Buggy behavior from SF, <1 or >128 ?

[Joerg Oster]

Sometimes SF (with 6 pcs Syzygy) gives Bxc5 completely winning (>128) and sometimes drawish (0.75).
Is there a bug in draw by repetition ? a Bug in the search ?
The original puzzle stated that Bxc5 is draw and Bf8 is winning.

For me Bxc5 is draw because : 1. Bxc5 Bc7 2. Bb6 Bd6 3. Ba5 Be7 4. Be1 Bd6 5. Bc3 Bb8 6. Bb4 Bc7 7. Ba5 Bd6 8. Kc4 Bb8 9. Bb4 Bc7 10. d6 Bb6 11. Kd5 Bd4 12. Ba3 Ba1 *

I think Bxc5 is winning.

I agree after a long analysis using texel. Bf8 is a forced mate in 207 moves or less and Bxc5 is a forced mate in 223 moves or less. (If texel could probe 6 men DTM TBs those numbers would most likely be much smaller.)

I have the same "bouncing scores" problem in texel too. It seems there are quite a lot of zugzwang positions deep in the search tree for this position, which makes it hard for engines to solve.

This may play a role, but I really think this position reveals the same problem as the one mentioned here: http://talkchess.com/forum/viewtopic.php?t=58258

Just that here it happens quite frequently.

If I don't save draw scores into TT, problem seems to be gone.
At least, I was not able to reproduce the score dropping so far.

I may be wrong, of course.