diff --git a/README.md b/README.md
index de7bd61..98e41c6 100644
--- a/README.md
+++ b/README.md
@@ -77,7 +77,6 @@ The lines in a file look exactly the same as regular entries in the prompt, but
 
 
 ## Todo (pre-release, in this order):
- - Export will break with :save, fix it.
  - Prevent macro-chaining recursion
  - Full-reduce option (expand all macros)
  - step-by-step reduction
diff --git a/lamb/nodes/functions.py b/lamb/nodes/functions.py
index 423bdb3..1f3236d 100644
--- a/lamb/nodes/functions.py
+++ b/lamb/nodes/functions.py
@@ -166,11 +166,14 @@ def prepare(root: lbn.Root, *, ban_macro_name = None) -> list:
 				if (n.input.name in bound_variables):
 					raise lbn.ReductionError(f"Bound variable name conflict: \"{n.input.name}\"")
 				else:
-					bound_variables[n.input.name] = lbn.Bound(n.input.name)
+					bound_variables[n.input.name] = lbn.Bound(
+						lamb.utils.remove_sub(n.input.name),
+						macro_name = n.input.name
+					)
 					n.input = bound_variables[n.input.name]
 
 			elif s == lbn.Direction.LEFT:
-				del bound_variables[n.input.name]
+				del bound_variables[n.input.macro_name] # type: ignore
 
 	return warnings
 
diff --git a/lamb/nodes/nodes.py b/lamb/nodes/nodes.py
index 70f877e..e231e20 100644
--- a/lamb/nodes/nodes.py
+++ b/lamb/nodes/nodes.py
@@ -333,11 +333,16 @@ class History(ExpandableEndNode):
 
 bound_counter = 0
 class Bound(EndNode):
-	def __init__(self, name: str, *, forced_id = None, runner = None):
+	def __init__(self, name: str, *, forced_id = None, runner = None, macro_name = None):
 		self.name = name
 		global bound_counter
 		self.runner = runner # type: ignore
 
+		# The name of the macro this bound came from.
+		# Always equal to self.name, unless the macro
+		# this came from had a subscript.
+		self.macro_name: str | None = macro_name
+
 		if forced_id is None:
 			self.identifier = bound_counter
 			bound_counter += 1
diff --git a/lamb/parser.py b/lamb/parser.py
index 0d32a81..1198eee 100755
--- a/lamb/parser.py
+++ b/lamb/parser.py
@@ -13,7 +13,7 @@ class LambdaParser:
 		# We still create macro objects from them, they are turned into
 		# bound variables after the expression is built.
 		self.pp_macro = pp.Word(pp.alphas + "_")
-		self.pp_bound = pp.Char(pp.srange("[a-z]"))
+		self.pp_bound = pp.Regex("[a-z][₀₁₂₃₄₅₆₈₉]*")
 		self.pp_name = self.pp_bound ^ self.pp_macro
 		self.pp_church = pp.Word(pp.nums)
 		self.pp_history = pp.Char("$")
diff --git a/lamb/utils.py b/lamb/utils.py
index 22a8a09..62617a0 100644
--- a/lamb/utils.py
+++ b/lamb/utils.py
@@ -5,6 +5,8 @@ from prompt_toolkit.key_binding import KeyBindings
 from prompt_toolkit import print_formatted_text as printf
 from importlib.metadata import version
 
+import re
+
 
 style = Style.from_dict({ # type: ignore
 	# Basic formatting
@@ -108,6 +110,9 @@ def show_greeting():
 		""
 	])), style = style)
 
+def remove_sub(s: str):
+	return re.sub("[₀₁₂₃₄₅₆₈₉]*", "", s)
+
 def base4(n: int):
 	if n == 0:
 		return [0]
@@ -119,9 +124,9 @@ def base4(n: int):
 
 def subscript(num: int):
 
-	# unicode subscripts  ₀₁₂₃
-	# usually look different than
-	# the rest, so we'll use base 4.
+	# unicode subscripts  ₀₁₂₃ and ₄₅₆₈₉
+	# usually look different,
+	# so we'll use base 4.
 	qb = base4(num)
 
 	sub = {