Mate scores should be indicated as 100000 + N for "mate in N moves", and -100000 - N for "mated in N moves".
The sign of N in the above should be reversed. From the viewpoint of the mover, a mate-in-4 is better than a mate-in-7 and a lose-in-3 is better than a lose-in-1.
I hate having to subtract in order to get the DTM. When I see 100006 (or 1000.06 as the GUI would display it), I see that it is a mate in 6 without further ado. A GUI that supports this convention would of course just display something like M6, #6 or mate-in-6. But this convention also guarantees reasonable display on GUIs that do not support it, once you realize you "1000." is just a funny spelling for "mate in" (and -"1000." for "mated in". Note that things like "327.57" for a mate in 6 fall in the category "unreasonable display"...
You should not read (and compare) them as if they were numerical scores. They are just text strings. I would have chosen "mate6" and "mated6" if that would not have broken display of all thinking output on all older interfaces.
It's really a manifestation of the dubious programming practice of overloading the semantics of a scalar making it more than just a scalar. It's a bad idea even though nearly all of us have done it, including me where in the original EPD specification I gave a mapping between mate/lose scores and centipawn evaluations.
There is no easy resolution which wouldn't break a lot of existing code.
Inside of Symbolic is the C++ class Analysis, an instance of which is used to store the result of a move selection process. There are a bunch of member variables: a PV move list, timers, usage data, authorship data, search termination identification, etc., -- and a score in micropawn units. Internally, everything gets used. Externally, most of what gets seen is just the PV and the score.
I'd like to mention that there is really no such thing as a -MateIn3 score. There is a LoseIn2 score which is sightly worse than a LoseIn3 score, but neither is the opposite of a MateIn3 score because it all depends on measuring the MateIn3 before or after it is played. The guy not on the move can't play a move, so he has no moves and therefore can't have any move evaluations.
Symbolic handles this by treating mate/lose scores relative to the direction of time. Moving forward in time (one ply deeper, passing a window), a MateIn3 score becomes a LoseIn2 score. Moving backward in time (towards the root, returning a result), a MateIn3 becomes a LoseIn3. At the bottom, a checkmated position has a score of Checkmated which is a synonym for LoseIn0.
The initial A/B window is not [-infinity..+infinity], but rather [LoseIn0..MateIn1+1]. Moving down one ply changes the window to [LoseIn1..MateIn1] then on the next ply to [LoseIn2..MateIn2], etc.
With the proper numerical mapping, faster mates are always greater than slower mates and slower loses are greater than faster loses.
