diff --git a/CHANGELOG.md b/CHANGELOG.md
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -1,6 +1,25 @@
 # Revision history for mmzk-typeid
 
 
+## 0.7.1.2 -- 2026-09-06
+
+* Fix an out-of-bounds read in the `Storable` instance when the prefix length byte is corrupt. The length is now validated before any bytes are read, and `poke` rejects prefixes that would overflow the record.
+
+* Fix the dead prefix validation in the `Binary` and `Storable` instances. The character check was a tautology and never rejected anything; malformed serialised `TypeID`s could produce values with invalid prefixes. Both instances now also run the full prefix validation on decoding.
+
+* Fix the unsafe parsers (`unsafeParseString`, `unsafeParseText`, `unsafeParseByteString`) splitting at the first underscore while the safe parsers split at the last. A spec-valid ID with an underscored prefix (e.g. `super_user_...`) was silently parsed into a wrong prefix and a garbage UUID.
+
+* Make `parseByteString` (and the `byteString2ID` class method) total: invalid UTF-8 in the prefix now yields a `TypeIDError` instead of throwing a `UnicodeException`.
+
+* `getTime` now returns 0 for non-v7 `TypeID`/`KindID`, as its documentation always claimed, instead of the top bits of the UUID.
+
+* Reduce the OS entropy syscalls in `Data.UUID.V7` by buffering reads (256-byte chunks), and avoid spinning the CPU when the clock steps backwards during batch generation.
+
+* `Storable.poke` now errors on prefixes longer than 63 characters instead of writing past the end of the record. Only reachable via the unsafe API.
+
+* Documentation fixes: the `Binary` format description, the `README` examples, and notes on the `decorateTypeID` and custom-timestamp behaviour.
+
+
 ## 0.7.1.1 -- 2026-05-13
 
 * Support `random` 1.3.
@@ -77,7 +96,7 @@
 
 ## 0.6.2.0 -- 2024-05-28
 
-* Fix the bug where the first 32768 `TypeID`s may not of the same timestamp.
+* Fix the bug where the first 32768 `TypeID`s may not be of the same timestamp.
 
 * Test on GHC 9.8.2.
 
diff --git a/LICENSE b/LICENSE
--- a/LICENSE
+++ b/LICENSE
@@ -1,6 +1,6 @@
 MIT License
 
-Copyright (c) 2023 Yitang Chen
+Copyright (c) 2026 MMZK1526
 
 Permission is hereby granted, free of charge, to any person obtaining a copy
 of this software and associated documentation files (the "Software"), to deal
diff --git a/README.md b/README.md
--- a/README.md
+++ b/README.md
@@ -14,7 +14,7 @@
 
 It also serves as a (temporary) UUIDv7 implementation in Haskell, since there are no official ones yet.
 
-If you notice any issues or have any suggestions, please feel free to open an issue or contact me via email.
+If you notice any issues or have any suggestions, please feel free to open an issue.
 
 ## Highlights
 
@@ -144,7 +144,7 @@
 For a full list of functions on `KindID`, see [Data.KindID](https://hackage.haskell.org/package/mmzk-typeid/docs/Data-KindID.html).
 
 ### Functions with More General Types
-`TypeID` and `KindID` shares many functions with the same name and functionality. So far, we are using qualified imports to diffentiate them (*e.g* `KID.fromString` and `TID.fromString`). Alternatively, we can use the methods of `IDConv` to use the same functions for both `TypeID` and `KindID`.
+`TypeID` and `KindID` shares many functions with the same name and functionality. So far, we are using qualified imports to diffentiate them (*e.g* `KID.parseString` and `TID.parseString`). Alternatively, we can use the methods of `IDConv` to use the same functions for both `TypeID` and `KindID`.
 
 ```Haskell
 {-# LANGUAGE DataKinds #-}
@@ -167,17 +167,17 @@
   print kindID
 
   -- Parse a TypeID from string:
-  case string2ID "mmzk_01h455vb4pex5vsknk084sn02q" :: Maybe TypeID of
+  case string2ID "mmzk_01h455vb4pex5vsknk084sn02q" :: Either TypeIDError TypeID of
     Left err     -> throwIO err
     Right typeID -> print typeID
 
   -- Parse a KindID from string:
-  case string2ID "mmzk_01h455vb4pex5vsknk084sn02q" :: Maybe (KindID "mmzk") of
+  case string2ID "mmzk_01h455vb4pex5vsknk084sn02q" :: Either TypeIDError (KindID "mmzk") of
     Left err     -> throwIO err
     Right kindID -> print kindID
 
   -- Parse a KindID from string (wrong prefix):
-  case string2ID "mmzk_01h455vb4pex5vsknk084sn02q" :: Maybe (KindID "foo") of
+  case string2ID "mmzk_01h455vb4pex5vsknk084sn02q" :: Either TypeIDError (KindID "foo") of
     Left err     -> throwIO err -- Will throw here as the prefix matches not
     Right kindID -> print kindID
 ```
diff --git a/mmzk-typeid.cabal b/mmzk-typeid.cabal
--- a/mmzk-typeid.cabal
+++ b/mmzk-typeid.cabal
@@ -1,6 +1,6 @@
 cabal-version:      2.4
 name:               mmzk-typeid
-version:            0.7.1.1
+version:            0.7.1.2
 
 synopsis:           A TypeID and UUIDv7 implementation for Haskell
 description:
@@ -8,7 +8,7 @@
   .
   The specification is available at https://github.com/jetpack-io/typeid.
   .
-  This library supports generating and parsing speç-conforming 'TypeID's, with the following additional features:
+  This library supports generating and parsing spec-conforming 'TypeID's, with the following additional features:
   .
     - Batch generating 'TypeID's with the same UUIDv7 timestamp
   .
@@ -31,8 +31,8 @@
 homepage:           https://github.com/MMZK1526/mmzk-typeid
 bug-reports:        https://github.com/MMZK1526/mmzk-typeid/issues
 license:            MIT
-author:             Yitang Chen <mmzk1526@outlook.com>
-maintainer:         Yitang Chen <mmzk1526@outlook.com>
+author:             MMZK1526
+maintainer:         MMZK1526
 category:           Data, UUID, UUIDv7, TypeID
 tested-with:
     GHC == 9.4.8
@@ -173,7 +173,9 @@
         binary,
         bytestring,
         containers >=0.6 && <1,
+        directory,
         entropy,
+        filepath,
         hashable,
         hint ^>=0.9,
         hspec ^>=2.11,
diff --git a/src/Data/KindID.hs b/src/Data/KindID.hs
--- a/src/Data/KindID.hs
+++ b/src/Data/KindID.hs
@@ -1,7 +1,6 @@
 -- |
 -- Module      : Data.KindID
 -- License     : MIT
--- Maintainer  : mmzk1526@outlook.com
 -- Portability : GHC
 --
 -- Similar to "Data.TypeID", but the type is statically determined in the type
diff --git a/src/Data/KindID/Class.hs b/src/Data/KindID/Class.hs
--- a/src/Data/KindID/Class.hs
+++ b/src/Data/KindID/Class.hs
@@ -4,7 +4,6 @@
 -- |
 -- Module      : Data.KindID.Class
 -- License     : MIT
--- Maintainer  : mmzk1526@outlook.com
 -- Portability : GHC
 --
 -- This module contains the type-level mechanisms that are used to define
diff --git a/src/Data/KindID/Internal.hs b/src/Data/KindID/Internal.hs
--- a/src/Data/KindID/Internal.hs
+++ b/src/Data/KindID/Internal.hs
@@ -3,7 +3,6 @@
 -- |
 -- Module      : Data.KindID.Internal
 -- License     : MIT
--- Maintainer  : mmzk1526@outlook.com
 -- Portability : GHC
 --
 module Data.KindID.Internal where
@@ -591,6 +590,7 @@
 checkKindIDV5 :: (ToPrefix prefix, ValidPrefix (PrefixSymbol prefix))
               => KindID' 'V5 prefix -> Maybe TypeIDError
 checkKindIDV5 = TID.checkTypeIDV5 . toTypeID
+{-# INLINE checkKindIDV5 #-}
 
 -- | Convert a 'TypeID'' to a 'KindID''. If the actual prefix does not match
 -- with the expected one as defined by the type, it does not complain and
diff --git a/src/Data/KindID/Unsafe.hs b/src/Data/KindID/Unsafe.hs
--- a/src/Data/KindID/Unsafe.hs
+++ b/src/Data/KindID/Unsafe.hs
@@ -1,12 +1,11 @@
 -- |
 -- Module      : Data.KindID.Unsafe
 -- License     : MIT
--- Maintainer  : mmzk1526@outlook.com
 -- Portability : GHC
 --
 -- Unsafe 'Data.KindID.V7.KindID' functions.
 --
--- It is a re-export of "Data.TypeID.V7.Unsafe".
+-- It is a re-export of "Data.KindID.V7.Unsafe".
 --
 module Data.KindID.Unsafe
   (
diff --git a/src/Data/KindID/V1.hs b/src/Data/KindID/V1.hs
--- a/src/Data/KindID/V1.hs
+++ b/src/Data/KindID/V1.hs
@@ -1,7 +1,6 @@
 -- |
 -- Module      : Data.KindID.V1
 -- License     : MIT
--- Maintainer  : mmzk1526@outlook.com
 -- Portability : GHC
 --
 -- 'Data.KindID.V7.KindID' with 'UUID'v1.
@@ -127,6 +126,7 @@
                  . (ToPrefix prefix, ValidPrefix (PrefixSymbol prefix))
                 => ByteString -> Either TypeIDError (KindIDV1 prefix)
 parseByteString = KID.parseByteString
+{-# INLINE parseByteString #-}
 
 -- | Parse a 'KindIDV1' from its 'String' representation, throwing an error when
 -- the parsing fails. It is 'string2IDM' with concrete type.
diff --git a/src/Data/KindID/V1/Unsafe.hs b/src/Data/KindID/V1/Unsafe.hs
--- a/src/Data/KindID/V1/Unsafe.hs
+++ b/src/Data/KindID/V1/Unsafe.hs
@@ -1,7 +1,6 @@
 -- |
 -- Module      : Data.KindID.V1.Unsafe
 -- License     : MIT
--- Maintainer  : mmzk1526@outlook.com
 -- Portability : GHC
 --
 -- Unsafe 'KindIDV1' functions.
diff --git a/src/Data/KindID/V4.hs b/src/Data/KindID/V4.hs
--- a/src/Data/KindID/V4.hs
+++ b/src/Data/KindID/V4.hs
@@ -1,7 +1,6 @@
 -- |
 -- Module      : Data.KindID.V4
 -- License     : MIT
--- Maintainer  : mmzk1526@outlook.com
 -- Portability : GHC
 --
 -- 'Data.KindID.V7.KindID' with 'UUID'v4.
@@ -135,6 +134,7 @@
                  . (ToPrefix prefix, ValidPrefix (PrefixSymbol prefix))
                 => ByteString -> Either TypeIDError (KindIDV4 prefix)
 parseByteString = KID.parseByteString
+{-# INLINE parseByteString #-}
 
 -- | Parse a 'KindIDV4' from its 'String' representation, throwing an error when
 -- the parsing fails. It is 'string2IDM' with concrete type.
diff --git a/src/Data/KindID/V4/Unsafe.hs b/src/Data/KindID/V4/Unsafe.hs
--- a/src/Data/KindID/V4/Unsafe.hs
+++ b/src/Data/KindID/V4/Unsafe.hs
@@ -1,7 +1,6 @@
 -- |
 -- Module      : Data.KindID.V4.Unsafe
 -- License     : MIT
--- Maintainer  : mmzk1526@outlook.com
 -- Portability : GHC
 --
 -- Unsafe 'KindIDV4' functions.
diff --git a/src/Data/KindID/V5.hs b/src/Data/KindID/V5.hs
--- a/src/Data/KindID/V5.hs
+++ b/src/Data/KindID/V5.hs
@@ -1,7 +1,6 @@
 -- |
 -- Module      : Data.KindID.V5
 -- License     : MIT
--- Maintainer  : mmzk1526@outlook.com
 -- Portability : GHC
 --
 -- 'Data.KindID.V7.KindID' with 'UUID'v5.
@@ -128,6 +127,7 @@
                  . (ToPrefix prefix, ValidPrefix (PrefixSymbol prefix))
                 => ByteString -> Either TypeIDError (KindIDV5 prefix)
 parseByteString = KID.parseByteString
+{-# INLINE parseByteString #-}
 
 -- | Parse a 'KindIDV5' from its 'String' representation, throwing an error when
 -- the parsing fails. It is 'string2IDM' with concrete type.
diff --git a/src/Data/KindID/V5/Unsafe.hs b/src/Data/KindID/V5/Unsafe.hs
--- a/src/Data/KindID/V5/Unsafe.hs
+++ b/src/Data/KindID/V5/Unsafe.hs
@@ -1,7 +1,6 @@
 -- |
 -- Module      : Data.KindID.V5.Unsafe
 -- License     : MIT
--- Maintainer  : mmzk1526@outlook.com
 -- Portability : GHC
 --
 -- Unsafe 'KindIDV5' functions.
diff --git a/src/Data/KindID/V7.hs b/src/Data/KindID/V7.hs
--- a/src/Data/KindID/V7.hs
+++ b/src/Data/KindID/V7.hs
@@ -1,7 +1,6 @@
 -- |
 -- Module      : Data.KindID.V7
 -- License     : MIT
--- Maintainer  : mmzk1526@outlook.com
 -- Portability : GHC
 --
 -- Similar to "Data.TypeID", but the type is statically determined in the type
@@ -204,6 +203,7 @@
                  . (ToPrefix prefix, ValidPrefix (PrefixSymbol prefix))
                 => ByteString -> Either TypeIDError (KindID prefix)
 parseByteString = KID.parseByteString
+{-# INLINE parseByteString #-}
 
 -- | Parse a 'KindID' from its 'String' representation, throwing an error when
 -- the parsing fails. It is 'string2IDM' with concrete type.
diff --git a/src/Data/KindID/V7/Unsafe.hs b/src/Data/KindID/V7/Unsafe.hs
--- a/src/Data/KindID/V7/Unsafe.hs
+++ b/src/Data/KindID/V7/Unsafe.hs
@@ -1,7 +1,6 @@
 -- |
 -- Module      : Data.KindID.V7.Unsafe
 -- License     : MIT
--- Maintainer  : mmzk1526@outlook.com
 -- Portability : GHC
 --
 -- Unsafe 'KindID' functions.
diff --git a/src/Data/TypeID.hs b/src/Data/TypeID.hs
--- a/src/Data/TypeID.hs
+++ b/src/Data/TypeID.hs
@@ -1,7 +1,6 @@
 -- |
 -- Module      : Data.TypeID
 -- License     : MIT
--- Maintainer  : mmzk1526@outlook.com
 -- Portability : GHC
 --
 -- An implementation of the 'TypeID' specification:
diff --git a/src/Data/TypeID/Class.hs b/src/Data/TypeID/Class.hs
--- a/src/Data/TypeID/Class.hs
+++ b/src/Data/TypeID/Class.hs
@@ -3,7 +3,6 @@
 -- |
 -- Module      : Data.TypeID.Class
 -- License     : MIT
--- Maintainer  : mmzk1526@outlook.com
 -- Portability : GHC
 --
 -- A module with the APIs for any 'Data.TypeID.V7.TypeID'-ish identifier type.
@@ -76,7 +75,9 @@
   -- | Parse the identifier from its string representation as a lazy
   -- 'ByteString'.
   byteString2ID :: ByteString -> Either TypeIDError a
-  byteString2ID = string2ID . T.unpack . decodeUtf8 . BSL.toStrict
+  byteString2ID bs = case decodeUtf8' (BSL.toStrict bs) of
+    Left _    -> Left TypeIDErrorUUIDError
+    Right txt -> string2ID (T.unpack txt)
   {-# INLINE byteString2ID #-}
 
   -- | Pretty-print the identifier to a 'String'.
diff --git a/src/Data/TypeID/Error.hs b/src/Data/TypeID/Error.hs
--- a/src/Data/TypeID/Error.hs
+++ b/src/Data/TypeID/Error.hs
@@ -1,7 +1,6 @@
 -- |
 -- Module      : Data.TypeID.Error
 -- License     : MIT
--- Maintainer  : mmzk1526@outlook.com
 -- Portability : GHC
 --
 -- TypeID Error type.
diff --git a/src/Data/TypeID/Internal.hs b/src/Data/TypeID/Internal.hs
--- a/src/Data/TypeID/Internal.hs
+++ b/src/Data/TypeID/Internal.hs
@@ -1,7 +1,6 @@
 -- |
 -- Module      : Data.TypeID.Internal
 -- License     : MIT
--- Maintainer  : mmzk1526@outlook.com
 -- Portability : GHC
 --
 module Data.TypeID.Internal where
@@ -48,7 +47,8 @@
 -- other than v7.
 --
 --  The constructor is not exposed to the public API to prevent generating
--- invalid 'TypeID''s.
+-- invalid 'TypeID''s. Note that the 'Generic' and 'Data' instances still allow
+-- constructing arbitrary (including invalid) values.
 data TypeID' (version :: UUIDVersion) = TypeID' Text UUID
   deriving (Eq, Ord, Data, Generic)
 
@@ -93,8 +93,8 @@
 -- | Since the specification does not formulate a concrete binary format, this
 -- instance is based on the following custom format:
 --
--- * The first 16 bytes are the suffix 'UUID' encoded in base32.
--- * The next byte is the length of the prefix encoded in a byte.
+-- * The first 16 bytes are the raw big-endian suffix 'UUID'.
+-- * The next byte is the number of 5-bit groups in the encoded prefix.
 -- * The next bytes are the prefix, each letter taking 5 bits, mapping \'a\' to
 --   1 and \'z\' to 26. The underscore \'_\' is mapped to 27.
 --
@@ -119,13 +119,19 @@
   get = do
     uuid          <- get
     len           <- getWord8
+    when (len > fromIntegral maxEncodedPrefixLen) $ fail "Binary: Prefix too long"
     encodedPrefix <- separate5BitInts <$> replicateM (fromIntegral len) getWord8
-    when (length encodedPrefix > 63) do fail "Binary: Prefix too long"
-    when (any (liftM2 (&&) (< 1) (> 26)) encodedPrefix) do
+    when (any (\v -> v < 1 || v > 27) encodedPrefix) do
       fail "Binary: Invalid prefix"
     let back 27 = 95
         back a  = a + 96
-    pure $ TypeID' (decodeUtf8 . BS.pack $ fmap back encodedPrefix) uuid
+    let bs = BS.pack $ fmap back encodedPrefix
+    prefix <- case decodeUtf8' bs of
+      Right p  -> pure p
+      Left _   -> fail "Binary: Invalid prefix"
+    case checkPrefix prefix of
+      Just err -> fail $ "Binary: " ++ show err
+      Nothing  -> pure $ TypeID' prefix uuid
   {-# INLINE get #-}
 
 -- | Similar to the 'Binary' instance, but the 'UUID' is stored in host endian.
@@ -140,27 +146,36 @@
 
   peek :: Ptr (TypeID' version) -> IO (TypeID' version)
   peek ptr = do
-    uuid          <- peek (castPtr ptr :: Ptr UUID)
-    len           <- fromIntegral <$> (peekByteOff ptr 16 :: IO Word8)
+    uuid <- peek (castPtr ptr :: Ptr UUID)
+    len  <- peekByteOff ptr uuidSize :: IO Word8
+    when (len > fromIntegral maxEncodedPrefixLen) $ fail "Storable: Prefix too long"
     encodedPrefix <- separate5BitInts
-                 <$> forM [1..len] \ix -> peekByteOff @Word8 ptr (16 + ix)
-    when (length encodedPrefix > 63) $ fail "Storable: Prefix too long"
-    when (any (liftM2 (&&) (< 1) (> 26)) encodedPrefix) do
+                 <$> forM [1 .. fromIntegral len] \ix -> peekByteOff @Word8 ptr (uuidSize + ix)
+    when (any (\v -> v < 1 || v > 27) encodedPrefix) do
       fail "Storable: Invalid prefix"
     let back 27 = 95
         back a  = a + 96
-    pure $ TypeID' (decodeUtf8 . BS.pack $ fmap back encodedPrefix) uuid
+    let bs = BS.pack $ fmap back encodedPrefix
+    prefix <- case decodeUtf8' bs of
+      Right p  -> pure p
+      Left _   -> fail "Storable: Invalid prefix"
+    case checkPrefix prefix of
+      Just err -> fail $ "Storable: " ++ show err
+      Nothing  -> pure $ TypeID' prefix uuid
   {-# INLINE peek #-}
 
   poke :: Ptr (TypeID' version) -> TypeID' version -> IO ()
   poke ptr (TypeID' prefix uuid) = do
-    poke (castPtr ptr) uuid
     let fore 95 = 27
         fore a  = a - 96
     let encodedPrefix = concat5BitInts . fmap fore . BS.unpack
                       $ encodeUtf8 prefix
-    pokeByteOff @Word8 ptr 16 (fromIntegral $ length encodedPrefix)
-    zipWithM_ (pokeByteOff ptr . (+ 16)) [1..] encodedPrefix
+    -- A prefix that is too long would overflow the fixed 60-byte record.
+    when (T.length prefix > 63 || length encodedPrefix > maxEncodedPrefixLen) $
+      error "Storable: Prefix too long"
+    poke (castPtr ptr) uuid
+    pokeByteOff @Word8 ptr uuidSize (fromIntegral $ length encodedPrefix)
+    zipWithM_ (pokeByteOff ptr . (+ uuidSize)) [1..] encodedPrefix
   {-# INLINE poke #-}
 
 instance Hashable (TypeID' version) where
@@ -180,7 +195,9 @@
   {-# INLINE getUUID #-}
 
   getTime :: TypeID' version -> Word64
-  getTime = V7.getTime . getUUID
+  getTime tid@(TypeID' _ uuid)
+    | validateWithVersion uuid V7 = V7.getTime uuid
+    | otherwise                   = 0
   {-# INLINE getTime #-}
 
 -- | Conversion between 'TypeID'' and 'String'/'Text'/'ByteString'.
@@ -439,6 +456,9 @@
 {-# INLINE genTypeIDV5 #-}
 
 -- | Obtain a 'TypeID'' from a prefix and a 'UUID'.
+--
+-- Note: the 'UUID' is not validated against the version of the 'TypeID''.
+-- Use 'checkID' if it needs to be validated.
 decorateTypeID :: Text -> UUID -> Either TypeIDError (TypeID' version)
 decorateTypeID prefix uuid = case checkPrefix prefix of
   Nothing  -> Right $ TypeID' prefix uuid
@@ -515,7 +535,8 @@
   (_, Just ("", _))          -> Left TypeIDExtraSeparator
   (_, Nothing)               -> TypeID' "" <$> decodeUUID bs
   (suffix, Just (prefix, _)) -> do
-    let prefix' = decodeUtf8 $ BSL.toStrict prefix
+    prefix' <- either (const $ Left TypeIDErrorUUIDError) Right
+               $ decodeUtf8' (BSL.toStrict prefix)
     case checkPrefix prefix' of
       Nothing  -> TypeID' prefix' <$> decodeUUID suffix
       Just err -> Left err
@@ -706,29 +727,32 @@
 -- | Parse a 'TypeID'' from its 'String' representation, but crashes when
 -- parsing fails.
 unsafeParseString :: String -> TypeID' version
-unsafeParseString str = case span (/= '_') str of
+unsafeParseString str = case spanEnd (/= '_') str of
   (_, "")              -> TypeID' "" . unsafeDecodeUUID $ fromString str
-  (prefix, _ : suffix) -> TypeID' (T.pack prefix)
+  (suffix, "_")        -> TypeID' "" . unsafeDecodeUUID $ fromString suffix
+  (suffix, prefix)     -> TypeID' (T.pack $ init prefix)
                         . unsafeDecodeUUID $ fromString suffix
 {-# INLINE unsafeParseString #-}
 
 -- | Parse a 'TypeID'' from its string representation as a strict 'Text', but
 -- crashes when parsing fails.
 unsafeParseText :: Text -> TypeID' version
-unsafeParseText text = case second T.uncons $ T.span (/= '_') text of
+unsafeParseText text = case second T.unsnoc . swap . runIdentity
+                  $ T.spanEndM (pure . (/= '_')) text of
   (_, Nothing)               -> TypeID' "" . unsafeDecodeUUID
                               . BSL.fromStrict $ encodeUtf8 text
-  (prefix, Just (_, suffix)) -> TypeID' prefix . unsafeDecodeUUID
+  (suffix, Just (prefix, _)) -> TypeID' prefix . unsafeDecodeUUID
                               . BSL.fromStrict . encodeUtf8 $ suffix
 {-# INLINE unsafeParseText #-}
 
 -- | Parse a 'TypeID'' from its string representation as a lazy 'ByteString',
 -- but crashes when parsing fails.
 unsafeParseByteString :: ByteString -> TypeID' version
-unsafeParseByteString bs = case second BSL.uncons $ BSL.span (/= 95) bs of
+unsafeParseByteString bs = case second BSL.unsnoc . swap $ BSL.spanEnd (/= 95) bs of
   (_, Nothing)               -> TypeID' "" $ unsafeDecodeUUID bs
-  (prefix, Just (_, suffix)) -> TypeID' (decodeUtf8 $ BSL.toStrict prefix)
-                              . unsafeDecodeUUID $ suffix
+  (suffix, Just (prefix, _)) ->
+    let prefix' = either (error . show) id . decodeUtf8' $ BSL.toStrict prefix
+    in  TypeID' prefix' . unsafeDecodeUUID $ suffix
 {-# INLINE unsafeParseByteString #-}
 
 -- | A helper for generating 'UUID'v1.
@@ -738,6 +762,17 @@
 
 -- Helpers
 
+-- | The maximum number of bytes needed to 5-bit-encode a 63-character prefix,
+-- i.e. @ceil(63 * 5 / 8)@.
+maxEncodedPrefixLen :: Int
+maxEncodedPrefixLen = 40
+{-# INLINE maxEncodedPrefixLen #-}
+
+-- | The size in bytes of the 'UUID' at the start of the 'Storable' record.
+uuidSize :: Int
+uuidSize = 16
+{-# INLINE uuidSize #-}
+
 concat5BitInts :: [Word8] -> [Word8]
 concat5BitInts
   = reverse . toBytes
@@ -810,7 +845,7 @@
   writeArray dest 8 $ ((base32Table ! (bs `BSL.index` 13)) `shiftL` 4) .|. ((base32Table ! (bs `BSL.index` 14)) `shiftR` 1)
   writeArray dest 9 $ ((base32Table ! (bs `BSL.index` 14)) `shiftL` 7) .|. ((base32Table ! (bs `BSL.index` 15)) `shiftL` 2) .|. ((base32Table ! (bs `BSL.index` 16)) `shiftR` 3)
   writeArray dest 10 $ ((base32Table ! (bs `BSL.index` 16)) `shiftL` 5) .|. (base32Table ! (bs `BSL.index` 17))
-  writeArray dest 11 $ ((base32Table ! (bs `BSL.index` 18)) `shiftL` 3) .|. (base32Table ! (bs `BSL.index` 19)) `shiftR` 2
+  writeArray dest 11 $ ((base32Table ! (bs `BSL.index` 18)) `shiftL` 3) .|. ((base32Table ! (bs `BSL.index` 19)) `shiftR` 2)
   writeArray dest 12 $ ((base32Table ! (bs `BSL.index` 19)) `shiftL` 6) .|. ((base32Table ! (bs `BSL.index` 20)) `shiftL` 1) .|. ((base32Table ! (bs `BSL.index` 21)) `shiftR` 4)
   writeArray dest 13 $ ((base32Table ! (bs `BSL.index` 21)) `shiftL` 4) .|. ((base32Table ! (bs `BSL.index` 22)) `shiftR` 1)
   writeArray dest 14 $ ((base32Table ! (bs `BSL.index` 22)) `shiftL` 7) .|. ((base32Table ! (bs `BSL.index` 23)) `shiftL` 2) .|. ((base32Table ! (bs `BSL.index` 24)) `shiftR` 3)
diff --git a/src/Data/TypeID/Unsafe.hs b/src/Data/TypeID/Unsafe.hs
--- a/src/Data/TypeID/Unsafe.hs
+++ b/src/Data/TypeID/Unsafe.hs
@@ -1,7 +1,6 @@
 -- |
 -- Module      : Data.TypeID.Unsafe
 -- License     : MIT
--- Maintainer  : mmzk1526@outlook.com
 -- Portability : GHC
 --
 -- Unsafe 'Data.TypeID.V7.TypeID' functions.
diff --git a/src/Data/TypeID/V1.hs b/src/Data/TypeID/V1.hs
--- a/src/Data/TypeID/V1.hs
+++ b/src/Data/TypeID/V1.hs
@@ -1,7 +1,6 @@
 -- |
 -- Module      : Data.TypeID.V1
 -- License     : MIT
--- Maintainer  : mmzk1526@outlook.com
 -- Portability : GHC
 --
 -- 'Data.TypeID.V7.TypeID' with 'UUID'v1.
diff --git a/src/Data/TypeID/V1/Unsafe.hs b/src/Data/TypeID/V1/Unsafe.hs
--- a/src/Data/TypeID/V1/Unsafe.hs
+++ b/src/Data/TypeID/V1/Unsafe.hs
@@ -1,7 +1,6 @@
 -- |
 -- Module      : Data.TypeID.V1.Unsafe
 -- License     : MIT
--- Maintainer  : mmzk1526@outlook.com
 -- Portability : GHC
 --
 -- Unsafe 'TypeIDV1' functions.
diff --git a/src/Data/TypeID/V4.hs b/src/Data/TypeID/V4.hs
--- a/src/Data/TypeID/V4.hs
+++ b/src/Data/TypeID/V4.hs
@@ -1,7 +1,6 @@
 -- |
 -- Module      : Data.TypeID.V4
 -- License     : MIT
--- Maintainer  : mmzk1526@outlook.com
 -- Portability : GHC
 --
 -- 'Data.TypeID.V7.TypeID' with 'UUID'v4.
diff --git a/src/Data/TypeID/V4/Unsafe.hs b/src/Data/TypeID/V4/Unsafe.hs
--- a/src/Data/TypeID/V4/Unsafe.hs
+++ b/src/Data/TypeID/V4/Unsafe.hs
@@ -1,7 +1,6 @@
 -- |
 -- Module      : Data.TypeID.V4.Unsafe
 -- License     : MIT
--- Maintainer  : mmzk1526@outlook.com
 -- Portability : GHC
 --
 -- Unsafe 'TypeIDV4' functions.
diff --git a/src/Data/TypeID/V5.hs b/src/Data/TypeID/V5.hs
--- a/src/Data/TypeID/V5.hs
+++ b/src/Data/TypeID/V5.hs
@@ -1,7 +1,6 @@
 -- |
 -- Module      : Data.TypeID.V5
 -- License     : MIT
--- Maintainer  : mmzk1526@outlook.com
 -- Portability : GHC
 --
 -- 'Data.TypeID.V7.TypeID' with 'UUID'v5.
diff --git a/src/Data/TypeID/V5/Unsafe.hs b/src/Data/TypeID/V5/Unsafe.hs
--- a/src/Data/TypeID/V5/Unsafe.hs
+++ b/src/Data/TypeID/V5/Unsafe.hs
@@ -1,7 +1,6 @@
 -- |
 -- Module      : Data.TypeID.V5.Unsafe
 -- License     : MIT
--- Maintainer  : mmzk1526@outlook.com
 -- Portability : GHC
 --
 -- Unsafe 'TypeIDV5' functions.
diff --git a/src/Data/TypeID/V7.hs b/src/Data/TypeID/V7.hs
--- a/src/Data/TypeID/V7.hs
+++ b/src/Data/TypeID/V7.hs
@@ -1,7 +1,6 @@
 -- |
 -- Module      : Data.TypeID.V7
 -- License     : MIT
--- Maintainer  : mmzk1526@outlook.com
 -- Portability : GHC
 --
 -- An implementation of the 'TypeID' specification:
diff --git a/src/Data/TypeID/V7/Unsafe.hs b/src/Data/TypeID/V7/Unsafe.hs
--- a/src/Data/TypeID/V7/Unsafe.hs
+++ b/src/Data/TypeID/V7/Unsafe.hs
@@ -1,7 +1,6 @@
 -- |
 -- Module      : Data.TypeID.V7.Unsafe
 -- License     : MIT
--- Maintainer  : mmzk1526@outlook.com
 -- Portability : GHC
 --
 -- Unsafe 'TypeID' functions.
diff --git a/src/Data/UUID/V7.hs b/src/Data/UUID/V7.hs
--- a/src/Data/UUID/V7.hs
+++ b/src/Data/UUID/V7.hs
@@ -1,7 +1,6 @@
 -- |
 -- Module      : Data.UUID.V7
 -- License     : MIT
--- Maintainer  : mmzk1526@outlook.com
 -- Portability : GHC
 --
 -- UUIDv7 implementation.
@@ -29,12 +28,14 @@
   , getEpochMilli
   ) where
 
+import           Control.Concurrent
 import           Control.Monad
 import           Control.Monad.IO.Class
 import           Data.Binary
 import           Data.Binary.Get
 import           Data.Binary.Put
 import           Data.Bits
+import qualified Data.ByteString as BS
 import qualified Data.ByteString.Lazy as BSL
 import           Data.IORef
 import           Data.Time.Clock.POSIX
@@ -67,14 +68,16 @@
 -- It is guaranteed that the first 32768 t'UUID's are generated at the same
 -- timestamp.
 genUUIDs :: MonadIO m => Word16 -> m [UUID]
-genUUIDs = liftIO . go True
+genUUIDs n = liftIO $ go True n Nothing
   where
-    go _ 0            = pure []
-    go mustSameTime n = do
-      timestamp   <- getEpochMilli
+    go _ 0 _          = pure []
+    go mustSameTime n mEntropy16 = do
+      timestamp <- getEpochMilli
       -- We set the first bit of the entropy to 0 to ensure that there's enough
       -- room for incrementing the sequence number.
-      entropy16   <- (.&. 0x7FFF) <$> getEntropyWord16
+      entropy16 <- case mEntropy16 of
+        Just e  -> pure e
+        Nothing -> (.&. 0x7FFF) <$> getEntropyWord16
       -- Calculate the maximum number of slots we can use for the current
       -- timestamp before the sequence number overflows.
       let getMaxSlots num seqNo = if 0xFFFF - seqNo < num
@@ -83,16 +86,16 @@
             else (num, seqNo + num)
       -- Get the sequence number corresponding to the current timestamp and the
       -- number of UUIDs we can generate.
-      (n', seqNo) <- atomicModifyIORef __state__ \(ts, seqNo) -> if
+      (n', seqNo) <- atomicModifyIORef' __state__ \(ts, seqNo) -> if
         | ts < timestamp -> let (n', entropy16') = getMaxSlots n entropy16
                             in  ((timestamp, entropy16'), (n', entropy16 + 1))
         | ts > timestamp -> ((ts, seqNo), (0, 0))
         | otherwise      -> let (n', entropy16') = getMaxSlots n seqNo
                             in  ((timestamp, entropy16'), (n', seqNo + 1))
       -- If we can't generate any UUIDs, we try again, hoping that the timestamp
-      -- has changed.
+      -- has changed. The threadDelay avoids pegging the CPU on clock step-backs.
       if n' == 0
-        then go mustSameTime n
+        then threadDelay 100 >> go mustSameTime n (Just entropy16)
         else do
           uuids <- forM [0..(n' - 1)] $ \curN -> do
             entropy64 <- getEntropyWord64
@@ -103,7 +106,7 @@
             pure . uncurry UUID $ runGet (join (liftM2 (,)) getWord64be) bs
           if n' == n
             then pure uuids
-            else (uuids ++) <$> go False (n - n')
+            else (uuids ++) <$> go False (n - n') Nothing
 
 -- | Generate a t'UUID'v7 with a custom timestamp (milliseconds since Unix
 -- epoch).
@@ -113,6 +116,8 @@
 --
 -- Note: a future timestamp will produce a valid t'UUID' that nonetheless
 -- fails 'validateWithTime'.
+--
+-- Note: a timestamp wider than the 48-bit time field is truncated.
 genUUIDWithTime :: MonadIO m => Word64 -> m UUID
 genUUIDWithTime ts = head <$> genUUIDsWithTime ts 1
 {-# INLINE genUUIDWithTime #-}
@@ -126,6 +131,8 @@
 --
 -- Note: a future timestamp will produce a valid t'UUID' that nonetheless
 -- fails 'validateWithTime'.
+--
+-- Note: a timestamp wider than the 48-bit time field is truncated.
 genUUIDWithTime' :: MonadIO m => Word64 -> m UUID
 genUUIDWithTime' timestamp = do
   entropy16 <- getEntropyWord16
@@ -153,6 +160,8 @@
 --
 -- Note: a future timestamp will produce valid t'UUID's that nonetheless
 -- fail 'validateWithTime'.
+--
+-- Note: a timestamp wider than the 48-bit time field is truncated.
 genUUIDsWithTime :: MonadIO m => Word64 -> Word16 -> m [UUID]
 genUUIDsWithTime timestamp = liftIO . go
   where
@@ -200,7 +209,7 @@
 -- | The global mutable state of (timestamp, sequence number).
 --
 -- The \"NOINLINE\" pragma is IMPORTANT! The logic would be flawed if it is
--- is inlined by its definition.
+-- is inlined by its definition. A @fork()@'ed child inherits this state.
 __state__ :: IORef (Word64, Word16)
 __state__ = unsafePerformIO (newIORef (0, 0))
 {-# NOINLINE __state__ #-}
@@ -250,14 +259,45 @@
   in (b3, b2, b1, b0)
 {-# INLINE splitWord64ToWord16s #-}
 
+-- | The global entropy buffer, refilled from the OS CSPRNG in chunks to avoid
+-- one syscall per t'UUID'.
+--
+-- The \"NOINLINE\" pragma is IMPORTANT! The logic would be flawed if it is
+-- is inlined by its definition. A @fork()@'ed child inherits this buffer.
+__entropy__ :: IORef [Word8]
+__entropy__ = unsafePerformIO (newIORef [])
+{-# NOINLINE __entropy__ #-}
+
+-- | The maximum size of a single OS entropy request. macOS caps
+-- @getentropy()@ at 256 bytes.
+maxEntropyRequestSize :: Int
+maxEntropyRequestSize = 256
+{-# INLINE maxEntropyRequestSize #-}
+
+-- | Draw @n@ bytes of entropy from the shared buffer, refilling it from the
+-- OS in chunks of 'maxEntropyRequestSize' when exhausted.
+getEntropyBytes :: Int -> IO [Word8]
+getEntropyBytes n = do
+  taken <- atomicModifyIORef' __entropy__ \buffer ->
+    if length buffer >= n
+      then (drop n buffer, Just (take n buffer))
+      else (buffer, Nothing)
+  case taken of
+    Just bytes -> pure bytes
+    Nothing    -> do
+      chunk <- BS.unpack <$> getEntropy (max n maxEntropyRequestSize)
+      atomicModifyIORef' __entropy__ \buffer ->
+        let newBuffer = buffer ++ chunk
+        in  (drop n newBuffer, take n newBuffer)
+
 getEntropyWord16 :: MonadIO m => m Word16
 getEntropyWord16 = liftIO do
-  bs <- BSL.fromStrict <$> getEntropy 2
+  bs <- BSL.fromStrict . BS.pack <$> getEntropyBytes 2
   pure $ runGet getWord16host bs
 {-# INLINE getEntropyWord16 #-}
 
 getEntropyWord64 :: MonadIO m => m Word64
 getEntropyWord64 = liftIO do
-  bs <- BSL.fromStrict <$> getEntropy 8
+  bs <- BSL.fromStrict . BS.pack <$> getEntropyBytes 8
   pure $ runGet getWord64host bs
 {-# INLINE getEntropyWord64 #-}
diff --git a/src/Data/UUID/Versions.hs b/src/Data/UUID/Versions.hs
--- a/src/Data/UUID/Versions.hs
+++ b/src/Data/UUID/Versions.hs
@@ -1,7 +1,6 @@
 -- |
 -- Module      : Data.UUID.Versions
 -- License     : MIT
--- Maintainer  : mmzk1526@outlook.com
 -- Portability : GHC
 --
 -- Supported t'UUID' versions for 'Data.TypeID.TypeID''.
diff --git a/test/Spec.hs b/test/Spec.hs
--- a/test/Spec.hs
+++ b/test/Spec.hs
@@ -6,6 +6,7 @@
 {-# OPTIONS_GHC -Wno-x-partial #-}
 #endif
 
+import           Control.Exception
 import           Control.Monad
 import           Data.Aeson
 import           Data.Binary (get, put)
@@ -17,6 +18,7 @@
 import           Data.KindID.V5 (KindIDV5)
 import           Data.KindID
 import           Data.KindID.Class
+import           Data.List (find, isInfixOf)
 import           Data.Map (Map)
 import qualified Data.Map as M
 import           Data.String
@@ -24,6 +26,7 @@
 import qualified Data.Text as T
 import           Data.Text.Encoding
 import           Data.TypeID
+import qualified Data.TypeID as TID
 import           Data.TypeID.Unsafe
 import           Data.TypeID.V1 (TypeIDV1)
 import           Data.TypeID.V4 (TypeIDV4)
@@ -87,10 +90,12 @@
     runStmt $ "foo :: ValidPrefix " <> show str <> " => () <- return ()"
     eval "foo"
   case result of
-    Left (WontCompile [e]) -> head (lines (errMsg e)) `shouldBe` expectedErr
-    Left (WontCompile _)   -> fail "Unexpected number of type errors!"
-    Left _                 -> fail "Impossible: cannot interpret!"
-    Right _                -> fail "Unexpected success!"
+    Left (WontCompile es) -> case find (isInfixOf expectedErr . errMsg) es of
+      Just _  -> pure ()
+      Nothing -> expectationFailure $ "Expected error containing "
+                    ++ show expectedErr ++ " but got: " ++ unlines (map errMsg es)
+    Left _                -> fail "Impossible: cannot interpret!"
+    Right _               -> fail "Unexpected success!"
 
 withCheck :: HasCallStack => (IDConv a, IDGen a) => IO a -> IO a
 withCheck action = do
@@ -120,7 +125,7 @@
   v5Test
 
 typeLevelTest :: Spec
-typeLevelTest = describe "Reject malformed KindID previx at compile time" do
+typeLevelTest = describe "Reject malformed KindID prefix at compile time" do
   it "rejects invalid alphabet" do
     prefixHasExpectedError "sZb" "The prefix \"sZb\" contains invalid character 'Z'!"
   it "rejects single underscore" do
@@ -367,6 +372,32 @@
         kid'' `shouldBe` kid'
         free ptr
 
+  describe "Unsafe parsing" do
+    let underscoredPrefixes = ["super_user", "user_post", "a_b_c"]
+    it "unsafeParseString roundtrips with the safe parsers on underscored prefixes" do
+      forM_ underscoredPrefixes \pref -> do
+        tid <- withCheck $ genID @TypeID pref
+        let str = TID.toString tid
+        unsafeParseString str `shouldBe` tid
+        Right tid `shouldBe` string2ID str
+    it "unsafeParseText roundtrips with the safe parsers on underscored prefixes" do
+      forM_ underscoredPrefixes \pref -> do
+        tid <- withCheck $ genID @TypeID pref
+        let str = T.pack $ TID.toString tid
+        unsafeParseText str `shouldBe` tid
+        Right tid `shouldBe` text2ID str
+    it "unsafeParseByteString roundtrips with the safe parsers on underscored prefixes" do
+      forM_ underscoredPrefixes \pref -> do
+        tid <- withCheck $ genID @TypeID pref
+        let bs = BSL.fromStrict . encodeUtf8 . T.pack $ TID.toString tid
+        unsafeParseByteString bs `shouldBe` tid
+        Right tid `shouldBe` byteString2ID bs
+    it "unsafe parsers crash on too-short suffix" do
+      -- 'show' forces deep evaluation of the lazy UUID fields
+      evaluate (show (unsafeParseString "no" :: TypeID)) `shouldThrow` anyException
+      evaluate (show (unsafeParseText "no" :: TypeID)) `shouldThrow` anyException
+      evaluate (show (unsafeParseByteString "no" :: TypeID)) `shouldThrow` anyException
+
 v7WithTimeTest :: Spec
 v7WithTimeTest = do
   let testTimestamp = 1234567890123 :: Word64
@@ -467,7 +498,7 @@
       tid <- withCheck $ genID @TypeIDV1 ""
       getPrefix tid `shouldBe` ""
       validateWithVersion (getUUID tid) V1 `shouldBe` True
-    it "can generate TypeIDV1 with insecure UUIDv4" do
+    it "can generate TypeIDV1 with stateless UUIDv7 generator" do
       tid   <- withCheck $ genID' @TypeIDV1 "mmzk"
       getPrefix tid `shouldBe` "mmzk"
       validateWithVersion (getUUID tid) V1 `shouldBe` True
@@ -476,11 +507,11 @@
         Left err  -> expectationFailure $ "Parse error: " ++ show err
         Right tid -> getPrefix tid `shouldBe` "mmzk"
     it "can parse TypeIDV1 from Text" do
-      case text2ID @TypeID "mmzk_00041061050r3gg28a1c60t3gf" of
+      case text2ID @TypeIDV1 "mmzk_00041061050r3gg28a1c60t3gf" of
         Left err  -> expectationFailure $ "Parse error: " ++ show err
         Right tid -> getPrefix tid `shouldBe` "mmzk"
     it "can parse TypeIDV1 from ByteString" do
-      case byteString2ID @TypeID "mmzk_00041061050r3gg28a1c60t3gf" of
+      case byteString2ID @TypeIDV1 "mmzk_00041061050r3gg28a1c60t3gf" of
         Left err  -> expectationFailure $ "Parse error: " ++ show err
         Right tid -> getPrefix tid `shouldBe` "mmzk"
 
@@ -669,11 +700,11 @@
         Left err  -> expectationFailure $ "Parse error: " ++ show err
         Right tid -> getPrefix tid `shouldBe` "mmzk"
     it "can parse TypeIDV4 from Text" do
-      case text2ID @TypeID "mmzk_00041061050r3gg28a1c60t3gf" of
+      case text2ID @TypeIDV4 "mmzk_00041061050r3gg28a1c60t3gf" of
         Left err  -> expectationFailure $ "Parse error: " ++ show err
         Right tid -> getPrefix tid `shouldBe` "mmzk"
     it "can parse TypeIDV4 from ByteString" do
-      case byteString2ID @TypeID "mmzk_00041061050r3gg28a1c60t3gf" of
+      case byteString2ID @TypeIDV4 "mmzk_00041061050r3gg28a1c60t3gf" of
         Left err  -> expectationFailure $ "Parse error: " ++ show err
         Right tid -> getPrefix tid `shouldBe` "mmzk"