Syzygy tablebase returning incorrect results?

For years I have used Nalimov tablebases to help me analyze endgame problems.  They have always worked well for this purpose, but they slow down the engine substantially during analysis of full games and they take up a lot of hard drive space, limiting me to the 3 piece through 5 piece set.

I read that the Syzygy tablebases use substantially less hard drive space and don't slow down the engine like the Nalimov tablebases do.  So I recently installed the Turbo 5 Syzygy tablebases put out by chessbase.

I kept the Nalimov tablebases installed so that I could compare results with Syzygy tablebases.  The first thing I noticed is that Syzygy tablebases don't return the number of moves to checkmate like the Nalimov tablebases do.  I researched this on the Chess.com forum and found a posting where an experienced user of the Syzygy product explained that this newer tablebase does not provide that information, so it is normal for the results not to display this information.

I decided that I don't necessarily need to know the number of moves to checkmate as long as the tablebase always returns the best move for the top result, followed by the next best move, and so on.  And it most cases it does. 

However, I have discovered that sometimes I am getting less optimal moves as the top result with the Syzygy tablebases, leading to a greater number of moves to checkmate.  This is easy to verify by comparing the same problem using the Nalimov tablebases.

For this testing I am using all chessbase products.  I have the Fritz 16 interface running the Stockfish 9 64 bit engine using the Syzygy tablebases, and a Fritz 14 interface running Stockfish 9 using the Nalimov tablebases.  FYI, I have tried using running other engines on these same interfaces but it made no difference in the results.  I also tried running the Fritz 14 interface with the Syzygy tablebases and it also made no difference.

Example 1 shows the Syzygy results for this position with black to move.   The top result is Kxf5, even though Qg5 is a better move because it immediately checkmates white.  This best move is the 4th choice returned by the Syzygy tablebases.  The program using the Nalimov tablebases returns the best move for the top choice. 

I wondered if the stockfish engine would be able to catch this problem and play Qg5# even though it was not the top result.   So I went back to the beginning of the problem and had the program play black while I played white.  Sure enough, the program used the top choice and selected Kxf5 instead of the more precise Qg5#.  Apparently when doing analysis the program assumes the tablebase is returning the best move as the top result and doesn't look at other options.

The interesting thing is that the Syzygy tablebase returned the best move as the top result on all the earlier moves in the problem.

In the second example, the Syzygy tablebase returned an inferior move immediately.

The top move recommended by Syzygy here is Kf6.  Comparing it to the results from the program running the Nalimov tablebases, I immediately notice that the best move is Qh5+, with 7 moves until black is checkmated with optimal play.  Looking down the list of moves in the Nalimov program, I discover that Kf6 is ranked #13 with 16 moves to checkmate!  I tried playing this problem against the computer in the program running the Syzygy tablebases, and once again the computer automatically used the top move recommended by the tablebase.

So, can anyone out there tell me what is going on here?  Are the Syzygy tablebases being sold by Chessbase not compiled correctly?  As I noted earlier, in many positions the Syzygy tablebases do return the best move for the top choice.  But because I found two positions where this was not the case so quickly, it is reasonable to assume that that inaccuracies will be returned in many, if not most, chess problems.

Bump

I'm not using any Chessbase GUIs or Chessbase tablebase files. (I'm using Arena and downloaded Syzygy & Nalimov files from independent sites.)

Example 1 tells me there's probably something wrong with at least one of your Syzygy files.

In Example 2, my Nalimov files say it's mate in 16, with any of three moves: Kf6, Kf5, or Qh5. Unless you can show that this position is really a mate in 9, I'm assuming my Nalimov info is correct. So it appears that at least one of your Nalimov files may be corrupt, also.

I'd start by checking the md5 hashes of your Syzygy and Nalimov files. I don't know if Chessbase has that capability built in. If not, you should be able to find md5 hash value files on a few servers offering independent downloads of Syzygy/Nalimov files.

Do you have both WDL and DTZ files of syzgy? Are you using 5 men or 6 men TB?

http://oics.olympuschess.com/tracker/index.php (this is where anyone can download for free legally)

 I have 6 men syzgy with both WDL and DTZ and both Leela and Stock see that the best move is   Qh5+. (SF says mate in 16, Leela say winning percentage as usual) 

And the first diagram, both engines see mate in 1.

If you have only WDL and no DTZ, engine will play perfectly but the distance to mate may be longer.

For example Mate in 40 instead of Mate in 20. ( But in perfect chess, there is no reward for shorter mate)

So theoretically, WDL only can still play perfectly!

If you do decide to check the md5 hash values of your tablebase files:

You could do it manually, which would take forever. For an automated method, I like to download the Md5Checker program. Put a copy of the Md5Checker program in your folder containing the tablebase files. Also, make sure the hash value file is also in that folder. (The hash value file ends with the xxxx.md5 file extension.) Double-click on the Md5Checker.exe file to start the program. Click on the "Add" button to add all of the tablebase files, then click on the "Start" button to start checking the files.

http://getmd5checker.com/

The Syzygy is flawed. More on this to follow if you would like to see more.

No, Syzgy can never be flawed.

In Syzgy, all possible moves are calculated with permutation. 100% correct or perfect chess.

In case it is flawed

1.  downloaded files are missing/ corrupted 

2. GUI is bugged or

3. engine probing is flawed in the code.