diff --git a/Frames.cabal b/Frames.cabal
--- a/Frames.cabal
+++ b/Frames.cabal
@@ -1,5 +1,5 @@
 name:                Frames
-version:             0.1.6
+version:             0.1.8
 synopsis:            Data frames For working with tabular data files
 description:         User-friendly, type safe, runtime efficient tooling for
                      working with tabular data deserialized from
@@ -82,9 +82,9 @@
     build-depends: base, Frames,
                    microlens, vector, text,
                    template-haskell,
-                   pipes >= 4.1.5 && < 4.3,
-                   Chart >= 1.5 && < 1.8,
-                   Chart-diagrams >= 1.5 && < 1.8,
+                   pipes >= 4.1.5 && < 4.4,
+                   Chart >= 1.5 && < 1.9,
+                   Chart-diagrams >= 1.5 && < 1.9,
                    diagrams-rasterific >= 1.3 && < 1.4,
                    diagrams-lib >= 1.3 && < 1.4,
                    readable, containers, statistics
@@ -98,9 +98,9 @@
   if flag(demos)
     build-depends: base, Frames,
                    microlens, vector, text, template-haskell,
-                   pipes >= 4.1.5 && < 4.3,
-                   Chart >= 1.5 && < 1.8,
-                   Chart-diagrams >= 1.5 && < 1.8,
+                   pipes >= 4.1.5 && < 4.4,
+                   Chart >= 1.5 && < 1.9,
+                   Chart-diagrams >= 1.5 && < 1.9,
                    diagrams-rasterific >= 1.3 && < 1.4,
                    diagrams-lib >= 1.3 && < 1.4,
                    readable, containers, statistics
@@ -115,7 +115,7 @@
   if flag(demos)
     build-depends: base, list-t, microlens, transformers, Frames,
                    vector, text, template-haskell, ghc-prim, readable,
-                   pipes >= 4.1.5 && < 4.3
+                   pipes >= 4.1.5 && < 4.4
   hs-source-dirs: demo
   default-language: Haskell2010
   ghc-options: -O2
@@ -129,7 +129,7 @@
     build-depends: base, Frames,
                    microlens, vector, text, template-haskell, readable,
                    foldl >= 1.1.0 && < 1.3,
-                   pipes >= 4.1.5 && < 4.3
+                   pipes >= 4.1.5 && < 4.4
   hs-source-dirs: demo
   default-language: Haskell2010
 
@@ -141,7 +141,7 @@
   if flag(demos)
     build-depends:    base, Frames,
                       foldl >= 1.1.0 && < 1.3,
-                      pipes >= 4.1.5 && < 4.3
+                      pipes >= 4.1.5 && < 4.4
   hs-source-dirs:   benchmarks
   default-language: Haskell2010
   -- ghc-options:      -O2
@@ -166,7 +166,7 @@
   hs-source-dirs:   benchmarks
   main-is:          InsuranceBench.hs
   build-depends:    base, criterion, Frames, transformers,
-                    pipes >= 4.1.5 && < 4.3
+                    pipes >= 4.1.5 && < 4.4
   ghc-options:      -O2
   default-language: Haskell2010
 
diff --git a/src/Frames/CSV.hs b/src/Frames/CSV.hs
--- a/src/Frames/CSV.hs
+++ b/src/Frames/CSV.hs
@@ -1,4 +1,5 @@
 {-# LANGUAGE BangPatterns,
+             CPP,
              DataKinds,
              FlexibleInstances,
              KindSignatures,
@@ -15,18 +16,22 @@
 -- necessary types so that you can write type safe programs referring
 -- to those types.
 module Frames.CSV where
+#if __GLASGOW_HASKELL__ < 710
 import Control.Applicative ((<$>), pure, (<*>))
-import Control.Arrow (first)
+import Data.Foldable (foldMap)
+import Data.Traversable (sequenceA)
+import Data.Monoid (Monoid)
+#endif
+
+import Control.Arrow (first, second)
 import Control.Monad (MonadPlus(..), when, void)
 import Control.Monad.IO.Class
 import Data.Char (isAlpha, isAlphaNum, toLower, toUpper)
-import Data.Foldable (foldMap)
-import Data.Maybe (fromMaybe, isNothing)
-import Data.Monoid ((<>), Monoid(..))
+import Data.Maybe (isNothing)
+import Data.Monoid ((<>))
 import Data.Proxy
 import qualified Data.Text as T
 import qualified Data.Text.IO as T
-import Data.Traversable (sequenceA)
 import Data.Vinyl (RElem, Rec)
 import Data.Vinyl.TypeLevel (RIndex)
 import Frames.Col
@@ -240,11 +245,11 @@
 -- * Template Haskell
 
 -- | Generate a column type.
-recDec :: ColumnTypeable a => [(T.Text, a)] -> Q Type
+recDec :: [(T.Text, Q Type)] -> Q Type
 recDec = appT [t|Record|] . go
   where go [] = return PromotedNilT
         go ((n,t):cs) =
-          [t|($(litT $ strTyLit (T.unpack n)) :-> $(colType t)) ': $(go cs) |]
+          [t|($(litT $ strTyLit (T.unpack n)) :-> $(t)) ': $(go cs) |]
 
 -- | Massage a column name from a CSV file into a valid Haskell type
 -- identifier.
@@ -376,7 +381,7 @@
 tableType' (RowGen {..}) csvFile =
     pure . TySynD (mkName rowTypeName) [] <$>
     (runIO (readColHeaders opts csvFile) >>= recDec')
-  where recDec' = recDec :: [(T.Text, a)] -> Q Type
+  where recDec' = recDec . map (second colType) :: [(T.Text, a)] -> Q Type
         colNames' | null columnNames = Nothing
                   | otherwise = Just (map T.pack columnNames)
         opts = ParserOptions colNames' separator (RFC4180Quoting '\"')
@@ -399,7 +404,7 @@
      optsDec <- valD (varP optsName) (normalB $ lift opts) []
      colDecs <- concat <$> mapM (uncurry $ colDec (T.pack tablePrefix)) headers
      return (recTy : optsTy : optsDec : colDecs)
-  where recDec' = recDec :: [(T.Text, a)] -> Q Type
+  where recDec' = recDec . map (second colType) :: [(T.Text, a)] -> Q Type
         colNames' | null columnNames = Nothing
                   | otherwise = Just (map T.pack columnNames)
         opts = ParserOptions colNames' separator (RFC4180Quoting '\"')
@@ -423,7 +428,7 @@
      return (recTy : optsTy : optsDec : colDecs)
      -- (:) <$> (tySynD (mkName n) [] (recDec' headers))
      --     <*> (concat <$> mapM (uncurry $ colDec (T.pack prefix)) headers)
-  where recDec' = recDec :: [(T.Text, a)] -> Q Type
+  where recDec' = recDec . map (second colType) :: [(T.Text, a)] -> Q Type
         colNames' | null columnNames = Nothing
                   | otherwise = Just (map T.pack columnNames)
         opts = ParserOptions colNames' separator (RFC4180Quoting '\"')
diff --git a/src/Frames/ColumnTypeable.hs b/src/Frames/ColumnTypeable.hs
--- a/src/Frames/ColumnTypeable.hs
+++ b/src/Frames/ColumnTypeable.hs
@@ -38,6 +38,8 @@
 instance Parseable Int where
 instance Parseable Float where
 instance Parseable Double where
+  -- Some CSV's export Doubles in a format like '1,000.00', filtering out commas lets us parse those sucessfully
+  parse = fmap Definitely . fromText . T.filter (/= ',')
 instance Parseable T.Text where
 
 -- | This class relates a universe of possible column types to Haskell
diff --git a/src/Frames/ColumnUniverse.hs b/src/Frames/ColumnUniverse.hs
--- a/src/Frames/ColumnUniverse.hs
+++ b/src/Frames/ColumnUniverse.hs
@@ -91,6 +91,16 @@
 -- | We use a join semi-lattice on types for representations. The
 -- bottom of the lattice is effectively an error (we have nothing to
 -- represent), @Bool < Int@, @Int < Double@, and @forall n. n <= Text@.
+--
+-- The high-level goal here is that we will pick the "greater" of two
+-- choices in 'bestRep'. A 'Definitely' parse result is preferred over
+-- a 'Possibly' parse result. If we have two distinct 'Possibly' parse
+-- results, we give up. If we have two distinct 'Definitely' parse
+-- results, we are in dangerous waters: all data is parseable at
+-- /both/ types, so which do we default to? The defaulting choices
+-- made here are described in the previous paragraph. If there is no
+-- defaulting rule, we give up (i.e. use 'T.Text' as a
+-- representation).
 lubTypeReps :: Parsed TypeRep -> Parsed TypeRep -> Maybe Ordering
 lubTypeReps (Possibly _) (Definitely _) = Just LT
 lubTypeReps (Definitely _) (Possibly _) = Just GT