diff --git a/testing/ms_linalg.py b/testing/1.20_manasteal/ms_linalg.py
similarity index 100%
rename from testing/ms_linalg.py
rename to testing/1.20_manasteal/ms_linalg.py
diff --git a/testing/ms_simulator.py b/testing/1.20_manasteal/ms_simulator.py
similarity index 100%
rename from testing/ms_simulator.py
rename to testing/1.20_manasteal/ms_simulator.py
diff --git a/testing/ms_sslow.py b/testing/1.20_manasteal/ms_sslow.py
similarity index 100%
rename from testing/ms_sslow.py
rename to testing/1.20_manasteal/ms_sslow.py
diff --git a/testing/optimization/analyze_items.py b/testing/optimization/analyze_items.py
new file mode 100644
index 0000000..e44dae8
--- /dev/null
+++ b/testing/optimization/analyze_items.py
@@ -0,0 +1,137 @@
+import json
+
+import numpy as np
+import matplotlib.pyplot as plt
+
+def max_id(item, id_name, invert=False):
+    """
+    Calculate the "max roll" for a given ID.
+
+    Parameters
+        Name        type        desc
+        ----------------------------------------------------------
+        item        json        Item json data
+        id_name     string      name of the ID to get
+        invert      bool        Whether to "invert" (raw cost and %cost have funny
+                                0.7-1.3 positive roll and 0.3-1.3 negative roll)
+    
+    Return:
+        val: float -- max roll id value.
+    """
+    id_val = item.get(id_name, 0)
+    if id_val == 0: # if the ID isn't present, its just going to be zero
+        return 0
+
+    if item.get('fixID', False):
+        # If the item is a fixed roll item, don't roll the ID.
+        return id_val
+
+    # roll the ID. Negative roll (and invert) max roll is 0.7; positive max is 1.3.
+    if bool(id_val < 0) != bool(invert):    # logical XOR
+        val = round(id_val * 0.7)
+    else:   #if bool(id_val > 0) != bool(invert):
+        val = round(id_val * 1.3)
+
+    if val == 0:    # if we rounded to zero, then restore the id as sign(base_val).
+        val = id_val / abs(id_val)
+    return val
+
+def mv(item, base_costs):
+    """
+    Compute mana value for an item.
+    Takes a maximum mana value
+        - assuming 1 melee value (3/3 mana steal = 1 mana value)
+        - assuming spells 1, 3, and 4 are cycle spells.
+    Ignores spell 2 for spell cost purposes.
+
+    Parameters
+        Name        type            desc
+        ----------------------------------------------------------
+        item        json            Item json data
+        base_costs  list[float]     base spell cost [spell1, spell2, spell3, spell4]
+
+    Return:
+        val: float -- mana value.
+    """
+    cost_reductions = sorted([
+        max_id(item, 'spRaw1', True) + base_costs[0]*max_id(item, 'spPct1', True)/100,
+        #max_id(item, 'spRaw2', True) + base_costs[1]*max_id(item, 'spPct2', True)/100,
+        max_id(item, 'spRaw3', True) + base_costs[2]*max_id(item, 'spPct3', True)/100,
+        max_id(item, 'spRaw4', True) + base_costs[3]*max_id(item, 'spPct4', True)/100,
+    ])
+    cost_mv = -sum(cost_reductions[:2])
+
+    return (
+        max_id(item, 'ms')/3
+        + max_id(item, 'mr')/5
+        + cost_mv
+    )
+
+###########################
+# constants for damage calc.
+elements = 'rnetwfa'
+raw_ids = ['sdRaw'] + [x+'SdRaw' for x in elements] + [x+'DamRaw' for x in elements]
+
+# these %boosts apply to all damages.
+percent_all_ids = ['sdPct', 'rSdPct']
+
+# this one  is a list of lists.
+# the mini lists are sub-sums, the big list gets max'd over (elemental damage works like this.)
+percent_max_id_groups = list(zip([x+'DamPct' for x in 'etwfa'] + [x+'SdPct' for x in 'etwfa']))  # exclude neutral lel
+###########################
+def damage(item, weapon_base):
+    """
+    Compute effective damage bonus.
+    Note that this assumes the weapon aligns with whatever bonus this item is giving.
+
+    Parameters
+        Name            type        desc
+        ----------------------------------------------------------
+        item            json        Item json data
+        weapon_base     float       weapon base dps
+
+    Return:
+        val: float -- raw damage bonus given (approximate) for the weapon.
+    """
+    total = sum(max_id(item, x) for x in raw_ids)
+    total += weapon_base * sum(max_id(item, x) for x in percent_all_ids) / 100
+    total += weapon_base * max(sum(max_id(item, y) for y in x) for x in percent_max_id_groups) / 100
+    return total
+
+
+#################################
+# NOTE: Edit these parameters! LOL i was lazy to make a CLI
+level_threshold = 80
+weapon_base = 700
+base_costs = [35, 20, 35, 35]
+item_type = 'leggings'
+
+# TODO: Changeme to point to a copy of wynnbuilder's compress.json file!
+items = json.load(open("../../compress.json"))['items']
+#################################
+
+
+# collect data from items.
+points = []
+names = dict()
+for item in items:
+    if item['type'] == item_type and item['lvl'] > level_threshold:
+        # Edit me to see other comparisons!
+        #point = (mv(item, base_costs), damage(item, weapon_base))
+        point = (max_id(item, 'spd'), item.get('hp', 0) + max_id(item, 'hpBonus'))
+
+        points.append(point)
+        # just some shenanigans to aggregate text that happens to fall on the same point.
+        if point in names:
+            names[point] += '\n'+item.get('displayName', item['name'])
+        else:
+            names[point] = item.get('displayName', item['name'])
+points = np.array(points)
+
+# plot points.
+plt.figure()
+plt.scatter(points[:, 0], points[:, 1])
+# and add annotations.
+for point, txt in names.items(): 
+    plt.annotate(txt, point)
+plt.show()