Is it a bug?

Accuracy: 0% Bullshit: probably 100%

Is it the fact that the " " and "-" are in difference Unicode subcategories (punctuation vs. separator)?

it's not a bug i guess...

http://msdn.microsoft.com/library/default.asp?url=/library/en-us/winui/winui/windowsuserinterface/resources/strings/stringreference/stringfunctions/comparestring.asp

Typically, strings are compared using what is called a "word sort" technique. In a word sort, all punctuation marks and other nonalphanumeric characters, except for the hyphen and the apostrophe, come before any alphanumeric character. The hyphen and the apostrophe are treated differently than the other nonalphanumeric symbols, in order to ensure that words such as "coop" and "co-op" stay together within a sorted list.

If the SORT_STRINGSORT flag is specified, strings are compared using what is called a "string sort" technique. In a string sort, the hyphen and apostrophe are treated just like any other nonalphanumeric symbols. Their positions in the collating sequence are before the alphanumeric symbols.

A self-consistent, reproducible result from a localised collation function is a bug if and only if it conflicts with the reasonable expectations of the users of that locale in the context of the collation. If all the users disagree, you have a bug. If a substantial population of users disagree then in fact you have one or more new groups who might need a new locale.

You'll need some native English speaking Americans. My guess is that they don't care either way, and so the current collation is not a bug.

There's no bug.  This is a documented feature called "word sort".  To avoid it, use SORT_STRINGSORT in dwCmpFlags.

http://tinyurl.com/t4k4
Typically, strings are compared using what is called a "word sort" technique. In a word sort, all punctuation marks and other nonalphanumeric characters, except for the hyphen and the apostrophe, come before any alphanumeric character. The hyphen and the apostrophe are treated differently than the other nonalphanumeric symbols, in order to ensure that words such as "coop" and "co-op" stay together within a sorted list.

"Hello-Bob" is word-sorted as "HelloBob", which comes AFTER "Hello Bob".
"-" is word-sorted as "", which comes BEFORE " ".

Some sort key weights...

"": 1 1 1 1
" ": 7 2 1 1 1 1
"-": 1 1 1 1 128 7 6 130
"ab": 14 2 14 9 1 1 1 1
"a b": 14 2 7 2 14 9 1 1 1 1
"a-b": 14 2 14 9 1 1 1 1 128 11 6 130

"a" is 14 2
"b" is 14 9
" " is 7 2

Note "a-b" sorts as "ab" plus a little bit (128 11 6 130)
Note "-" sorts as "" plus that same little bit (128 11 6 130)

The hyphen-and-apostrophe rule makes me think this is for sorting names... so Jo-Anne comes next to Joanne, and O'Brady comes next to OBrady.

Oops in my previous comment... the "little bit" for "-" is 128 7 6 130, which is not quite the same as the "little bit" for "a-b" (128 11 6 130) (perhaps because of where in the string the hyphen occurs?)

There are primary weights, secondary weights, and tertiary weights:

primary weights: a < b
secondary weights: a < A
tertiary weights: "" < ' < -

These are visible in the sort keys:
a: (0e 02) 01 01 [] 01 01 {}
a': (0e 02) 01 01 [] 01 01 {80 0b 06 80}
a-: (0e 02) 01 01 [] 01 01 {80 0b 06 82}
A: (0e 02) 01 01 [12] 01 01 {}
A': (0e 02) 01 01 [12] 01 01 {80 0b 06 80}
A-: (0e 02) 01 01 [12] 01 01 {80 0b 06 82}

(primary weight) 01 01 [secondary weight] 01 01 {tertiary weight}

The x0101 is used as a boundary between weights because (guess mode = ON)
* a x00 would terminate the byte array and
* x0101 is LESS than any weight -- which is necessary to make shorter strings sort as "less than" longer strings with the same prefix (a < a')

a: (0e 02) 01 [] 01 {} 01 ? 01 <>
a': (0e 02) 01 [] 01 {} 01 ? 01 <80 0b 06 80>
á: (0e 02) 01 [0e] 01 {} 01 ? 01 <>
A: (0e 02) 01 [] 01 {12} 01 ? 01 <>
b: (0e 09) 01 [] 01 {} 01 ? 01 <>

(primary weight)
01
[secondary weight]
01
{tertiary weight}
01
?missing quaternary weight?
01
<quaternary or quintary weight>

primary weight: a < b
secondary weight: a < A
tertiary weight: a < á
quaternary weight: a < a' (in a WORD sort)

Hmmm... what goes where the question mark is?

Oops, I switched secondary and tertiary weight... in my previous comment: A < á.

> what goes where the question mark is:
Ah, here we go:
http://blogs.msdn.com/michkap/archive/2005/06/15/429279.aspx

Primary weight is "UW"
Secondary weight is "DW"
Tertiary weight is "CW"
Quaternary weight is "SW"
Quintary weight is unnamed (the sort key in that post has nothing between the last 01 and the final 00)

Alas, SW happens to be empty.

UW... I got nothing.
DW... "diacritic weight", n'est-pas?
CW... "case weight", perhaps?
SW... "something weight" (but WHAT?)
(anonymous)... "word sort weight", if you will

Heh, re-reading the comments to that post, I missed this gem:
"some implementations assume the byte array is NULL terminated since it is doc-ed as such"

OK, found UW and SW:
http://download.microsoft.com/download/d/1/8/d18fc51c-1f18-436e-8e1e-312a47353a77/DBA319.ppt

Slide 24:
[all Unicode] 0x01 [all Diacritic] 0x01 [all Case] 0x01 [all Special] 0x01 0x0

So U is Unicode, and S is Special.

Slide 26 has examples of strings with SW... but I can't copy them :'(

Here's a screenshot (from PowerPoint Viewer 2003 on Windows 2000)

http://www.geocities.com/mvaneerde/sw-example.gif