An R package for finding optimized armor combinations in Dark Souls Remastered. It ships exact stats for every armor piece in the game, computing any upgrade level with the game's own upgrade rates, a combinatorial search that scores every valid head/chest/hands/legs combination against a set of weighted priorities and constraints, and an interactive Shiny app built on top of that search. It can also chart trade-offs: the most of a stat any combination can reach at every armor weight.
install.packages("darksoulsarmor", repos = c("https://ricewhitlam.r-universe.dev", "https://cloud.r-project.org"))
or, to build from source (needs build tools for the C++ code):
devtools::install_github("ricewhitlam/darksoulsarmor", build_vignettes = TRUE)
Requires R 4.3 or later.
build_vignettes = TRUE installs the "scoring" vignette referenced below; without it, the same content is readable on GitHub at vignettes/scoring.Rmd.
The app also runs entirely in your browser, with nothing to install, at https://ricewhitlam.github.io/darksoulsarmor/. The first visit downloads R itself and takes a little while; later visits are faster.
The easiest way to use the package is the Shiny app:
library(darksoulsarmor)
armor.application()
Adjust the settings in the sidebar (which armor pieces to consider, upgrade levels, rings, weapons, movement type, minimum stats, and how much each stat should matter) and click "Refresh Armor Data". The Results tab lists the top combinations. The Trade-offs tab charts the most of a chosen stat (the score, Poise, or one defense or resistance) that any combination can reach at every armor weight, with lines where the movement type changes. Click a row or point for links to the armor pieces on the Dark Souls wiki. "Download Armor Data" saves an Excel workbook of both tabs and the settings behind them. The in-app "User Guide" button explains every input in detail.
The same search is available directly as a function, for scripting or exploring results outside the app:
library(darksoulsarmor)
result <- get.optimal.armor.combos(
max.table.size = 5,
endurance.level = 40,
weapons = c(right.1 = "Great Club"), # weight 12
movement = "Mid",
minima = c(POISE = 30), # require at least 30 poise
weights = c(16, 16, 16, 16, 8, 8, 8, 4, 4, 4)
)
result$data[, .(SCORE_QUALITY, HEAD, CHEST, HANDS, LEGS, ARMOR_POISE, PCT_LOAD)]
#> SCORE_QUALITY HEAD CHEST HANDS
#> <char> <char> <char> <char>
#> 1: Top 1 in 193 Crown of the Great Lord Sage Robe Smough's Gauntlets
#> 2: Top 1 in 181 Smough's Helm Smough's Armor Antiquated Gloves
#> 3: Top 1 in 170 Big Hat Sage Robe Smough's Gauntlets
#> 4: Top 1 in 166 Bloated Head Sage Robe Smough's Gauntlets
#> 5: Top 1 in 165 Ornstein's Helm Smough's Armor Witch Gloves
#> LEGS ARMOR_POISE PCT_LOAD
#> <char> <num> <num>
#> 1: Smough's Leggings 42 0.49750
#> 2: Gold-Hemmed Black Skirt 49 0.50000
#> 3: Smough's Leggings 42 0.49750
#> 4: Smough's Leggings 42 0.49125
#> 5: Gold-Hemmed Black Skirt 44 0.49750
See ?get.optimal.armor.combos for the full set of settings (which areas/classes make a piece available, upgrade levels, movement type, ring bonuses, per-stat minimums, and per-stat weights). get.all.armor.combos() returns every valid combination unscored, for your own analysis.
get.armor.tradeoffs() finds the best value of one stat at each armor-weight limit, with the same settings as get.optimal.armor.combos(). For example, the lightest armor that reaches each PVE poise breakpoint with Mid movement:
curve <- get.armor.tradeoffs(metric = "POISE", endurance.level = 40, weapons = c(right.1 = "Great Club"), movement = "Mid", weights = c(16, 16, 16, 16, 8, 8, 8, 4, 4, 4))
first <- sapply(c(21, 31, 46), function(poise) which(curve$data$BEST_VALUE >= poise)[1])
curve$data[first, .(ARMOR_WEIGHT_LIMIT, BEST_VALUE, HEAD, CHEST, HANDS, LEGS)]
#> ARMOR_WEIGHT_LIMIT BEST_VALUE HEAD CHEST HANDS
#> <num> <num> <char> <char> <char>
#> 1: 7.8 21 Giant Helm No Chest No Hands
#> 2: 12.5 31 Havel's Helm No Chest Knight Gauntlets
#> 3: 18.7 46 Havel's Helm No Chest Giant Gauntlets
#> LEGS
#> <char>
#> 1: Hollow Soldier Waistcloth
#> 2: Hollow Soldier Waistcloth
#> 3: Hollow Soldier Waistcloth
weight.step sets the spacing of the limits (default 0.1), and min.armor.weight/max.armor.weight set the range; by default it runs from 0 to what the movement type allows. Among combinations tied on the stat, the best-scoring one is returned.
get.tradeoff.efficiency() describes a curve in four kinds of piece, to show where extra weight pays off:
curve <- get.armor.tradeoffs(metric = "CURSE_RES", endurance.level = 40, movement = "Fat", weights = c(16, 16, 16, 16, 8, 8, 8, 4, 4, 4))
efficiency <- get.tradeoff.efficiency(curve)
efficiency$data[, .(TYPE, FROM, TO, GAIN, RATIO_TO_AVERAGE)]
#> TYPE FROM TO GAIN RATIO_TO_AVERAGE
#> <char> <num> <num> <num> <num>
#> 1: line 0.0 5.0 114 6.7247191
#> 2: boundary 4.9 5.0 11 NA
#> 3: line 5.0 28.4 49 0.6176174
#> 4: jump 28.4 28.5 15 NA
#> 5: flat 28.5 52.5 0 0.0000000
Here curse resistance pays off steeply up to 5.0 armor weight, at 6.7 times the curve's average. It then buys less than average up to 28.4, jumps by 15 at 28.5, and gains nothing after that.
jumps and flats (each "few", "some" (the default) or "many") set the sensitivity at which these features are identified. The rules and their defaults were tuned by judgement to describe typical curves well, so the description is a heuristic rather than an exact measure.
Every combination gets a score built from ten stats (physical/strike/slash/thrust/magic/fire/lightning defense, and bleed/poison/curse resistance). weights sets how much each counts: either one weight per stat, or, with weights.mode = "grouped", one per group (physical defenses, elemental defenses and resistances, each counting as the average of its stats). stat.scale sets what one unit of each stat counts as before it's weighted: its standard deviation across every armor combination ("sd", the default, so that stats that vary more between sets don't dominate), its highest value ("max"), or one point ("raw"). The resulting SCORE is standardized to mean 0 and variance 1 across every combination, and SCORE_QUALITY converts it into an exact rarity among every combination at the selected upgrade levels, like "Top 1 in 40" or "Bottom 1 in 40". See vignette("scoring") for exactly how that's computed. Scores are directly comparable across different filter and constraint choices as long as the weights, stat.scale and weights.mode are the same.
The underlying armor stats live in create_rda/*.csv and are compiled into the package's shipped data by create_rda/create_rda.R — see that directory for details on the build process and its own dependencies.
GPL (>= 2)