However, I along with many others here do not have this line memorized. Many found the key moves, and I too saw the idea, thus proving that without study or memorized lines one can devise a winning plan.

Lets not talk about the first move. 13....♙xg5 14.♕xh8 is just what happens and seems to be Black's best shot. White position is horrible when unable to castle. Got the points there.

Now I tripped over the ♘ on d5. I took it with the ♗ first to take in a more flexible piece. Looks like I went wrong there: 14....♗xd5 15.♙xd5 ♕a5+ and White is pretty much free to play 16.♔d1 without too much danger. 16.♔f1 is met by ♕d2. Although I did see ♕d2 I wasn't confinced, not even with the ♗ on e6 instead of the ♘ on e7. Only now I notice that not only d3 but also e2 and b2 are under attack after that move and White is very bad there.

I also noticed that 16....♗g4+ in the main line would win. I'm not going indepth there, just because I didn't. I got stuck too much on 14....♗xd5 and 16.♔f1 that I did not really calculate the main line, but at least that seemed to win with the c-file for Black, etc.

<znprdx: absolutely agree <openingspecialist>>

I disagree with the comments suggesting that this was not a good puzzle, because white could have drawn. The draw was only possible because of two mistakes by the black: 20...Bxh1 and 23...f6.

Otherwise, the line played by Zsuzsa up to move 20 was absolutely winning, but should've been followed by 20. ... Qa4+ (see my previous posting for details). Note that 20...Bxd5 suggested by <Some call me Tim> is mostly defused with 21.Qh3 (attacking R on c8)

<Creg> had a good comment, worth repeating: <Once the white king is exposed and you can see that your remaining pieces can begin entering into the attack quickly you should realize that your opponent is simply under pressure and that you hold a strong initiative.> This is completely in agreement with Qa4+ continuing the attack instead of Bxa1?, prematurely cashing in the exchange.

Unfortunately, I agree. I know that we should use our brains and not engines, but in this case, I can't help but present some Fritz analysis. The top two moves were 13...Bxd5 and 13...hxg5. The practical and boring Bxd5 gives black the advantage, the bold and sensational hxg5 evaluates as equal. As a basketball analogy, a coach would chew out a player who goes for a risky (but crowd-pleasing) behind the back blind pass when a simple bounce pass would suffice. So which move represents the "solution" to the puzzle?

1. (-0.77): 13...Bxd5 14.exd5 hxg5 15.Qxh8 Qa5+ 16.Kd1 Qa4+ 17.Kc1 e4 18.Qh3 exd3 19.Qxd3 Nxd5 20.Qe4+ Ne7 21.Re1 axb5 22.Kb1

May I suggest that you turn on the "Deep analysis" mode in Fritz and let it crank for a while, pondering what to play as 13...?

Letting the engine run on the initial position is usually not the best thing to do in complex situations. One has to do some sliding analysis - pick a line, run it forward, then move back one ply at a time, while letting the program re-analyze it, and back-track to the initial position. This should be done in the "infinite analysis" mode, and typically identifies mistakes made by the software in selecting best moves at the deep levels of analysis. Big enough hash memory allocation is of utmost importance here; it enables the engine to retain valuations of deep moves while sliding back. It may take several iterations to get the real results.

<MAJ> Don't bother. You were right. Another 10 minutes and 1 ply deeper in Infinite Analysis mode, and 13...hxg5 comes out on top. I guess it *is* a good puzzle then ;-) I have even more admiration for Susan.

1. (-1.78): 13...hxg5 14.Qxh8 Nxd5 15.0-0 Nf4 16.Rad1 Nxd3 17.Rxd3 axb5 18.Rd2 Qf6 19.Nc2 Qf4 20.Rxd4 exd4

2. (-1.02): 13...Bxd5

My recollection is that with sliding forward/backward, the valuation at move 13 goes all the way to -5 or better. Here's the interesting part:

On the sliding forward analysis, the moves 13-20 appear to be pretty much best for both sides. However, a really deep and complete analysis starting at 20 ...? results in valuation -9 or so. Now one could ask: if moves 13-20 are optimal, then how come the valuation at move 13 is only -5?

The reason is that engines do a lot of search space pruning. For example, if two alternatives evaluate at (-2) and (-5) (with relatively shallow analysis), the one evaluating at (-5) will be abandoned. The only way to revisit it is to do a deep analysis on the (-2) branch and, for example, prove that it really is (-6).

Once the position at move 20 evaluates to (-9), sliding back analysis uncovers a lot of (shallowly analyzed) branches that appear to evaluate at (-5) or so. Descending into those branches would be very time-consuming (so I did not bother). Furthermore, the limited memory for hash (where the engine stores valuation of examined positions to avoid repeating the searches) causes the engine to discard the least-used portion of that hash, so that sliding back forces it to re-examine the discarded lines (hence the time-consuming behavior).

Oh, well. The bottom line is that engine is still only a tool, and needs careful guidance. I found that sliding forward/backward produces results much faster than just letting the engine run, but selection of when to stop analysis at any particular move is tricky.

This is why it rather annoys me when people disparage computer-aided analysis as requiring no human thought - there is still a fair amount of brain effort needed there, at least in positions that do not result in an immediate mate (where immediate is about 15 or fewer moves).