Forum is still broken, by the way. My direct response to Jon went missing.Due to the nature of the game changes in eval at the leaves of the search tree can change the eval much farther up. And that is why "theory" is always changing.
For example, I have recently been looking at this game:
[Event "EU/TC9/final"]
[Site "ICCF"]
[Date "2014.06.10"]
[Round "?.3"]
[White "Borstnik, Ales"]
[Black "Kupsys, Alfonsas"]
[Result "1-0"]
1. e4 e6 2. d4 d5 3. Nc3 Nf6 4. e5 Nfd7 5. f4 c5 6. Nf3 Nc6 7. Be3 cxd4 8. Nxd4
Qb6 9. Qd2 Qxb2 10. Rb1 Qa3 11. Bb5 Nxd4 12. Bxd4 a6 13. Bxd7 Bxd7 14. Rb3 Qe7
15. Rxb7 Rc8 16. O-O Qa3 17. Rfb1 Bc5 18. Ne2 h6 19. f5 exf5 20. h3 Bc6
21. R7b3 Bxd4 22. Nxd4 Qc5 23. Rg3 g5 24. e6 fxe6 25. Rc3 Qa7 26. Qe3 Bd7
27. Rcb3 Qc5 28. Rb7 Rh7 29. Qe5 Kf8 30. R1b6 Kg8 31. Rxe6 Qxd4 32. Qxd4 Bxe6
33. Qf6 Bf7 34. Qxa6 Re8 35. a4 g4 36. Qd3 1-0
Here I think 18. .. h6 is a novelty and it looks like a good one, but after 20. h3! Black is busted. However, actually Black does not appear to have any really adequate moves at move 18 (.. Bb4+ for example gives him a difficult position too), so in fact he should probably deviate earlier. It appears recent games favor 15. .. Qd8, as in Caruana-Robson, US chmp 2016. However, this whole line is very complex. I am quite sure you can't search forward far earlier in the game and get an accurate eval for it.
--Jon
These are the choices that black could have made to improve his game. By the last one, it may already have been too late:
[d]r1bqkb1r/pp1n1ppp/2n1p3/3pP3/3N1P2/2N1B3/PPP3PP/R2QKB1R b KQkq - acd 39; acs 7200; bm Bc5; c3 "Qb6"; move# 08b; cce -65; ce -17; pm Bc5 {4495} Qb6 {357} Nxd4 {207} Bb4 {99} a6 {44} Be7 {33} Nc5 {30} Nb6 {22} Qa5 {4} g6 {2} Qc7 {1} g5 {1}; pv Bc5 Qd2 O-O Be2 a6 Bf3 Nxd4 Bxd4 Bxd4 Qxd4 b5 O-O Rb8 Ne2 b4 Qe3 a5 Nd4 Qb6 Rfb1 Qc5 Re1 Qb6 Rac1 Bb7 Rb1 Rfc8 c3 Ba6 cxb4 axb4 f5 Re8 Rbd1 exf5 Nxf5 Rxe5 Qxb6 Rxe1+ Kf2 Nxb6 Rxe1 Nc4 Bxd5; white_wins 1963; black_wins 1285; draws 1413;
[d]r3kb1r/1R1bqppp/p3p3/3pP3/3B1P2/2N5/P1PQ2PP/4K2R b Kkq - acd 46; acs 7200; bm Qh4+; c3 "Rc8"; move# 15b; ce -29; pm Qh4+; pv Qh4+ Bf2;
[d]2r1kb1r/1R1bqppp/p3p3/3pP3/3B1P2/2N5/P1PQ2PP/5RK1 b k - acd 48; acs 7200; bm Qd8; c3 "Qa3"; move# 16b; ce -27; pm Qd8; pv Qd8 f5;
[d]2r1k2r/1R1b1ppp/p3p3/2bpP3/3B1P2/q7/P1PQN1PP/1R4K1 b k - acd 47; acs 7200; bm Bxd4+; c3 "h6"; move# 18b; ce -43; pm Bxd4+; pv Bxd4+ Nxd4 Rc4 f5 exf5 e6 fxe6 Rb8+ Rc8 Rxc8+ Bxc8 Nxe6 Qd6 Nxg7+ Kf7 Nh5 Re8 Nf4 d4 Ne2 Re4 Rd1 Bb7 Ng3 Rf4 h3 d3 Qxd3 Qxd3 Rxd3 Bc8 Rc3 Bd7 Rc7 Ke6 Ra7 Ra4 Nxf5 Bc6 g4 Be4 Rxh7 Bxf5 Rh6+ Kf7 gxf5 Rxa2 Rc6 Ra5 f6 Rh5 Kg2 a5 Ra6 Rc5 Kg3 Rxc2 h4 Rc3+ Kf4 a4 h5 a3;