diff --git a/src/Lexer.hs b/src/Lexer.hs
index 77fc84a..3b04ecc 100644
--- a/src/Lexer.hs
+++ b/src/Lexer.hs
@@ -42,6 +42,7 @@ data Token
   | Escape
   | EmphasisOpen
   | EmphasisClose
+  | Expression
   | Header Level
   | MonospaceOpen
   | MonospaceClose
@@ -82,28 +83,44 @@ lexText = go
     Parsec.optionMaybe Parsec.eof >>= \case
       Just _ -> pure []
       Nothing -> do
-        toks <-
-          choice $
-            Parsec.try
-              <$> [ mathMultiline
-                  , mathInline
-                  , escape -- maths go before escape to avoid mismatch
-                  , headers
-                  , newlineToken
-                  , spaceToken
-                  , link
-                  , labeledLink
-                  , module_
-                  , anchor
-                  , numericEntity
-                  , textElement
-                  , quotes
-                  , birdTrack
-                  , other
-                  ]
+        toks <- topLevel
         rest <- go
         pure (toks <> rest)
 
+{- FOURMOLU_DISABLE -}
+  topLevel =
+    -- backtracking here so we always have a chance to try "other", the "catch-all-leave-to-parser-to-deal-with" choice
+    -- TODO: is this desirable? do we throw lexer error at all?
+    try
+      ( choice
+          -- Sorted in
+          -- - longest to shortest parse path
+          -- - highest frequency to lowest frequency (for performance?)
+          -- - more exact to more freeform (the latter can be the former but not vice versa)
+          [ spaceToken
+          , newlineToken
+
+          , try module_
+          , quotes
+          , try expression
+          , birdTrack
+
+            -- starts with "\"
+          , try mathMultiline
+          , try mathInline
+          , escape
+
+          , headers
+          , labeledLink
+          , link
+          , anchor
+          , numericEntity
+          , textElement
+          ]
+      )
+      <|> other
+{- FOURMOLU_ENABLE -}
+
 -- Tokens
 
 textElement :: Parser [LocatedToken]
@@ -143,16 +160,26 @@ delimited openP closeP openTok closeTok = asList <$> delimitedAsTuple (openTok <
   asList (a, tok, b) = [a, tok, b]
 
 delimitedNoTrailing :: Parser open -> Parser close -> Token -> Parser [LocatedToken]
-delimitedNoTrailing openP closeP openTok = asList <$> delimitedAsTuple (openTok <$ openP) (void closeP)
+delimitedNoTrailing openP closeP openTok =
+  asList <$> delimitedAsTuple (openTok <$ openP) (void closeP)
  where
   asList (a, tok, _) = [a, tok]
 
 delimitedSymmetric :: Parser a -> Token -> Token -> Parser [LocatedToken]
 delimitedSymmetric s = delimited s s
 
+--- End of line // end of file
 eol :: Parser ()
 eol = void "\n" <|> void "\r\n" <|> Parsec.eof
 
+-- Start of line // start of file
+sol :: Parser ()
+sol = do
+  p <- getPosition
+  if sourceColumn p == 1
+    then pure ()
+    else fail "Not at start of line/document"
+
 header1 :: Lexer
 header1 = delimitedNoTrailing "= " eol (Header One)
 
@@ -239,8 +266,15 @@ mathMultiline = delimited "\\[" "\\]" MathMultilineOpen MathMultilineClose
 mathInline :: Lexer
 mathInline = delimited "\\(" "\\)" MathInlineOpen MathInlineClose
 
+-- TODO: make sure this starts at column 0?
 birdTrack :: Lexer
-birdTrack = delimitedNoTrailing ">> " eol BirdTrack
+birdTrack = delimitedNoTrailing (sol <* "> ") eol BirdTrack
+
+-- TODO: also match following lines iff:
+-- they start with alphanum
+-- they're not empty
+expression :: Lexer
+expression = delimitedNoTrailing (sol <* ">>> ") eol Expression
 
 escape :: Lexer
 escape = delimitedNoTrailing "\\" eol Escape
diff --git a/test/Spec.hs b/test/Spec.hs
index 2040e2f..3db44f1 100644
--- a/test/Spec.hs
+++ b/test/Spec.hs
@@ -18,7 +18,7 @@ main = hspec $ do
     describe "minimal" do
       it "handles unicode" unicode
       it "escapes" escaping
-      it "maths" math
+      it "maths" maths
       it "anchors" anchor
       it "space chars" space
       it "bare string" someString
@@ -29,6 +29,7 @@ main = hspec $ do
       it "bird tracks" birdTracks
       it "module names" modules
       it "quotes" quotes
+      it "expressions" expressions
       it "numeric entity" numericEntity
       it "ignores nesting" ignoreNesting
 
@@ -91,8 +92,8 @@ anchor =
     `shouldLexTo` [ (1, 1, Anchor "myAnchor")
                   ]
 
-math :: IO ()
-math = do
+maths :: IO ()
+maths = do
   "\\[some math\\]"
     `shouldLexTo` [ (1, 1, MathMultilineOpen)
                   , (1, 3, Token "some math")
@@ -128,10 +129,35 @@ ignoreNesting =
                   ]
 
 birdTracks :: Expectation
-birdTracks =
-  ">> code"
+birdTracks = do
+  "> code line"
     `shouldLexTo` [ (1, 1, BirdTrack)
-                  , (1, 4, Token "code")
+                  , (1, 3, Token "code line")
+                  ]
+  " > not code"
+    `shouldLexTo` [ (1, 1, Space)
+                  , (1, 2, Token ">")
+                  , (1, 3, Space)
+                  , (1, 4, Token "not")
+                  , (1, 7, Space)
+                  , (1, 8, Token "code")
+                  ]
+
+expressions :: Expectation
+expressions = do
+  ">>> eval this"
+    `shouldLexTo` [ (1, 1, Expression)
+                  , (1, 5, Token "eval this")
+                  ]
+  " >>> not eval this"
+    `shouldLexTo` [ (1, 1, Space)
+                  , (1, 2, Token ">>>")
+                  , (1, 5, Space)
+                  , (1, 6, Token "not")
+                  , (1, 9, Space)
+                  , (1, 10, Token "eval")
+                  , (1, 14, Space)
+                  , (1, 15, Token "this")
                   ]
 
 quotes :: Expectation