Troll420'sDraft Decoded - A Mathematical Analysis of LCIMathematically Evaluating Card Performance by Draft Position
Now that Lost Caverns of Ixalan (LCI) has been drafted, I’d like to look how individual cards are performing. I created a model to evaluate performance. The basic idea is to get a numerical analysis of what cards are good and which are bad. We’ll call the numerical score Draft Value (DV). The model is based on data from 17Lands.
1.0 Modelling Card Performance
1.1: Background
Game In Hand Win Rate (GIH WR) was invented by 17lands and is the most used metric for card evaluation. While GIH WR is an excellent starting point it doesn't tell the full story.
Its main limitation is not accounting for draft position. Early picks have a higher opportunity cost than late picks because you have to pass on other good cards early in the pack. For example, in Midnight Hunt Neonate's Rush had a higher GIH WR (56.4%) than Moonrager's Slash (55.5%). However, Slash was taken 4th while Rush was taken 10th. GIH WR indicates that Rush is the “better” card, but in context Slash is clearly better.
In every set some cards that are taken late have a high GIH WR. Others are drafted early and have a low GIH WR. GIH WR cannot tell determine which card is intrinsically better. Would those late gems still perform as early picks? Taken lower down, would those poor performers put up good results?
Another limitation is synergy/role assignment. Every set has unique mechanics with payoff cards + enablers. Payoff cards can put up a high WR with a low draft position when taken for a deck that already has enablers. Alternatively, some cards are deck dependent (e.g. combat tricks in aggro) and can have a high WR because players only take them when they fit the deck. In such cases, a high GIH WR doesn't necessarily mean the card should be a high pick.
The underlying idea of this model is to see performance based on draft position. In the past, people have attempted to adjust GIH WR by draft position. The main difference in this model's approach is the weighting. Some mathematical operations are performed to make Draft Value depend more heavily on GIH WR than ATA. In other words, instead of an equal balance between win rate and draft position, Draft Value is primarily win rate with an adjustment for position.
Before getting into the model, I’d like to establish a few things:
1) You can skip straight to section 2.0 to avoid math (math is for blockers). I recommend getting a high level understanding of sections 1.1 to 1.3. The math gets more intricate in 1.4.
2) My background is in physics not statistics. Don’t expect Frank Karsten levels of accuracy. If you are confident with statistics I’d love to hear your feedback.
1.2: Variables
17 lands is a fantastic resource. The data can be filtered by player skill (top/middle/bottom). You could make an argument to filter for top players only. However, this model is intended for all users, so I’ll be including all data points. We’ll be using two key variables: Games in Hand Win Rate (GIH WR) and Average Taken At (ATA).
GIH WR gives an idea of how cards perform as covered above. Generally a card with a 60% win rate would be a higher pick then a 58%. However, if the 60% card is a 1-of (curve topper) and the 58% is something run in multiples (cheap removal) the pick order changes. Fortunately, GIH WR already adjusts for number of copies drawn.
Average Taken At (ATA) is a way to adjust for pick order. A model that consumes ATA can then adjust out the primary limitation of GIH WR. Basically want to reward cards that are drafted highly and punish cards that are taken late.
1.3: Rough Equation
The rough idea is pretty simple (Formula 1):
A higher pick should correspond with a higher Draft Value. Note that a “higher pick” means a smaller ATA: although 15 is a bigger number than 1, 15th is a lower pick than 1st. Dividing by ATA accomplishes the goal. A larger denominator yields a smaller fraction. A card taken 4th would score higher than a card taken 12th for the same reason that a quarter of a pizza is bigger than a twelfth of a pizza.
Similarly, a higher GIH WR should correspond with a higher Draft Value. Using it as the numerator works. A bigger pizza gives bigger slices.
One concern is over drafted cards being overrated by DV (lower ATA -> higher DV). However, over drafted cards usually have a worse GIH WR as a consequence of being taken over superior cards. Can assume over drafted cards have their elevated ATA offset by a lower GIH WR and proceed with the model.
1.4: Adjustments
Formula 1 isn’t a fair operation. It’s essentially dividing apples by oranges. GIH WR is a percentage, ATA is a rank from 1-15 (e.g. draft order). We can illustrate the issue with an arbitrary example:
1.4.1 Example of Issues
Imagine two cards with identical win rates are drafted 10th and 15th. In absolute terms, ATA increases by 5 units (15 - 10 = 5). However, in relative terms this is a 50% increase (15/10 - 1 = 0.5).
Now supposed two cards with identical ATA have GIH WRs of 50% and 55%. In absolute terms, GIH WR also increases by 5 units (55 - 50 = 5). In relative terms, we only have a 10% change (55/50 - 1 = 0.1).
The relative change is what drives DV. Therefore, in this example DV is more sensitive to the change in ATA than in GIH WR. This doesn’t hold up to scrutiny. The difference between being drafted at 10th and 15th is minimal – it already wheeled. However, a 55% win rate is massively different than 50%. This is basically the gap between draft chaff and a playable card.
1.4.2 Normalizing GIH WR
We need an apples-to-apples comparison. This can be accomplished by normalizing GIH WR (we’ll denote this by putting a fancy hat over GIH WR in the formula). Instead of win rate out of 100%, I want to rank cards between the best performing and worst performing. That is, take the card with the highest GIH WR as the upper bound (we’ll call this HW for Highest Win Rate) and the card with the lowest as the lower bound (similarly LW). Can then normalize GIH WR between LW and HW (Formula 2):
In historical sets the best card has a GIH WR of ~70% and the worst has ~40% (HW = 70, LW = 40). We can use these bounds as a good approximation in the model for future sets as well. This formula won’t work for cards that break the bounds. However, you don’t need a model to tell you that a card with a GIH WR > 70% is good and one with < 40% sucks so this isn’t an issue in practice. You could always manually determine unique bounds set-by-set.
Let’s revisit the 50% vs 55% example. On a scale of 40 to 70, 50 converts to 33% while 55 converts to 50% (just plug the numbers into Formula 2). This is a relative difference of about 50%, which is a more accurate description of the gap between a 55% and 50% GIH WR. That is, normalizing GIH WR instead of using the raw values makes the model more sensitive to changes in GIH WR.
1.4.3 Adjusting ATA
I also want to clean up ATA before inputting it into the model.
Technically could also rescale ATA from the highest picked card to the lowest picked card. However, since ATA typically spans from 1 point something to 13 point something the implied 1-15 scale is fine.
In 1.4.1 we saw how an identical absolute change can yield very different relative changes depending on the starting point. Basically, adding 1 unit to a small number matters more than adding 1 unit to a big number. ATA only spans from 1-15 which are all very small numbers. This means the model is very sensitive to a change in draft position. For instance, the difference between 3rd and 4th is 33% (4/3 - 1).
My first thought was to take the square root of ATA to smooth out the deltas. Note that 3*3 = 9 but also -3*-3 = 9 so the square root of 9 is actually +/-3. Negative roots rarely have practical meanings (e.g. a negative draft value) so only consider positive roots. Going back to the 3rd and 4th pick example would now give sqrt(4)/sqrt(3) - 1 =15%
As mentioned in the background, we want the model to be weighted more heavily towards GIH WR than ATA. ATA is driven by perception while GIH WR is based on reality. A card may be drafted higher because people think it is better than it actually is. Shrinking ATA via the square root means GIH WR dominates final DV as desired.
After back testing using Midnight Hunt (see Appendix A) I found the cubed root to be a better fit than the square root. The basic idea was ensuring that GIH WR was more important than ATA. The original square root method resulted in the DV rank being closer to the ATA rank than the GIH WR rank. In our example we’d now have cbrt(4)/cbrt(3) - 1 =10%
The table below evaluates the difference in value between consecutive picks. For example, the relative difference between a 1st and 2nd pick is 100%. The model instead considers the 1st pick to be 26% more valuable than the 2nd pick. Qualitatively, I actually think the square root better describes the relative change in value between picks. However, back testing told me to use the cubed root and I’m not good enough at statistics to argue against it.
Draft Position | Raw | Square Root | Cubed Root |
1 | 100% | 41% | 26% |
2 | 50% | 22% | 14% |
3 | 33% | 15% | 10% |
4 | 25% | 12% | 8% |
5 | 20% | 10% | 6% |
6 | 17% | 8% | 5% |
7 | 14% | 7% | 5% |
8 | 13% | 6% | 4% |
9 | 11% | 5% | 4% |
10 | 10% | 5% | 3% |
11 | 9% | 4% | 3% |
12 | 8% | 4% | 3% |
13 | 8% | 4% | 3% |
14 | 7% | 4% | 2% |
15 |
|
|
|
This gives Formula 3:
1.4.4 Final Equation
Putting everything together, we get
We can restate this (Formula 4):
You could go even further. Instead of comparing the raw Draft Value you could get the distribution of DV for all cards. Using the distribution, you could get the z-score for each card. Comparing z-scores is fairer than raw values. However, DV isn’t normally distributed so I think it’s easier to use the raw scores.
This model has several limitations:
1) Minimal back testing (see Appendix A). Assumed cubed root was best fit for ATA. With more back testing, you could fit a better function
2) Assumed GIH WR was the best metric for raw value. Could make arguments for GP WR, GD WR, GNS WR, IWD. Similarly could have used ALSA instead of ATA
3) Ignores deck context. For instance, my set review was focused on RDW. This model gets the overall Draft Value of a card (regardless of deck archetype). A card with a high DV might not actually be a good aggro card and vice versa
4) Minimal predictive value. In a set with a bunch of small creatures, Shock would score very highly. However, this cannot be used to predict Shock’s performance if it was reprinted in a future set. For example, if the new set had mostly big creatures Shock would be a low pick. This doesn’t mean the model is wrong, just that different sets have different metas
5) Arguably punishes cards for being under drafted. A card with a high GIH WR may still get a low DV due a poor ATA. However, in such instances the high GIH WR is often due to the low ATA. Even when this is not the case rather than being a limitation it may be a way to identify undervalued cards (high GIH WR with a low DV). Converse applies to over drafted cards
6) GIH WR and ATA are not independent. This falls under the opportunity cost discussed earlier. Perhaps a more advanced model could adjust for correlations
7) “Rare drafting” can skew ATA. For example, players taking a rare early just to get 20 gems. DV is best when comparing rares-to-rares and commons-to-commons
8) Although the top and bottom cards are at 40% and 70%, most cards lie between 45% and 65%. Maybe the GIH WR normalization should use percentiles/z-scores
Again, the main purpose of the model is to help understand high level card performance. Feeding the results into an AI would probably accomplish nothing. A human may be able to use the results to draw qualitative conclusions. It’s a tool to help players make decisions. DV should not be used as the sole factor in determining a pick.
2.0 Modelling Early LCI Red Performance
For today, I'll be using Draft Value to take a look at early LCI performance in Red. I'll be highlighting some cards that are difficult to analyze going purely by GIH WR. I'll only be looking at commons and uncommons since we don't have enough of a sample for the rares. If people like the model, I can run it on the other colors as well.
Name | GIH WR | ATA | GIH WR Adj | ATA Adj | DV | Rank WR | Rank ATA | Rank DV |
Geological Appraiser | 61.2% | 3.43 | 0.7 | 1.5 | 46.9 | 1 | 2 | 1 |
Abrade | 58.4% | 3.42 | 0.6 | 1.5 | 40.7 | 5 | 1 | 2 |
Belligerent Yearling | 58.6% | 3.55 | 0.6 | 1.5 | 40.6 | 3 | 4 | 3 |
Scytheclaw Raptor | 57.9% | 3.46 | 0.6 | 1.5 | 39.4 | 9 | 3 | 4 |
Triumphant Chomp | 57.2% | 3.97 | 0.6 | 1.6 | 36.2 | 13 | 5 | 5 |
Dreadmaw's Ire | 58.0% | 4.88 | 0.6 | 1.7 | 35.4 | 8 | 7 | 6 |
Diamond Pick-Axe | 58.1% | 5 | 0.6 | 1.7 | 35.3 | 7 | 8 | 7 |
Enterprising Scallywag | 56.3% | 4.72 | 0.5 | 1.7 | 32.4 | 18 | 6 | 8 |
Plundering Pirate | 57.7% | 6.46 | 0.6 | 1.9 | 31.7 | 10 | 11 | 9 |
Saheeli's Lattice | 57.3% | 6.11 | 0.6 | 1.8 | 31.5 | 12 | 9 | 10 |
Volatile Wanderglyph | 58.5% | 7.72 | 0.6 | 2.0 | 31.2 | 4 | 16 | 11 |
Goblin Tomb Raider | 58.8% | 8.2 | 0.6 | 2.0 | 31.1 | 2 | 21 | 12 |
Etali's Favor | 58.4% | 9.12 | 0.6 | 2.1 | 29.4 | 5 | 24 | 13 |
Sunfire Torch | 57.2% | 8.06 | 0.6 | 2.0 | 28.6 | 13 | 19 | 14 |
Dinotomaton | 56.4% | 7.7 | 0.5 | 2.0 | 27.7 | 16 | 15 | 15 |
Sunshot Militia | 56.8% | 8.39 | 0.6 | 2.0 | 27.6 | 15 | 22 | 16 |
Idol of the Deep King | 55.5% | 7.2 | 0.5 | 1.9 | 26.8 | 19 | 13 | 17 |
Rumbling Rockslide | 55.3% | 6.93 | 0.5 | 1.9 | 26.8 | 20 | 12 | 18 |
Ancestors' Aid | 57.5% | 10.56 | 0.6 | 2.2 | 26.6 | 11 | 29 | 19 |
Panicked Altisaur | 56.4% | 8.87 | 0.5 | 2.1 | 26.4 | 16 | 23 | 20 |
Dowsing Device | 55.2% | 7.79 | 0.5 | 2.0 | 25.6 | 21 | 17 | 21 |
Rampaging Ceratops | 53.9% | 6.21 | 0.5 | 1.8 | 25.2 | 24 | 10 | 22 |
Seismic Monstrosaur | 55.1% | 7.99 | 0.5 | 2.0 | 25.2 | 22 | 18 | 23 |
Burning Sun Cavalry | 54.9% | 9.34 | 0.5 | 2.1 | 23.6 | 23 | 25 | 24 |
Goldfury Strider | 53.9% | 8.12 | 0.5 | 2.0 | 23.1 | 24 | 20 | 25 |
Curator of Sun's Creation | 52.6% | 7.63 | 0.4 | 2.0 | 21.3 | 28 | 14 | 26 |
Brazen Blademaster | 53.5% | 9.56 | 0.5 | 2.1 | 21.2 | 26 | 27 | 27 |
Calamitous Cave-In | 52.5% | 9.38 | 0.4 | 2.1 | 19.8 | 30 | 26 | 28 |
Daring Discovery | 53.3% | 12.05 | 0.4 | 2.3 | 19.3 | 27 | 32 | 29 |
Hotfoot Gnome | 52.6% | 10.44 | 0.4 | 2.2 | 19.2 | 28 | 28 | 30 |
Child of the Volcano | 49.5% | 10.68 | 0.3 | 2.2 | 14.4 | 31 | 30 | 31 |
2.1 Top Performing Red Commons/Uncommons
When the set was spoiled, everyone was bullish on Abrade, Geological Appraiser, and Belligerent Yearling. DV ranks these as the top 3 commons/uncommons in Red. Players are taking them highly and winning with them. Scytheclaw Raptor, Triumphant Chomp, and Dreadmaw's Ire aren't far behind.
It's hard to draw any meaningful insights here. These cards all looked good during spoilers. The raw GIH WR ranks them highly and drafters are taking them highly. The model loves them just as much as players. Good cards being good should come as no surprise. Drafters should keep doing what they're doing here.
2.2 Under Drafted Cards
2.2.1 Goblin Tomb Raider
By GIH WR this little guy is a B-level common (2nd). However, he has a very low ATA (21st). Are his results due to him not being drafted ahead of anything or is Goblin Guide (at home) a top priority?
His Draft Value of 31.1 is respectable, ranking 12th in Red. Although his GIH WR of 58%+ is somewhat due to his low ATA, 31.1 DV is really nice for a common (40+ is almost exclusively Rares and Mythics).
A 58% win rate isn't sustainable as a higher pick. The model likes the underlying context enough to still take him higher than most commons. I wouldn't fall into the trap of taking him in the first 3 picks. That said, if you are in Red aggro you need a good reason to pass this dude.
Based on the model, I would draft this as a C+ while it's currently being treated as a C-
2.2.2 Volatile Wanderglyph
This 2-drop has the 4th best GIH WR despite being drafted the 16th highest. Again, we have some tension. He's being drafted as a C- while GIH WR rates him as a B.
DV ranks him at a respectable 11th. Note that the 9-12 range in DV is very tight (31.X). The conclusion is similar to Goblin Tomb Raider. This card should be drafted higher than it's being taken, just not quite as high as the GIH WR would indicate.
Qualitatively, this card contributes in a lot of ways. A 2/2 rummager would be a C- or C in most sets. Rummaging is a great way to enable descend, one of the sets core mechanics. Red has a decent number of artifact payoffs especially when paired with Blue (one of the top early colors).
Another C+. Playable in any Red deck, straight up good in an artifact or descend deck.
2.2.3 Sunfire Torch
13th in GIH WR and 19th in ATA. You'd expect DV to split the difference, but it actually ranks 14th, right in line with the GIH WR. The removal being rather poor explains this performance.
Equipment can be invalidated by a high density of opposing removal (nothing to equip to) which isn't a concern in this set. Moreover, with players depending on blockers to stabilize the cheap stats boost can really open attacks and disrupt your opponent's plan. The sacrifice ability isn't just for dealing those last 2 points, it can even double as removal in a set that sorely needs interaction.
C to C+ depending on how aggressive you are
2.3 Over Drafted Cards
2.3.1 Rampaging Ceratops
This big dino is the 10th highest drafted Red common/uncommon but has just the 24th best GIH WR. Is this a case of a good card being taken too highly or is Ceratops simply not a top performer?
DV ranks this in a lowly 22nd. In other words, the raw GIH WR is being very fair to this card. Qualitatively, I do like the stat line. However, Red has a ton of great cheap cards in this set. Based on the early data, it's advisable to scoop up cheap cards early and fill out curve toppers with later picks. This is still a fine 5-drop, you just shouldn't prioritize 5-drops in Red this set.
Treat this as a C-
2.3.2 Rumbling Rockslide & Idol of the Deep King
These cards rank very similarly in GIH WR, ATA, and therefore DV. They have the 19th and 20th best GIH WR with being the 12th and 13th highest in ATA. DV ranks them 17th and 18th.
The removal in this set isn't very good, so it's easy to see why these are being taken relatively early. I wouldn't go as far as calling these over drafted. Rather, I think they're being misevaluated.
Early in the draft, if a pack has premium removal (e.g. Abrade) and a great threat (e.g. Belligerent Yearling) it's tempting to take the threat since you can get Rockslide or Idol later in the pack. However, based on the data I'd take the premium removal over the threat. The replacement level threats proxy the premium threats better than the replacement level removal proxies the premium removal.
To put it simply, snap up Abrade and Triumphant Chomp instead of counting on Rockslide and Idol as replacements.
Appendix A – Back testing
I did a quick back test using the Red cards in Midnight Hunt since the model was inspired by the comparison between Moonrager's Slash and Neonate's Rush. Data is all from 17 lands (awesome website).
I began with the square root model (see table A3). After calculating DV, I ranked each card by GIH WR, ATA, and DV. It looked like the DV rank was closer to the ATA rank than GIH WR rank. To test this, I got the distance (absolute difference) between DV rank and ATA/GIH WR rank for each card. On average, DV was closer to ATA (average distance of 3.1 ranks) than GIH WR (4.1 ranks). Using the cubed root gave better results (see table A2).
Fitting z-scores would give an even clearer picture than the rank. Cards that are far apart in rank may be close in value if the data is clustered. However, none of GIH WR, ATA, and DV are normally distributed. I was fine to use ranks as an estimate. This is the biggest area for development for back testing.
Table A1 shows granular results. GIH WR ranks Sunstreak Phoenix 19th while DV ranks it 7th. Qualitatively, I’d take Phoenix highly. While it was certainly over drafted (ATA ranked 2nd), 19th definitely undersells it. Meanwhile, Neonate's Rush ranked 8th in GIH WR but 17th in DV and middle of the road is where I’d place it. Thermo-Alchemist scored 5th in GIH WR and 10th in DV. In the right deck Alchemist was a phenomenal 2 drop but (especially in pack 1) I wouldn’t draft it over the top 9 cards. DV also loves Fangblade Brigand (15th vs 25th in GIH WR) and I would rank it closer to 15 than 25.
Conversely, I think DV underscored Famished Foragers at 25th compared to 17th in GIH WR. Red wants to double spell. 18th seems generous for Purifying Dragon (27th in GIH WR). I’d probably take solid 2 drops like Flame Channeler, Obsessive Astronomer, and Festival Crasher over it. 20th in DV is too low for Festival Crasher. I suspect this is because it was a great pick for a spells deck and a mediocre one otherwise. In other words, it was either sought after or ignored which makes the ATA (and therefore DV) unreliable. Finally, 27th is too high for Geistflame Reservoir, rare drafting likely skewed the ATA & DV.
Overall DV ranks MID’s Red cards in a reasonable order. Definitely not perfect, but pretty fair. DV struggles at evaluating contextual cards. It is best suited for ranking staples.
Table A1: Midnight Hunt Red DV – Cubed Root Model
Name | GIH WR | ATA | DV | Rank WR | Rank ATA | Rank DV |
---|---|---|---|---|---|---|
Moonveil Regent | 61.8% | 1.3 | 66.2 | 1 | 1 | 1 |
Reckless Stormseeker | 59.9% | 1.7 | 55.8 | 2 | 4 | 2 |
Burn Down the House | 58.0% | 1.8 | 49.3 | 3 | 5 | 3 |
Light Up the Night | 57.7% | 2.0 | 46.8 | 4 | 6 | 4 |
Bloodthirsty Adversary | 56.2% | 1.6 | 46.4 | 9 | 3 | 5 |
Smoldering Egg | 56.8% | 2.3 | 42.7 | 6 | 7 | 6 |
Sunstreak Phoenix | 54.1% | 1.5 | 40.9 | 19 | 2 | 7 |
Play with Fire | 56.5% | 4.0 | 34.7 | 7 | 9 | 8 |
Cathartic Pyre | 55.5% | 3.3 | 34.7 | 10 | 8 | 9 |
Thermo-Alchemist | 57.0% | 4.7 | 33.8 | 5 | 15 | 10 |
Moonrager's Slash | 55.5% | 4.2 | 31.9 | 10 | 12 | 11 |
Falkenrath Pit Fighter | 54.2% | 4.1 | 29.5 | 17 | 11 | 12 |
Lunar Frenzy | 55.0% | 5.0 | 29.3 | 13 | 16 | 13 |
Burn the Accursed | 54.6% | 6.1 | 26.6 | 14 | 19 | 14 |
Fangblade Brigand | 53.0% | 4.4 | 26.5 | 25 | 13 | 15 |
Seize the Storm | 54.4% | 6.2 | 26.1 | 15 | 20 | 16 |
Neonate's Rush | 56.4% | 9.6 | 25.7 | 8 | 29 | 17 |
Purifying Dragon | 52.4% | 4.4 | 25.3 | 27 | 14 | 18 |
Flame Channeler | 53.4% | 5.8 | 24.8 | 21 | 18 | 19 |
Festival Crasher | 55.2% | 8.5 | 24.8 | 12 | 25 | 20 |
Spellrune Painter | 53.0% | 5.3 | 24.8 | 25 | 17 | 21 |
Obsessive Astronomer | 53.3% | 6.5 | 23.8 | 22 | 21 | 22 |
Voldaren Ambusher | 53.1% | 7.0 | 22.8 | 24 | 23 | 23 |
Voldaren Stinger | 54.3% | 9.5 | 22.5 | 16 | 28 | 24 |
Famished Foragers | 54.2% | 9.6 | 22.2 | 17 | 30 | 25 |
Stolen Vitality | 53.5% | 10.3 | 20.7 | 20 | 34 | 26 |
Geistflame Reservoir | 49.7% | 4.0 | 20.4 | 38 | 9 | 27 |
Electric Revelation | 53.3% | 10.9 | 20.0 | 22 | 37 | 28 |
Falkenrath Perforator | 52.2% | 9.3 | 19.4 | 28 | 26 | 29 |
Ardent Elementalist | 51.8% | 8.4 | 19.3 | 31 | 24 | 30 |
Immolation | 52.0% | 9.3 | 19.0 | 29 | 27 | 31 |
Lambholt Harrier | 51.9% | 9.8 | 18.5 | 30 | 32 | 32 |
Brimstone Vandal | 51.5% | 10.0 | 17.8 | 33 | 33 | 33 |
Abandon the Post | 51.7% | 11.9 | 17.1 | 32 | 40 | 34 |
Mounted Dreadknight | 51.1% | 10.6 | 16.8 | 34 | 36 | 35 |
Harvesttide Infiltrator | 50.7% | 9.7 | 16.7 | 36 | 31 | 36 |
Pack's Betrayal | 51.1% | 11.7 | 16.3 | 34 | 38 | 37 |
Raze the Effigy | 50.7% | 11.8 | 15.6 | 36 | 39 | 38 |
Village Watch | 47.8% | 7.0 | 13.6 | 40 | 22 | 39 |
Tavern Ruffian | 48.4% | 10.6 | 12.8 | 39 | 35 | 40 |
Table A2: DV Distance from GIH WR and ATA by Model
Card | Square Root | Cubed Root | ||
DV - WR | DV - ATA | DV - WR | DV - ATA | |
Moonveil Regent | 0 | 0 | 0 | 0 |
Reckless Stormseeker | 0 | 2 | 0 | 2 |
Burn Down the House | 0 | 2 | 0 | 2 |
Light Up the Night | 1 | 1 | 0 | 2 |
Bloodthirsty Adversary | 5 | 1 | 4 | 2 |
Smoldering Egg | 1 | 0 | 0 | 1 |
Sunstreak Phoenix | 13 | 4 | 12 | 5 |
Play with Fire | 2 | 0 | 1 | 1 |
Cathartic Pyre | 2 | 0 | 1 | 1 |
Thermo-Alchemist | 5 | 5 | 5 | 5 |
Moonrager's Slash | 1 | 1 | 1 | 1 |
Falkenrath Pit Fighter | 5 | 1 | 5 | 1 |
Lunar Frenzy | 0 | 3 | 0 | 3 |
Burn the Accursed | 2 | 3 | 0 | 5 |
Fangblade Brigand | 11 | 1 | 10 | 2 |
Seize the Storm | 2 | 3 | 1 | 4 |
Neonate's Rush | 12 | 9 | 9 | 12 |
Purifying Dragon | 12 | 1 | 9 | 4 |
Flame Channeler | 2 | 1 | 2 | 1 |
Festival Crasher | 10 | 3 | 8 | 5 |
Spellrune Painter | 7 | 1 | 4 | 4 |
Obsessive Astronomer | 1 | 0 | 0 | 1 |
Voldaren Ambusher | 1 | 0 | 1 | 0 |
Voldaren Stinger | 9 | 3 | 8 | 4 |
Famished Foragers | 9 | 4 | 8 | 5 |
Stolen Vitality | 7 | 7 | 6 | 8 |
Geistflame Reservoir | 14 | 15 | 11 | 18 |
Electric Revelation | 7 | 8 | 6 | 9 |
Falkenrath Perforator | 2 | 4 | 1 | 3 |
Ardent Elementalist | 3 | 4 | 1 | 6 |
Immolation | 2 | 4 | 2 | 4 |
Lambholt Harrier | 2 | 0 | 2 | 0 |
Brimstone Vandal | 0 | 0 | 0 | 0 |
Abandon the Post | 4 | 4 | 2 | 6 |
Mounted Dreadknight | 1 | 1 | 1 | 1 |
Harvesttide Infiltrator | 2 | 3 | 0 | 5 |
Pack's Betrayal | 3 | 1 | 3 | 1 |
Raze the Effigy | 2 | 1 | 2 | 1 |
Village Watch | 1 | 17 | 1 | 17 |
Tavern Ruffian | 1 | 5 | 1 | 5 |
Average | 4.1 | 3.1 | 3.2 | 3.9 |
Table A3: Midnight Hunt Red DV – Square Root Model
Name | GIH WR | ATA | DV | Rank WR | Rank ATA | Rank DV |
Moonveil Regent | 61.8% | 1.3 | 63.2 | 1 | 1 | 1 |
Reckless Stormseeker | 59.9% | 1.7 | 51.2 | 2 | 4 | 2 |
Burn Down the House | 58.0% | 1.8 | 44.7 | 3 | 5 | 3 |
Bloodthirsty Adversary | 56.2% | 1.6 | 43.0 | 9 | 3 | 4 |
Light Up the Night | 57.7% | 2.0 | 41.6 | 4 | 6 | 5 |
Sunstreak Phoenix | 54.1% | 1.5 | 38.1 | 19 | 2 | 6 |
Smoldering Egg | 56.8% | 2.3 | 37.3 | 6 | 7 | 7 |
Cathartic Pyre | 55.5% | 3.3 | 28.4 | 10 | 8 | 8 |
Play with Fire | 56.5% | 4.0 | 27.6 | 7 | 9 | 9 |
Thermo-Alchemist | 57.0% | 4.7 | 26.1 | 5 | 15 | 10 |
Moonrager's Slash | 55.5% | 4.2 | 25.1 | 10 | 12 | 11 |
Falkenrath Pit Fighter | 54.2% | 4.1 | 23.3 | 17 | 11 | 12 |
Lunar Frenzy | 55.0% | 5.0 | 22.4 | 13 | 16 | 13 |
Fangblade Brigand | 53.0% | 4.4 | 20.7 | 25 | 13 | 14 |
Purifying Dragon | 52.4% | 4.4 | 19.7 | 27 | 14 | 15 |
Burn the Accursed | 54.6% | 6.1 | 19.6 | 14 | 19 | 16 |
Seize the Storm | 54.4% | 6.2 | 19.2 | 15 | 20 | 17 |
Spellrune Painter | 53.0% | 5.3 | 18.8 | 25 | 17 | 18 |
Flame Channeler | 53.4% | 5.8 | 18.5 | 21 | 18 | 19 |
Neonate's Rush | 56.4% | 9.6 | 17.7 | 8 | 29 | 20 |
Obsessive Astronomer | 53.3% | 6.5 | 17.4 | 22 | 21 | 21 |
Festival Crasher | 55.2% | 8.5 | 17.4 | 12 | 25 | 22 |
Voldaren Ambusher | 53.1% | 7.0 | 16.5 | 24 | 23 | 23 |
Geistflame Reservoir | 49.7% | 4.0 | 16.2 | 38 | 9 | 24 |
Voldaren Stinger | 54.3% | 9.5 | 15.5 | 16 | 28 | 25 |
Famished Foragers | 54.2% | 9.6 | 15.2 | 17 | 30 | 26 |
Stolen Vitality | 53.5% | 10.3 | 14.0 | 20 | 34 | 27 |
Ardent Elementalist | 51.8% | 8.4 | 13.6 | 31 | 24 | 28 |
Electric Revelation | 53.3% | 10.9 | 13.4 | 22 | 37 | 29 |
Falkenrath Perforator | 52.2% | 9.3 | 13.4 | 28 | 26 | 30 |
Immolation | 52.0% | 9.3 | 13.1 | 29 | 27 | 31 |
Lambholt Harrier | 51.9% | 9.8 | 12.7 | 30 | 32 | 32 |
Brimstone Vandal | 51.5% | 10.0 | 12.2 | 33 | 33 | 33 |
Harvesttide Infiltrator | 50.7% | 9.7 | 11.5 | 36 | 31 | 34 |
Mounted Dreadknight | 51.1% | 10.6 | 11.4 | 34 | 36 | 35 |
Abandon the Post | 51.7% | 11.9 | 11.3 | 32 | 40 | 36 |
Pack's Betrayal | 51.1% | 11.7 | 10.8 | 34 | 38 | 37 |
Raze the Effigy | 50.7% | 11.8 | 10.4 | 36 | 39 | 38 |
Village Watch | 47.8% | 7.0 | 9.9 | 40 | 22 | 39 |
Tavern Ruffian | 48.4% | 10.6 | 8.6 | 39 | 35 | 40 |