diff --git a/src/main/kotlin/ru/dbotthepony/mc/otm/core/math/Clustering.kt b/src/main/kotlin/ru/dbotthepony/mc/otm/core/math/Clustering.kt
index 22b244067..a4faa615f 100644
--- a/src/main/kotlin/ru/dbotthepony/mc/otm/core/math/Clustering.kt
+++ b/src/main/kotlin/ru/dbotthepony/mc/otm/core/math/Clustering.kt
@@ -13,11 +13,11 @@ interface Cluster<V : Any> {
 }
 
 private class ClusterValue<V : Comparable<V>>(val value: V, var cluster: MutableCluster<V>, var error: V) {
-	inline fun updateError(abs: (V) -> V, minus: (V, V) -> V) {
+	fun updateError(abs: (V) -> V, minus: (V, V) -> V) {
 		error = abs(minus(cluster.center, value))
 	}
 
-	inline fun maybeSwitchCluster(cluster: MutableCluster<V>, abs: (V) -> V, minus: (V, V) -> V): Boolean {
+	fun maybeSwitchCluster(cluster: MutableCluster<V>, abs: (V) -> V, minus: (V, V) -> V): Boolean {
 		if (cluster == this.cluster) return false
 
 		val newError = abs(minus(cluster.center, value))
@@ -26,6 +26,8 @@ private class ClusterValue<V : Comparable<V>>(val value: V, var cluster: Mutable
 			error = newError
 			this.cluster.values.remove(this)
 			cluster.values.add(this)
+			cluster.generation++
+			this.cluster.generation++
 			this.cluster = cluster
 			return true
 		} else {
@@ -34,12 +36,16 @@ private class ClusterValue<V : Comparable<V>>(val value: V, var cluster: Mutable
 	}
 }
 
-private class MutableCluster<V : Comparable<V>>(var center: V, expectedSize: Int) {
-	val values = ObjectArrayList<ClusterValue<V>>(min(100, expectedSize))
+private class MutableCluster<V : Comparable<V>>(var center: V) {
+	val values = ObjectArrayList<ClusterValue<V>>()
+	var generation = 0
+	var centerCalculatedAtGeneration = -1
 
-	inline fun calculateCenter(identity: V, plus: (V, V) -> V, minus: (V, V) -> V, divInt: (V, Int) -> V, abs: (V) -> V): Boolean {
-		if (values.isEmpty()) return false
+	fun calculateCenter(identity: V, plus: (V, V) -> V, minus: (V, V) -> V, divInt: (V, Int) -> V, abs: (V) -> V): Boolean {
+		if (values.isEmpty) return false
+		if (centerCalculatedAtGeneration == generation) return true
 
+		centerCalculatedAtGeneration = generation
 		var value = identity
 		values.forEach { value = plus(value, it.value) }
 		val old = center
@@ -51,9 +57,37 @@ private class MutableCluster<V : Comparable<V>>(var center: V, expectedSize: Int
 
 		return true
 	}
+
+	fun pullFrom(clusters: MutableList<MutableCluster<V>>, identity: V): Boolean {
+		var candidate: ClusterValue<V>? = null
+
+		for (eCluser in clusters) {
+			if (eCluser.values.size > 1) {
+				for (value in eCluser.values) {
+					if (candidate == null || candidate.error < value.error) {
+						candidate = value
+					}
+				}
+			}
+		}
+
+		if (candidate != null) {
+			values.add(candidate)
+			candidate.cluster.values.remove(candidate)
+			this.generation++
+			candidate.cluster.generation++
+			candidate.cluster = this
+			center = candidate.value
+			candidate.error = identity
+			clusters.add(this)
+			return true
+		} else {
+			return false
+		}
+	}
 }
 
-private inline fun <V : Comparable<V>> Iterable<V>.clusterize(
+private fun <V : Comparable<V>> Iterable<V>.clusterize(
 	random: RandomGenerator,
 	initialClusters: Int = 1,
 	identity: V,
@@ -63,49 +97,43 @@ private inline fun <V : Comparable<V>> Iterable<V>.clusterize(
 	abs: (V) -> V,
 	heuristics: (min: V, max: V, error: V) -> Boolean,
 ): List<Cluster<V>> {
+	require(initialClusters > 0) { "Invalid amount of initial clusters: $initialClusters" }
 	val itr = iterator()
 
 	if (!itr.hasNext())
 		return listOf()
 
-	var expectedSize = 1
+	val clusters = ObjectArrayList<MutableCluster<V>>(initialClusters)
+
+	for (i in 0 until initialClusters) {
+		clusters.add(MutableCluster(identity))
+	}
+
+	val values = ObjectArrayList<ClusterValue<V>>()
 	var min = itr.next()
 	var max = min
 
-	while (itr.hasNext()) {
-		val value = itr.next()
+	values.add(ClusterValue(min, clusters.random(random), min))
+	values[0].cluster.values.add(values[0])
 
+	for (value in itr) {
 		min = minOf(min, value)
 		max = maxOf(max, value)
-		expectedSize++
+
+		val cluster = clusters.random(random)
+		val wrapped = ClusterValue(value, cluster, value)
+		values.add(wrapped)
+		cluster.values.add(wrapped)
 	}
 
 	if (min == max) {
 		return listOf(Cluster.Impl(listOf(min), min))
 	}
 
-	var targetClusters = initialClusters
-	val clusters = ObjectArrayList<MutableCluster<V>>(initialClusters)
-	val values = ObjectArrayList<ClusterValue<V>>(expectedSize)
-
 	while (true) {
-		clusters.clear()
-		values.clear()
-
 		var converged = false
 		var oversaturation = false
 
-		for (i in 0 until targetClusters) {
-			clusters.add(MutableCluster(identity, expectedSize))
-		}
-
-		for (value in this) {
-			val cluster = clusters.random(random)
-			val wrapped = ClusterValue(value, cluster, value)
-			values.add(wrapped)
-			cluster.values.add(wrapped)
-		}
-
 		clusters.forEach { it.calculateCenter(identity, plus, minus, divInt, abs) }
 
 		while (!converged) {
@@ -127,26 +155,7 @@ private inline fun <V : Comparable<V>> Iterable<V>.clusterize(
 				}
 
 				for (cluster in emptyClusters) {
-					var candidate: ClusterValue<V>? = null
-
-					for (eCluser in clusters) {
-						if (eCluser.values.size > 1) {
-							for (value in eCluser.values) {
-								if (candidate == null || candidate.error < value.error) {
-									candidate = value
-								}
-							}
-						}
-					}
-
-					if (candidate != null) {
-						cluster.values.add(candidate)
-						candidate.cluster.values.remove(candidate)
-						candidate.cluster = cluster
-						cluster.center = candidate.value
-						candidate.error = identity
-						clusters.add(cluster)
-					} else {
+					if (!cluster.pullFrom(clusters, identity)) {
 						oversaturation = true
 						break
 					}
@@ -156,8 +165,10 @@ private inline fun <V : Comparable<V>> Iterable<V>.clusterize(
 
 		val maxError = values.maxOf { it.error }
 
-		if (!oversaturation && targetClusters < values.size && heuristics(min, max, maxError)) {
-			targetClusters++
+		if (!oversaturation && clusters.size < values.size && heuristics(min, max, maxError)) {
+			val cluster = MutableCluster(identity)
+			check(cluster.pullFrom(clusters, identity)) { "Newly created cluster couldn't pull first value from other clusters (expected at least one cluster with 2 values in it)" }
+			clusters.add(cluster)
 		} else {
 			return clusters.stream()
 				.filter { it.values.isNotEmpty() }