diff --git a/dist/habitrpg-shared.js b/dist/habitrpg-shared.js
index 190d0db602..1fdb40587a 100644
--- a/dist/habitrpg-shared.js
+++ b/dist/habitrpg-shared.js
@@ -11688,6 +11688,9 @@ var process=require("__browserify_process");(function() {
             currVal = task.value < -47.27 ? -47.27 : task.value > 21.27 ? 21.27 : task.value;
             nextDelta = Math.pow(0.9747, currVal) * (direction === 'down' ? -1 : 1);
             if (task.type !== 'reward') {
+              if (user.preferences.automaticAllocation === true && user.preferences.allocationMode === 'taskbased') {
+                user.stats.training[task.attribute] += nextDelta;
+              }
               adjustAmt = nextDelta;
               if (direction === 'up' && task.type !== 'reward' && !(task.type === 'habit' && !task.down)) {
                 adjustAmt = nextDelta * (1 + user._statsComputed.str * .004);
@@ -11963,7 +11966,7 @@ var process=require("__browserify_process");(function() {
       */
 
       autoAllocate: function() {
-        var diff, ideal, preference, suggested, tallies;
+        var diff, ideal, preference, suggested;
         switch (user.preferences.allocationMode) {
           case "flat":
             suggested = Math.min(user.stats.str, user.stats.int, user.stats.con, user.stats.per);
@@ -12005,23 +12008,21 @@ var process=require("__browserify_process");(function() {
             }
             break;
           case "taskbased":
-            tallies = _.reduce(user.tasks, (function(m, v) {
-              m[v.attribute || 'str'] += v.value;
-              return m;
-            }), {
-              str: 0,
-              int: 0,
-              con: 0,
-              per: 0
-            });
-            suggested = _.reduce(tallies, (function(m, v, k) {
-              if (v > tallies[m]) {
-                return k;
-              } else {
-                return m;
+            suggested = _.findKey(user.stats.training, (function(val) {
+              if (val === _.max(user.stats.training)) {
+                return val;
               }
-            }), 'str');
-            return suggested;
+            }));
+            user.stats.training.str = 0;
+            user.stats.training.int = 0;
+            user.stats.training.con = 0;
+            user.stats.training.per = 0;
+            if (suggested === void 0) {
+              return "str";
+            } else {
+              return suggested;
+            }
+            break;
           default:
             return "str";
         }
diff --git a/script/index.coffee b/script/index.coffee
index 16b060629c..73e3f2ca25 100644
--- a/script/index.coffee
+++ b/script/index.coffee
@@ -687,6 +687,7 @@ api.wrap = (user) ->
             else task.value
           nextDelta = Math.pow(0.9747, currVal) * (if direction is 'down' then -1 else 1)
           unless task.type is 'reward'
+            if (user.preferences.automaticAllocation is true and user.preferences.allocationMode is 'taskbased') then user.stats.training[task.attribute] += nextDelta
             adjustAmt = nextDelta
             # ===== STRENGTH =====
             # (Only for up-scoring, ignore up-onlies and rewards)
@@ -965,11 +966,17 @@ api.wrap = (user) ->
           # Get the difference between the ideal attribute spread according to level, and the user's current spread.
           diff = [(user.stats[preference[0]]-ideal[0]),(user.stats[preference[1]]-ideal[1]),(user.stats[preference[2]]-ideal[2]),(user.stats[preference[3]]-ideal[3])]
           suggested = _.findIndex(diff, ((val) -> if val is _.min(diff) then true)) # Returns the index of the first attribute that's furthest behind the ideal
-          if suggested is -1 then return "str" else return preference[suggested] # If _findIndex failed, we'd get a -1...
-        when "taskbased" # old logic, temp
-          tallies = _.reduce user.tasks, ((m,v)-> m[v.attribute or 'str'] += v.value;m), {str:0,int:0,con:0,per:0}
-          suggested = _.reduce tallies, ((m,v,k)-> if v>tallies[m] then k else m), 'str'
-          return suggested
+          if suggested is -1 then return "str" else return preference[suggested] # If _.findIndex failed, we'd get a -1...
+        when "taskbased"
+          suggested = _.findKey(user.stats.training, ((val) -> if val is _.max(user.stats.training) then val)) # Returns the stat that's been trained up the most this level
+          # FIXME Reset training for this level. Tried _.each but couldn't get it to take.
+          user.stats.training.str = 0
+          user.stats.training.int = 0
+          user.stats.training.con = 0
+          user.stats.training.per = 0
+          if suggested is undefined then return "str" else return suggested # Failed _.findkey gives undefined
+          # tallies = _.reduce user.tasks, ((m,v)-> m[v.attribute or 'str'] += v.value;m), {str:0,int:0,con:0,per:0}
+          # suggested = _.reduce tallies, ((m,v,k)-> if v>tallies[m] then k else m), 'str'
         else return "str" # if all else fails, dump into STR
 
     updateStats: (stats) ->