diff --git a/README.md b/README.md
index 353e9ad..39a0cfe 100644
--- a/README.md
+++ b/README.md
@@ -24,7 +24,7 @@ position in the second keyed alphabet, DRACONDI achieves the following
 improvements:
 
 - the initial offset is now affecting the rest of the ciphertext;
-- the keyspace size has been increased to about 225 bits;
+- the keyspace size has been increased to about 176 bits;
 - repeated plaintext digraphs no longer encrypt into the same ciphertext;
 - repeated letters in the keyphrase now affect the keyed alphabet;
 - the ciphertext is (optionally) divided into the groups of 5 that no longer
@@ -40,6 +40,22 @@ The entire DRACONDI cipher can be described in three sections: key preparation,
 encryption and decryption. Additionally, all key and message material which is
 alphabetical must first undergo the sanitization part (i.e. lowercasing it and
 removing all non-alphabetical characters not belonging to the `[a-z]` range).
+Unlike the original CONDI, all insignificant whitespace is to be removed in the
+plaintext message, and all significant whitespace is to be replaced with any
+rarely used letter, like `x`, `z` or `q`.
+
+### Numeric information encoding (optional)
+
+It is recommended, although not required (even for software implementations),
+that the numeric digits from 0 to 9 are encoded with `a` to `j` respectively
+and the numbers are framed within `xn` and `nx` digraphs. If the literal
+spelling of a number is shorter, it is preferred to this representation.
+For example, while 2 can be encoded as `xncnx`, the word `two` is just shorter
+and is preferred. However, 13 is preferred to be encoded as `xnbdnx`, as the
+word `thirteen` already is longer than this encoding. In case the length is
+the same, the general recommendation is not to use any encoding. So, as an
+example, the number 11 is just written as `eleven` as opposed to `xnbbnx` that
+has the same 6-character length. Same for 20: `twenty` instead of `xncanx`.
 
 ### Key preparation