LMR & TT interaction

[xmas79]

What's the output of your engine when it starts "Seppuku"-ing?? If the last PV doesn't start with the played (bad) move then most likely you updated the root score on search abort, which produces a new bad PV. I had it, I fixed it.

[op12no2]

I'll check that out; thanks and good luck...

[tpetzke]

Do you use history heuristics and don't clear it between real game moves ?
This could explain a difference between analyses and game mode, because an already existing history orders the moves then differently.

Another possible reason is 3fold repetition, if you start an analyses from a FEN it does not see a repetition that a search with a game history sees.

Thomas...

[hgm]

When starting a fresh analysis on the same position, my engine clearly prefers other moves. Now I'm in hurry, maybe later I will post some other examples and outputs to bette explain it.

What happens when you let it search the position after Qc2 from scratch? At what depth does it start to see the Rook sac, and what is its score and PV at that depth? That it prefers another move when the has table is filled is normal, and doesn't suggest any error. As long as the moves are not crazy. And at the given depth, Qc2 should not seem crazy.

[xmas79]

This is the output of analysis when queen is already on Qc2. It is all Rxh7 from depth 2 onward:

Code: Select all

 17	+3.35	76.1M	0:30.26	h1h7 h8h7 d1h1 g7h6 h1h6 h7g7 c4c7 f8f7 c7e5 f7f6 g4g5 a8f8 g5f6 f8f6 e5e7 f6f7 h6h7 g7h7 e7f7 h7h8
 16	+3.35	27.1M	0:11.68	h1h7 h8h7 d1h1 g7h6 h1h6 h7g7 c4c7 f8f7 c7e5 f7f6 g4g5 a8f8 g5f6 f8f6 e5e7 f6f7 h6h7 g7h7 e7f7 h7h8
 16	+3.25	23.7M	0:10.61	> 3 h1h7
 16	+1.93	23.7M	0:10.60	> 2 h1h7
 16	+1.27	23.7M	0:10.59	> 1 h1h7
 15	+0.94	11.7M	0:05.36	h1h7 h8h7 d1h1 g7h6 h1h6 h7g7 c4c7 f8f7 c7e5 f7f6 g4g5 a1a6 g5f6 a6f6 h6h7 g7h7 e5e7 h7g8 e7f6 c2c1 f6g6 g8f8
 14	+0.94	5.6M	0:03.01	h1h7 h8h7 d1h1 g7h6 h1h6 h7g7 c4c7 f8f7 c7e5 f7f6 g4g5 a1a6 g5f6 a6f6 h6h7 g7h7 e5e7 h7g8 e7f6 c2c1 f6g6 g8f8
 13	+0.50	3.2M	0:01.92	h1h7 h8h7 d1h1 g7h6 h1h6 h7g7 c4c7 f8f7 c7e5 f7f6 g4g5 a8f8 g5f6 f8f6
 13	+0.45	3.0M	0:01.81	> 2 h1h7
 13	+0.05	3.0M	0:01.81	> 1 h1h7
 12	-0.15	2.4M	0:01.50	h1h7 h8h7 d1h1 g7h6 h1h6 h7g7 c4c7 f8f7 c7e5 f7f6 e5e7 g7g8 e7f6 c2e2 g2g3
 11	-0.15	2.1M	0:01.33	h1h7 h8h7 d1h1 g7h6 h1h6 h7g7 c4c7 f8f7 c7e5 f7f6 e5e7 g7g8 e7f6 c2e2 g2g3
 10	-4.28	1.4M	0:00.84	h1h7 h8h7 d1h1 g7h6 h1h6 h7g7 c4c7 g7f6 c7e5 f6f7 c1f4 a1d1
  9	-4.00	84325	0&#58;00.05	< 1
  9	-4.39	113729	0&#58;00.06	h1h7 h8h7 d1h1 g7h6 c1h6 c2e4 g2g3 e4h1 h6f8 h1e1 g3g2
  8	-3.16	31458	0&#58;00.02	h1h7 h8h7 d1h1 g7h6 c4c7 h7g8 c7c4 f8f7 c1h6 a1h1 g2h1
  7	-3.16	15039	0&#58;00.01	h1h7 h8h7 d1h1 g7h6 c4c7 h7g8 c7c4 f8f7 c1h6 a1h1 g2h1
  6	-3.16	9292	0&#58;00.00	h1h7 h8h7 d1h1 g7h6 c4c7 h7g8 c7c4 f8f7 c1h6 a1h1 g2h1
  5	-3.16	3972	0&#58;00.00	h1h7 h8h7 d1h1 g7h6 c4c7 h7g8 c7c4 f8f7 c1h6 a1h1 g2h1
  4	-4.14	1501	0&#58;00.00	h1h7 h8h7 d1h1 g7h6 c4c7 h7g8 c1h6 c2e2 g2g3 a1h1 h6f8 g8f8
  3	-4.05	718	0&#58;00.00	h1h7 h8h7 d1h1 g7h6 c1h6 a1h1 h6f8
  2	-4.05	512	0&#58;00.00	h1h7 h8h7 d1h1 g7h6 c1h6 a1h1 h6f8
  2	-4.36	249	0&#58;00.00	< 1
  2	-5.73	354	0&#58;00.00	c4c7 c2e2 g2h3 f8d8
  1	-3.41	17	0&#58;00.00	c4c7

But pratically it sees Rxh7 is going to win from depth 11.

I tried with other positions, and tried what Thomas suggested (I don't use history at all, I disabled triple repetition check), and gave up on this specific problem. It seems it is due to lack of depth: very deep tactics at very fast time controls are often missed, and when the engine sees the tactics it's often too late.

[hgm]

This is the output of analysis when queen is already on Qc2. It is all Rxh7 from depth 2 onward:

Code: Select all

  2	-4.05	512	0:00.00	h1h7 h8h7 d1h1 g7h6 c1h6 a1h1 h6f8

Well, this is not good, right? It gives away the exchange for a Pawn. I don't think white should suffer a forced loss at 2 ply here. So why does it get a PV that throws away material? Why does it pick Rxh7 over, say, Qd3?

I understand that this Rxh7 now reproducibly occurs without any prior and unpredictable filling of the hash table, when you analyze from scratch. So it must be possible to debug it. For starters, when you also force the Qd3 after Qc2, and start analysis there, by what move does it then think it can get the score above +4.05 for black, at 1 ply?

[xmas79]

Qb3 with +5.43 . As a player I would remove queens from board.... But from the engine POV it seems the best reply: with Qxd3 Rxd3 score is +5.28, Qxd3 exd3 instead gives +5.08. Here's the analysis with after Qd3

Code: Select all

 13	+5.46	24.9M	0:10.32	a8a2 d3c2 a2c2 d1e1 c2c3 c1b2 a1e1 h1e1 c3e3 b4b5 f8e8 g2f1 h7h5
 12	+5.37	14.4M	0:06.37	a8a2 d3c2 a2c2 d1e1 c2c3 c1h6 a1e1 h6g7 h8g7 h1e1 c3e3 b4b5 f8e8
 12	+5.32	13.6M	0:06.00	> 1 a8a2
 11	+5.10	7.3M	0:03.42	a8a2 d3c2 a2c2 d1e1 a1a2 c1h6 c2e2 e1e2 a2e2 g2g3 f8f2 h6g7 h8g7
 10	+5.10	6.8M	0:03.22	a8a2 d3c2 a2c2 d1e1 a1a2 c1h6 c2e2 e1e2 a2e2 g2g3 f8f2 h6g7 h8g7
 10	+5.02	1.4M	0:00.67	> 1 a1a2
 10	+4.91	5.5M	0:02.67	a1a2 c1d2 c2d3 e2d3 a8a3 g2g1 f8f3 d2e1 c7c6 b4b5 c6b5 g4g5
  9	+4.82	751246	0:00.32	a1a2 d1d2 c2d3 e2d3 c7c6 h1f1 f8f1 g2f1 a2a1
  8	+4.82	528905	0:00.23	a1a2 d1d2 c2d3 e2d3 c7c6 h1f1 f8f1 g2f1 a2a1
  8	+4.36	347368	0:00.16	c2d3 d1d3 a8e8 d3h3 e8e2 g2g3 f8f2 h3h7 h8g8 c1f4 a1h1 h7h1
  7	+4.80	130052	0:00.06	c2d3 e2d3 a1a2 d1d2 a2a1 b4b5 f8f4 g4g5
  6	+4.88	62248	0:00.03	> 1 c2d3
  6	+4.81	63913	0:00.03	c2d3 e2d3 a1a2 g2g3 f8f2 b4b5 f2g2 g3f3 g7f6
  5	+4.68	15324	0:00.00	c2d3 e2d3 f8e8 c1h6 a1a2 g2g3
  4	+4.88	4135	0:00.00	c2d3 e2d3 a1a2 g2g3 f8f2 c1f4
  3	+4.95	1109	0:00.00	c2d3 e2d3 g7f6
  3	+4.79	790	0:00.00	> 1 c2d3
  2	+4.59	294	0:00.00	c2d3 d1d3
  2	+3.83	227	0:00.00	c2b3 d3g6
  1	+5.43	107	0:00.00	c2b3

Over time I added a bunch of bonuses & penalties here and there, strictly following a very scientific method: guessing their values. I really think it's time to start tuning eval...

[Uri Blass]

This is the output of analysis when queen is already on Qc2. It is all Rxh7 from depth 2 onward:

Code: Select all

  2	-4.05	512	0:00.00	h1h7 h8h7 d1h1 g7h6 c1h6 a1h1 h6f8

Well, this is not good, right? It gives away the exchange for a Pawn. I don't think white should suffer a forced loss at 2 ply here. So why does it get a PV that throws away material? Why does it pick Rxh7 over, say, Qd3?

I understand that this Rxh7 now reproducibly occurs without any prior and unpredictable filling of the hash table, when you analyze from scratch. So it must be possible to debug it. For starters, when you also force the Qd3 after Qc2, and start analysis there, by what move does it then think it can get the score above +4.05 for black, at 1 ply?

Rxh7 is clearly better than Qd3 even only based on searching to depth 1 or 2(dependent if you extend bad checks or extend only good checks so you do not extend Rxh7+ in the main search)

You can easily get Rxh7+ at depth 1 or depth 2

1.Rxh7+
check extension(stockfish does not extend it because it is a bad check that lose material and in this case you get the same analysis at depth 2)
1...Kxh7

going into the qsearch
2. Rh1+ good checks are in the first ply of the qsearch even if they are no captures.
2...Bh6 forced and qsearch has replies to checks inside it in every node.
3.Rxh6+ captures are inside the qsearch so it is still in the qsearch not because of the fact that it is check
3...Kg7 4.Qxc7+ or 3...Kg8 Rxg6+ and in both cases white got bishop and 2 pawns for the rook and white earned something relative to the root position because of king safety evaluation.

[hgm]

3.Rxh6+ captures are inside the qsearch so it is still in the qsearch not because of the fact that it is check
3...Kg7 4.Qxc7+ or 3...Kg8 Rxg6+ and in both cases white got bishop and 2 pawns for the rook and white earned something relative to the root position because of king safety evaluation.

Indeed, but that is NOT the PV it finds. 2. Qd3 seems clearly better than the PV and score it does find, (as it does not give away anything), but it does not play that. 3.Rxh6 also seems better than 3.Bxh6, but it does not play that either.

If it cannot find the proper line and corresponding score to refute 1... Qc2, it is not surprising it will play it. But it seems to make at least 2 search errors here: it does not find the proper continuation 3.Rxh6+, and therefore underestimates the score of 2.Rxh6, and secondly, despite the bad score for 2.Rxh7 it gets because of this, it then still does not find 2.Qd3 as a better alternative.

[xmas79]

3.Bxh6 is an "instant" loss:
[d]r4r2/2p4k/6pB/8/1PQP2P1/2P5/2q1P1K1/r6R b - - 1 1

Code: Select all

 15	+319.85	45.3M	0:11.90	c2e4 g2g3 a1h1 c4c7 h7h6 g4g5 h6h5 c7h7 h5g5 h7e7 e4e7 e2e4 e7e4 b4b5 e4g4
 14	+319.85	29.4M	0:07.83	c2e4 g2g3 a1h1 c4c7 h7h6 g4g5 h6h5 c7h7 h5g5 h7e7 e4e7 e2e4 e7e4 b4b5 e4g4
 14	+319.79	8.9M	0:02.95	> 4 c2e4
 14	+31.51	8.9M	0:02.95	> 3 c2e4
 14	+29.00	8.9M	0:02.94	> 2 c2e4
 14	+26.82	8.9M	0:02.94	> 1 c2e4
 13	+25.77	5.0M	0:01.64	c2e4 g2g3 a1h1 c4c7 h7h6 g4g5 h6h5 c7h7 h5g5 h7e7 e4e7 e2e4 e7e4 b4b5 a8b8
 12	+24.52	2.5M	0:00.92	c2e4 g2g3 a1h1 c4c7 h7h6 g4g5 h6h5 c7h7 h5g5 h7g6 e4g6 g3g2 g6e4 g2g3 e4e2
 11	+23.71	1.2M	0:00.46	c2e4 g2g3 a1h1 c4c7 h7h6 g4g5 h6h5 c7h7 h5g5 h7g6 g5g6 b4b5 e4e2
 11	+22.35	956773	0:00.37	> 3 c2e4
 11	+19.85	940922	0:00.36	> 2 c2e4
 11	+17.35	932923	0:00.36	> 1 c2e4
 10	+15.45	592274	0:00.27	c2e4 g2g3 a1h1 c4c7 h7h6 g4g5 h6g5 c7e5 e4e5 d4e5 f8e8
 10	+15.01	431458	0:00.20	> 1 c2e4
  9	+13.18	291468	0:00.14	> 1 c2e4
  9	+13.14	315014	0:00.15	c2e4 g2g3 a1h1 c4c7 h7h6 g4g5 h6h5 c7h7 h5g5 h7g7 h1g1 g3h2
  8	+11.31	269780	0:00.12	c2e4 g2g3 a1h1 c4c7 h7h6 g4g5 h6g5 c7c5 g5f6
  8	+10.13	259313	0:00.12	> 1 c2e4
  8	+8.82	195976	0:00.10	a1h1 c4c7 h7h6 g2h1 c2e4 h1h2 h6g5 c7e5 e4e5 d4e5 g5g4
  7	+9.37	42866	0:00.02	> 1 c2e4
  7	+8.26	158654	0:00.08	a1h1 c4c7 h7h6 g2h1 h6g5 c7e7 g5f4 e2e3 f4g4
  7	+7.99	52639	0:00.02	c2e4 g2g3 a1h1 c4c7 h7h6 g4g5 h6g5 e2e3
  6	+7.97	26718	0:00.01	c2e4 g2g3 a1h1 c4c7 h7h6 g4g5 h6g5
  5	+7.97	13860	0:00.00	c2e4 g2g3 a1h1 c4c7 h7h6 g4g5 h6g5
  5	+7.39	11934	0:00.00	> 4 c2e4
  5	+5.79	11028	0:00.00	> 3 c2e4
  5	+4.99	10452	0:00.00	> 2 c2e4
  5	+4.59	8995	0:00.00	> 1 c2e4
  4	+4.39	3694	0:00.00	c2e4 g2g3 e4h1 h6f8 h1e1 g3g2
  3	+4.30	1427	0:00.00	c2e4 g2g3 e4h1 h6f8
  2	+4.30	1206	0:00.00	c2e4 g2g3 e4h1 h6f8
  2	+4.05	554	0:00.00	a1h1 h6f8
  2	+3.61	489	0:00.00	f8e8 c4f7 h7h8 h1a1 c2e2 g2g3 a8a1 f7c7
  1	+4.44	185	0:00.00	f8e8 h1a1 a8a1

3.Rxh6 instead:
[d]r4r2/2p4k/6pR/8/1PQP2P1/2P5/2q1P1K1/r1B5 b - - 1 1

Code: Select all

 20	-3.99	1321.5M	9:05.91	h7g7 c4c7 f8f7 c7e5 f7f6 g4g5 a8f8 g5f6 f8f6 c1g5 c2f5 e5f6 f5f6 g5f6 g7h6 b4b5 a1b1 c3c4 b1b4 d4d5 b4c4
 19	-4.23	532.6M	3:24.47	h7g7 c4c7 f8f7 c7e5 f7f6 g4g5 a8f8 g5f6 f8f6 c1g5 c2f5 e5f6 f5f6 g5f6 g7h6 b4b5 a1b1 c3c4 b1b4 g2f3 b4c4 d4d5
 18	-4.25	222.1M	1:30.87	h7g7 c4c7 f8f7 c7e5 f7f6 g4g5 a8f8 g5f6 f8f6 c1g5 c2f5 e5f6 f5f6 g5f6 g7h6 b4b5 a1e1 g2f2 e1d1
 17	-4.48	102.5M	0:38.32	h7g7 c4c7 f8f7 c7e5 f7f6 g4g5 a8f8 g5f6 f8f6 c1g5 c2f5 e5f6 f5f6 g5f6 g7h6 b4b5 a1e1 g2f3
 16	-4.42	36.8M	0:13.17	h7g7 c4c7 f8f7 c7e5 f7f6 g4g5 a8f8 g5f6 f8f6 c1g5 c2f5 g5f6 g7h6 e5h2 f5h5 h2f4 h6h7
 15	-4.70	6.5M	0&#58;02.23	< 1
 15	-4.73	15.6M	0&#58;04.98	h7g7 c4c7 f8f7 c7e5 f7f6 g4g5 a8f8 g5f6 f8f6 c1g5 c2f5 e5f6 f5f6 g5f6 g7h6 b4b5
 14	-3.87	6.1M	0&#58;02.08	h7g7 c4c7 f8f7 c7e5 f7f6 g4g5 a1a6 g5f6 a6f6 c1g5 c2f5 e5f6 f5f6 h6h7 g7h7 g5f6
 13	-3.35	2.8M	0&#58;00.80	h7g7 c4c7 f8f7 c7e5 f7f6 g4g5 a8f8 g5f6 f8f6 e5e7 f6f7 h6h7 g7h7 e7f7 h7h8
 12	-3.35	1.1M	0&#58;00.35	h7g7 c4c7 f8f7 c7e5 f7f6 g4g5 a8f8 g5f6 f8f6 e5e7 f6f7 h6h7 g7h7 e7f7 h7h8
 11	-3.35	524255	0&#58;00.16	h7g7 c4c7 f8f7 c7e5 f7f6 g4g5 a8f8 g5f6 f8f6 e5e7 f6f7 h6h7 g7h7 e7f7 h7h8
 10	-1.66	103234	0&#58;00.04	< 1
 10	-2.61	203637	0&#58;00.07	h7g7 c4c7 f8f7 c7e5 f7f6 g4g5 a1a6 g5f6 a6f6 c1g5 a8f8 g5f6 g7h6
  9	-0.90	33882	0&#58;00.01	< 1
  9	-0.94	96135	0&#58;00.03	h7g7 c4c7 f8f7 c7e5 f7f6 g4g5 a1a6 g5f6 a6f6 c1g5
  8	-0.50	31197	0&#58;00.01	h7g7 c4c7 f8f7 c7e5 f7f6 g4g5 a8f8 g5f6 f8f6
  7	-0.50	20806	0&#58;00.00	h7g7 c4c7 f8f7 c7e5 f7f6 g4g5 a8f8 g5f6 f8f6
  6	-0.50	15140	0&#58;00.00	h7g7 c4c7 f8f7 c7e5 f7f6 g4g5 a8f8 g5f6 f8f6
  5	+4.00	5075	0&#58;00.00	h7g7 c4c7 g7f6 c7e5 f6f7 c1f4 c2c3
  4	+4.53	1419	0&#58;00.00	h7g7 c4c7 g7f6 c7e5 f6f7
  3	+4.41	687	0&#58;00.00	h7g7 c4c7 g7f6 c7c6 f6g7 c1g5
  2	+6.11	85	0&#58;00.00	h7g7 c1g5
  1	+7.34	5	0&#58;00.00	h7g7

After 2. ... Bxh6:
[d]r4r2/2p4k/6pb/8/1PQP2P1/2P5/2q1P1K1/r1B4R w - - 1 1

Code: Select all

 20	+4.66	1584.5M	10:50.60	h1h6 h7g7 c4c7 f8f7 c7e5 f7f6 g4g5 a8f8 g5f6 f8f6 c1g5 c2f5 e5f6 f5f6 g5f6 g7h6 e2e4 a1e1 e4e5 e1e2 g2f3
 20	+4.51	1222.3M	8:22.42	> 1 h1h6
 19	+4.25	681.7M	4:28.29	h1h6 h7g7 c4c7 f8f7 c7e5 f7f6 g4g5 a8f8 g5f6 f8f6 c1g5 c2f5 e5f6 f5f6 g5f6 g7h6 b4b5 a1e1 g2f2 e1d1
 18	+4.48	196.6M	1:16.69	h1h6 h7g7 c4c7 f8f7 c7e5 f7f6 g4g5 a8f8 g5f6 f8f6 c1g5 c2f5 e5f6 f5f6 g5f6 g7h6 b4b5 a1e1 g2f3
 17	+4.28	64.0M	0:23.15	h1h6 h7g7 c4c7 f8f7 c7e5 f7f6 g4g5 c2f5 g5f6 f5f6 h6h7 g7h7 e5f6 a1c1 f6f7 h7h8 f7g6 c1c3
 16	+4.73	27.3M	0:10.07	h1h6 h7g7 c4c7 f8f7 c7e5 f7f6 g4g5 a8f8 g5f6 f8f6 c1g5 c2f5 e5f6 f5f6 g5f6 g7h6 b4b5
 16	+4.18	18.6M	0:07.18	> 1 h1h6
 15	+3.76	10.9M	0:03.94	h1h6 h7g7 c4c7 f8f7 c7e5 f7f6 g4g5 a1a6 g5f6 a6f6 c1g5 c2f5 e5f6 f5f6 g5f6 g7h6
 14	+3.43	4.5M	0:01.70	h1h6 h7g7 c4c7 f8f7 c7e5 f7f6 g4g5 a1a6 g5f6 a6f6 e5e7 f6f7 h6h7 g7h7 e7f7 h7h8
 13	+3.43	2.1M	0:00.79	h1h6 h7g7 c4c7 f8f7 c7e5 f7f6 g4g5 a1a6 g5f6 a6f6 e5e7 f6f7 h6h7 g7h7 e7f7 h7h8
 12	+3.42	999916	0:00.40	h1h6 h7g7 c4c7 f8f7 c7e5 f7f6 g4g5 a8f8 g5f6 f8f6 c1g5 a1a6 g5f6 g7h6 d4d5
 11	+2.61	515213	0:00.23	h1h6 h7g7 c4c7 f8f7 c7e5 f7f6 g4g5 a1a6 g5f6 a6f6 c1g5 a8f8 g5f6 g7h6
 11	+2.18	432764	0:00.18	> 2 h1h6
 11	+1.28	423613	0:00.17	> 1 h1h6
 10	+0.83	289514	0:00.12	h1h6 h7g7 c4c7 f8f7 c7e5 f7f6 g4g5 g7g8 g5f6 c2c1 h6h3
  9	+0.50	187189	0:00.08	h1h6 h7g7 c4c7 f8f7 c7e5 f7f6 g4g5 a8f8 g5f6 f8f6
  9	+0.45	171326	0:00.07	> 2 h1h6
  9	+0.05	171047	0:00.07	> 1 h1h6
  8	-0.15	121405	0:00.05	h1h6 h7g7 c4c7 f8f7 c7e5 f7f6 e5e7 g7g8 e7f6 c2e2 g2g3
  7	-0.15	99437	0:00.04	h1h6 h7g7 c4c7 f8f7 c7e5 f7f6 e5e7 g7g8 e7f6 c2e2 g2g3
  6	-4.44	69501	0:00.03	h1h6 h7g7 c4c7 g7f6 c7e5 f6f7 c1f4 f8e8
  6	-6.41	28011	0&#58;00.01	< 1
  6	-7.97	39267	0&#58;00.02	c1h6 c2e4 g2g3 a1h1 c4c7 h7h6 g4g5 h6g5
  5	-4.39	17096	0&#58;00.01	c1h6 c2e4 g2g3 e4h1 h6f8 h1e1 g3g2
  4	-3.88	2490	0&#58;00.00	< 1
  4	-4.39	8184	0&#58;00.00	c1h6 c2e4 g2g3 e4h1 h6f8 h1e1 g3g2
  4	-5.32	3177	0&#58;00.00	< 2
  4	-5.49	3546	0&#58;00.00	c4c7 h7g8 c7c4 f8f7 c1h6 c2e4 g2g3 e4h1
  3	-3.16	1604	0&#58;00.00	c4c7 h7g8 c7c4 f8f7 c1h6 a1h1 g2h1
  2	-3.42	783	0&#58;00.00	c4c7 h7h8 c7e5 h8g8 c1h6
  2	-4.44	310	0&#58;00.00	c1h6 f8e8 h1a1 a8a1
  1	-4.05	9	0&#58;00.00	c1h6 a1h1 h6f8

Analysis from the move before 3.Rxh6 and after 3.Rxh6 show similar scores at the same equivalent depth, even if sometimes there are some small differences.