EVs, one already gets to a 96% chance of survival if Zapdos crits Snorlax during the rest loop, a pretty good number. But to fully survive, one has to invest a whopping 52 more EVs.
Now let's assume that you can easily find a spread that can satisfy at least one benchmark desired, that you can use in part B.
(B) Optimization
Some prelude:
It has been suggested that it is currently trendy not to invest HP in Snorlax. This is seen in many teams in the team sharing thread (Exhibit 1 & team) (Exhibit 2 & team) (Exhibit 3 & team).
I want to propose that having minimal HP is suboptimal. Now, I acknowledge that I may have overlooked something in my proceeding arguments and I may be mistaken in part. The point is not to call out players for their lack of understanding, but to highlight that even experienced players who have been thoroughly immersed in ADV can still overlook some small details, hopefully learn something from this thread, and turn the game around with a lax surviving at 1%. If there are reasons having low HP is better, I will be glad to hear them.
[B1] More HP is always better, given any two spreads that provide the same physical and special bulk.
Benefits: Switching Snorlax into Blissey's Seismic Toss is a very common interaction. Side points: higher damage on Counter, lower recoil on Double-Edge.
Costs: More Leech Seed healing from opponent. Less healing from same side Celebi.
For Curselax, the point is to last long enough to set up all Curses and OHKO everything, so trying to win any momentary damage trade through Leech Seed is not worth it. For offensive lax, switching into Blissey to take Seismic Tosses is usually more relevant than the small gains Celebi gets from increased HP. Every 4 HP EV counts towards reduced % Seismic Toss damage, but Leech Seed (12.5% HP transferred) only provides extra healing every 32 HP EVs. Just to get it out of the way, Pain Split is essentially never a point of contention as their use on even Misdreavus/Weezing/Dusclops is rare and insignificant.
[B2] An optimization based on "effective bulk" can always be performed. Unless (rarely) ignoring one defensive stat completely, the optimization will place EVs in at least two of the three defensive stats,
but almost never Def and SpD without HP.
Optimization Example 1: CurseLax Spread
Let us go back to the spread posted by the curious beginner in the Beginner's Lounge thread,
- 12 HP
- 96 Atk
- 184 Def
- 216+ SpD
The damage formula suggests that the "effective" physical / special bulk is just HP*Def / HP*SpD (the stats). This means that as long as two spreads have the same HP*Def, they would receive roughly the same physical damage.
Optimization Steps
Notation: EVs assumed to be divided by 4 already
(1) Set the Atk benchmark. The remaining EVs such that Total EVs = HP EVs + Def EVs + SpD EVs. Then you can express the "effective" Def and SpD as such:
(2) Determine the desired benchmark on one defensive stat based on a random spread that works that you have found with the damage calc. It could be that through the damage calc, you have found some combination of defense and HP lets you survive say 3 Dugtrio EQs after 1 Curse while you're resting. So let's say it is the physical defense benchmark. Solve for the Def EVs required to hit that given benchmark Def0 = HP*Def. This is a quadratic equation. Choose the root that gives non-negative values of HP/Def/SpD.
In this case, you can see that the effective SpD is maximized when SpD EVs are 0. However, this is not achievable because the HP would exceed 252 EVs. Still, it is important to know the trend that the smaller the number of SpD EVs, the higher the special bulk.
(5) Plot the HP EVs to find the optimal HP
To stay within the bounds of HP EVs < 252/4 = 63, we need about 19*4 = 76 SpD EVs
(6) Plot the Def EVs to check for sanity.
The final spread is 252 HP / 96 Atk / 84 Def / 76 SpD (Careful).
Now compare the physical and special damage that Snorlax receives, say from Dugtrio and Zapdos:
Original
Optimized
252 Atk 30 IVs Choice Band Dugtrio Earthquake vs.
12 HP / 184 Def Snorlax: 197-232 (42.4 - 50%) -- 0.4% chance to 2HKO after sandstorm damage and Leftovers recovery
252 HP / 84 Def Snorlax: 222-262 (42.3 - 50%) -- 0.4% chance to 2HKO after sandstorm damage and Leftovers recovery
252 SpA 30 IVs Zapdos Thunderbolt vs.
12 HP / 216+ SpD Snorlax: 105-124 (22.6 - 26.7%) -- 29.4% chance to 4HKO after sandstorm damage and Leftovers recovery
252 HP / 76+ SpD Snorlax: 118-139 (22.5 - 26.5%) -- 20.4% chance to 4HKO after sandstorm damage and Leftovers recovery
You get more HP for similar/slightly more bulk after optimization!
Optimization Example 2: Offensive Snorlax Spread
Consider Exhibit 1 & team using the spread
- 0 HP
- 252+ Atk
- 112 Def
- 144 SpD
We get an optimized spread of 228 HP / 252 Atk / 28 SpD (Adamant), but the calcs show a slight imbalance in the Def and SpDef damage. I suspect that these discrepancies are caused by flooring and the small additive part in the damage formula.
Anyway, we can retune the EVs by hand to provide more Def. One possible retuning would be 156 HP / 252 Atk / 52 Def / 48 SpD (There are HP values closer to 228 that work too but I wanted to show a spread that is convincingly better in both Def/SpD).
Original
Raw Optimal
Retuned
252 SpA 30 IVs Zapdos Thunderbolt vs.
0 HP / 144 SpD Snorlax: 123-145 (26.6 - 31.4%) -- 24.1% chance to 4HKO after Leftovers recovery
228 HP / 28 SpD Snorlax: 136-160 (26.2 - 30.8%) -- 6.1% chance to 4HKO after Leftovers recovery
156 HP / 48 SpD Snorlax: 133-157 (26.6 - 31.4%) -- 18.7% chance to 4HKO after Leftovers recovery
252 Atk 30 IVs Choice Band Dugtrio Earthquake vs.
0 HP / 112 Def Snorlax: 215-253 (46.6 - 54.8%) -- 10.9% chance to 2HKO after Leftovers recovery
228 HP / 0 Def Snorlax: 250-295 (48.2 - 56.9%) -- 41% chance to 2HKO after Leftovers recovery
156 HP / 52 Def Snorlax: 232-274 (46.4 - 54.8%) -- 10.5% chance to 2HKO after Leftovers recovery
TLDR: Compare Red and Green.
Again, we get more HP for similar/slightly more bulk after optimization!
[B3] Observation: optimized spreads tend to have a significant amount of HP EVs.
Following B1 and the examples in B2, we see that optimized spreads have a significant amount of HP EVs. This becomes increasingly true the more defensive a Snorlax spread is
One metric would be to consider the average fractional change in bulk induced by adding 4 EVs in each defensive stat compared to 8 EVs in HP.
Starting from the base stats, it is true that the average fractional change would be higher by just investing in each defensive stat. However, Snorlax would likely not have much more than 252 EVs in offensive stats, in which case there likely comes a point after adding enough Def/SpD EVs there is more marginal benefit from placing 8 EVs in HP.
Starting from the base stats, it is true that the average fractional change would be higher by just investing in each defensive stat. However, Snorlax would likely not have much more than 252 EVs in offensive stats, in which case there likely comes a point after adding enough Def/SpD EVs there is more marginal benefit from placing 8 EVs in HP.
Caveats
The graphs show that effective bulk of one stat tends to increase by about 3-5% if the other stat is kept constant after optimization from a spread without HP to one with HP, but it is unclear to me why the improvement is not as manifest in the final calculations. Perhaps the flooring and additive factor in the damage formula matters, but it is not easy to model without a brute force calculation.
Closing
I hope I've put forth some convincing arguments. I'm attaching a Mathematica notebook (if you're a college student, there's a good chance your college has a subscription). It's not necessary though - the plotting can be done in any numerical coding package. Happy to get feedback, whether on the benchmarking / optimizing part of this post!
Last edited: