by Steve Lopez

Letters...I get letters...

Over fifty of them a week, in fact. The majority are product queries or requests for help using a feature of a specific program. But once in a while, I get an e-mail in response to a particular article I've written or one which offers a useful new tip that never occurred to me (yes, Virginia, I sometimes do still get surprised by the sheer flexibility and utility of ChessBase and Fritz). So this week, we'll dip into the electronic mail pouch and see what turns up. We'll be going back over a year for some of these; I've been meaning to do a mail roundup, but never got around to it until now. The letters are presented in italics in this article and, by the way, I've sometimes added my own comments in brackets; you'll easily spot these as I've marked them with my initials.

Frequent correspondent Bob Durrett is a one-man font of ideas for using ChessBase software. We often correspond on the methodology of using chess software (as opposed to using specific features, although that topic sometimes comes up, too). In February, 2001, Bob offered the following suggestions for evaluating opening variations:

While looking at ChessBaseMega2001 annotated games, I have found it useful to take into account the strengths of the players for each variation, sub-variation, etc. For this purpose, I have classified GMs and chessmasters as follows:


Notice the similarity to grades given in [U.S. -- SL] schools. I also have plusses and minuses. Kasparov gets an A+ whereas someone with a 2715 rating would get an A-.

Opening lines played between A+ players must be taken very seriously whereas opening lines played by C masters should be taken with a lot of caution. Lines where the winner has a much higher grade than the loser should be analyzed by Fritz to see where the loser made his first serious error. Then the line can be evaluated more realistically. It helps me to evaluate variations and sub-variations. I use Player's Encyclopedia all the time for this purpose.

I'm frequently asked by class-level players (of which I am one, by the way) how you'd evaluate an opening line. Of course, you could use the raw statistics that you'd find in an opening tree but, as I've ranted many times in previous columns, statistics can lie like a cheap rug. Bob's offered us a handy yardstick for "grading" the value of an opening variation. While such a grade can also be a tad misleading (for example, an A+ line might be a heck of a good opening, but also might require the chess knowledge of a 2700+ rated grandmaster to play the resulting middlegame properly), grading openings in such a way can be a useful yardstick for gauging their effectiveness (especially for the higher-rated players among us).

In a later e-mail, Bob elaborated on his "grading" method:

Many times, the outcome of the game is a win for one side, but Fritz evaluation of the final position shows an even game, or a plus for the losing side. When that happens, it seems reasonable to assume that this usually means that the flag fell, or something like that. It follows that checking the final position (both with Fritz and by independent evaluation since Fritz sometimes "messes up") of each and every line is important if the line is to be incorporated into one's opening repertoire.

Excellent point, Bob, which also addresses another question I'm sometimes asked regarding games that a player loses despite a seemingly winning position. We have to keep in mind that grandmasters are governed by the clock, just like "weekend warriors" in a Saturday afternoon USCF quad. GMs do sometimes overstep the time controls or just find themselves in severe time pressure (Anatoly Karpov is notorious for the latter). And regarding the idea of Fritz "messing up", see the ETN issue for March 10, 2002, under point #10, as well as the discussion of the Power Book in the issue for April 15, 2001. It happens -- computers do misevaluate positions, especially those involving long-term planning.

On a related topic, Bob offered this a day later:

During the recent tournament at the Corus Wijk ann Zee 2001, it was noted by several of the people monitoring the games at the Internet Chess Club that chess engines tended to show a pawn advantage in endgames where one side was up a pawn. That seems fine, but the endgames in question were theoretically drawn. The chess engines were unaware of the latest "Endgame Theory." In several cases, one side was up by more than one pawn but the game was still theoretically drawn. The engines simply didn't "realize" that. A similar situation was noted by some of the observers in some of the middlegames. The chess engines were indicating a plus for the side which was in trouble. The chess engines gave a misleading evaluation because they could not see the long-range consequences in the positions. Fortunately, at ICC, top-grandmaster evaluations were provided for some of the games. Similar situations may well have occurred at other sites.

This suggests caution when using chess engines, like Fritz, in evaluating lines in annotated games. The chess engines findings are great as a starting point in evaluation of positions, but are no substitute for independent evaluation or grandmaster assessment. (Engines are especially useful for spotting near-term tactics.)

Most people do not own extremely fast multi-processor computers with the latest Deep Fritz software nor do they own any of the other programs evaluated at the recent "Cadaques 2001 Super-program Supertournament." Instead, they probably have Fritz 6.0 or another similar program and are using a modest single-processor computer. Furthurmore, many people may not have the patience and time to let their program reach a great depth before they stop the evaluation. [Bullseye! More on this later -- SL]

Therefore, one should be guided by the following principle when using chess engines to evaluate lines:

"Grandmasters above about 2650 are smarter than modern-day chess engines, especially when using single-processor PCs. The assessments of these top grandmasters, given in games they annotate, should be taken much more seriously than assessments made by chess engines."

"Make sure the GM analysis being relied on is from a 2650+ GM, or accept the analysis with caution in those cases where the GM's analysis contradicts the chess engine's evaluation." (This assumes the engines are allowed to reach an evaluation depth of at least 13 ply or, better yet, 17 or 18 in critical positions.)

Bob's hit us with a lot here, so let's go a step at a time. First of all, it's definitely true that chess engines will sometimes misevaluate positions. But there are extenuating circumstances in some cases. For an example (regarding the evaluation of gambit lines), see the ETN for June 3, 2001. And chess engines are notoriously bad at endgames, which led directly to the creation of the Thompson and Nalimov endgame databases/tablebases, which have frequently been discussed in ETN.

Bob's basic point about thinking for oneself ("independent evaluation") is excellent, and is something I've been harping on since the start of ETN five years ago. This tenet also applies to analysis provided by human grandmasters (there have been several notable cases in which a GM's published evaluation has been shown to be deliberately misleading, so as to not "tip his hand" to potential future opponents. Also, see Andy Soltis' excellent column on chess annotation in the April 2002 issue of Chess Life magazine [on page 12]); you sometimes can't accept a human annotator's evaluations at face value either.

Also note what Bob says regarding "patience" when having an engine analyze a position. In general, the longer an engine analyzes, the better the analysis will be. This isn't always the case, of course -- the position might be a closed one that an engine can chew on from now until the cows come home and still not be able to evaluate properly -- but it's a good rule of thumb to allow your engine to analyze as long as possible. How long should you let an engine analyze? Your hardware has everything to do with this, of course. I can easily and routinely get 13 plies out of my Pentium III 800 MHz unit, but this might be the impossible dream for a Pentium I 90MHz machine. Bob suggests 17 or 18 plies but this requires some lengthy analysis even on my decent computer; on the other hand, it's trivial on a multi-processor machine. Let your conscience be your guide when deciding on an analysis time/depth, but if you're in the routine habit of letting an engine chew on a position for only about 15 seconds maximum, you're kidding yourself.

As for Bob's "principle" and "corollary", I'll leave it up to you to decide whether or not to follow it. 2650 seems a bit high to me personally, but keep in mind that I'm a gambiteer -- "Super GMs" rarely play the openings that are my personal bread and butter -- so I have a slightly different take on this issue than do most people.

Another letter came in from a different Bob -- Bob Lambert -- who disagrees with a comment I made in the April 1, 2000 issue of ETN:

...I strongly disagree with your recommendation that "The best games to analyze and investigate are your losses" as you state in April 1, 2001. You also say "You can analyze and study your victories (and there are sometimes very valid reasons for doing so -- such as the occasions when you later realize you played like a complete moron and essentially lucked out at the end), but most of the time this analysis just amounts to useless ego-stroking ".

For Class B players and below ( and I would guess even higher rated players), it's absolutely essential to analyze YOUR WINS. I've found all too often that I thought I played great, only to have Fritz point out TACTICAL misses by my opponent that allowed me to win. I remember a specific game, where it showed that my opponent missed a mate-in-1. I would have never seen this because I was too busy glowing about my extraordinary play. Additionally, it will point out tactics that would have allowed me to win much faster. Since Tactics is probably the single most important part of the game for Class B players and below, analyzing all rated games non-Blitz games with Fritz6 is mandatory. (If you have time, you can do the Blitz games also, but Blitz is Blitz, not chess).

I won't enter the "Is Blitz chess really chess?" debate (you can take that up with Mr. Lambert), but I basically agree with his disagreement. Ideally, you should go over all of your games, both wins and losses, analyzing on your own or with computer assistance. But that's not always possible, particularly for folks who play a lot of chess. Consequently, I simplified things by saying that you should go over your losses. Note my comment about "useless ego-stroking"; I've known far too many players (particularly youngsters) who go over just their victories and toss their losing gamescores into the trashcan. I believe that these players do themselves a great disservice, as one can learn a lot more from their losses than from their victories -- it was at these players that my remarks were primarily aimed.

So, yes, Bob Lambert's right -- you should analyze all of your games if possible. But if you can't analyze them all and you need to pick and choose among them, I still contend that losses are more fruitful than victories when choosing the ones to analyze with a chessplaying engine.

There are a few more letters that I'd like to share with you and we'll look at them in next week's ETN. Until then, have fun!

You can e-mail me with your comments, suggestions, and analysis for Electronic T-Notes.