diff --git a/src/Data/Random/Generics.hs b/src/Data/Random/Generics.hs
index b856bf9..b06ae8c 100644
--- a/src/Data/Random/Generics.hs
+++ b/src/Data/Random/Generics.hs
@@ -1,4 +1,27 @@
 -- | Generic Boltzmann samplers.
+--
+-- Here, the words "/sampler/" and "/generator/" are used interchangeably.
+--
+-- Given an algebraic datatype:
+--
+-- > data A = A1 B C | A2 D
+--
+-- a Boltzmann sampler is recursively defined by choosing a constructor with
+-- some fixed distribution, and /independently/ generating values for the
+-- corresponding fields with the same method.
+--
+-- A key component is the aforementioned distribution, defined for every type
+-- such that the resulting generator produces a finite value in the end. These
+-- distributions are obtained from a precomputed object called /oracle/, which
+-- we will not describe further here.
+--
+-- Oracles depend on the target size of the generated data (except for singular
+-- samplers), and can be fairly expensive to compute repeatedly, hence some of
+-- the functions below attempt to avoid (re)computing too many of them even
+-- when the required size changes.
+--
+-- When these functions are specialized, oracles are memoized and will be
+-- reused for different sizes.
 
 module Data.Random.Generics (
   Size',
@@ -52,9 +75,6 @@ import Data.Random.Generics.Internal.Types
 -- * Main functions
 
 -- $sized
--- When these functions and their @_With@ counterparts below are specialized,
--- the numerical /oracles/ are computed once and for all, so they can be reused
--- for different sizes.
 --
 -- === Suffixes
 --
@@ -81,7 +101,7 @@ import Data.Random.Generics.Internal.Types
 --
 --     These generators filter out values whose sizes are not within some
 --     interval. In the first two sections, that interval is implicit:
---     @[(1-'epsilon')*size, (1+'epsilon')*size]@, for @'epsilon' = 0.1@.
+--     @[(1-'epsilon')*size', (1+'epsilon')*size']@, for @'epsilon' = 0.1@.
 --
 --     The generator restarts as soon as it has produced more constructors than
 --     the upper bound, this strategy is called /ceiled rejection sampling/.
diff --git a/src/Data/Random/Generics/Internal.hs b/src/Data/Random/Generics/Internal.hs
index ed48a13..6f34b6c 100644
--- a/src/Data/Random/Generics/Internal.hs
+++ b/src/Data/Random/Generics/Internal.hs
@@ -18,6 +18,7 @@ data SG r = SG
   , runSmallG :: Points -> r
   } deriving Functor
 
+-- | Number of pointing iterations.
 type Points = Int
 
 rangeSG :: SG r -> (Size, Maybe Size)
@@ -67,7 +68,7 @@ makeR aliases a = fmap (flip runRejectT) (make aliases a)
 -- | The size of a value is its number of constructors.
 --
 -- Here, however, the 'Size'' type is interpreted differently to make better
--- use of QuickCheck's size parameter, provided by the 'Test.QuickCheck.sized'
+-- use of QuickCheck's size parameter provided by the 'Test.QuickCheck.sized'
 -- combinator, so that we generate non-trivial data even at very small size
 -- values.
 --
@@ -85,8 +86,9 @@ makeR aliases a = fmap (flip runRejectT) (make aliases a)
 -- will target sizes close to @2 + delta@;
 -- the offset becomes less noticeable as @delta@ grows to infinity.
 --
--- For finite types with sizes in @[minSize, maxSize]@, the target
--- expected size is obtained from a 'Size'' in @[0, 99]@ by an affine mapping.
+-- For finite types with sizes in @[minSize, maxSize]@, the target expected
+-- size is obtained by clamping a 'Size'' to @[0, 99]@ and applying an affine
+-- mapping.
 type Size' = Int
 
 rescale :: SG r -> Size' -> Double
diff --git a/src/Data/Random/Generics/Internal/Types.hs b/src/Data/Random/Generics/Internal/Types.hs
index 30e6206..2c26fd7 100644
--- a/src/Data/Random/Generics/Internal/Types.hs
+++ b/src/Data/Random/Generics/Internal/Types.hs
@@ -151,7 +151,7 @@ class Monad m => MonadRandomLike m where
   incr :: m ()
   incr = return ()
 
-  -- | @doubleR@: generates values in @[0, upperBound)@.
+  -- | @doubleR upperBound@: generates values in @[0, upperBound]@.
   doubleR :: Double -> m Double
 
   -- | @integerR upperBound@: generates values in @[0, upperBound-1]@.