Oddly enough, this chapter is the first one in this “guide” that actually attempts to do
any real guiding. So far I’ve focused only on principles and explanations. But the
knowledge of why you’re doing something often makes all the difference in getting yourself
to do it and do it well.
To start constructing custom model games, start up a new lichess study. I’ve included
relevant links in the tools section. Once in the study, turn on Stockfish analysis and set
‘Multiple Lines’ to five. I always find it really helpful to see the various lines that
Stockfish is considering to see what the common ideas are across lines and how its eval drops with
different options. Get to the opening position of the line for which you want to build a game.
You’ll be selecting each move both for the “Hero” side (the side you’re looking to emulate)
and the “Villain” side (the side playing moves that you might have to deal with). There are
different principles for how you go about selecting moves for each side.
Having made a lot of custom model games, and having spent a lot of time committing them
to memory, I have developed a number of principles that should be used in making a good one.
I know very well the issues that spring up from trying to learn a poorly-designed custom model game,
and I want to help you to avoid them.
Principles for the Hero’s moves, in priority order:
- They should be strong moves. Stockfish’s evaluation should be used heavily here, because
Stockfish is stronger than any human - either you or the other people playing games
without computer assistance. Stockfish eval isn’t the only factor you should use when
choosing moves for your games, but it should be a really significant one. When Stockfish’s
opinion differs from what the bulk of masters play, I have almost always benefited from
figuring out why Stockfish prefers its choice. As a note, when there is a committal move
of some sort, or when Stockfish’s move seems strange or different than what strong players
play, then make sure you give the engine plenty of time to think it over - you don’t want
a superficial Stockfish evaluation to become the reason you make a bad move choice for
years to come.
- They should adhere to clear plans. Sometimes, Stockfish likes to move pieces back and forth
and - try as I might - even after looking at a position deeply the reasons are extremely
obscure to me. In that event, I should not include the move in my model game. It is better
to use a move I understand than one I don’t, even if the move I understand isn’t Stockfish’s
first choice. Of course, the best situation is where I gain understanding and then include a
move, but sometimes Stockfish seems a little befuddled by a position too. Don’t hard-code
befuddlement into the games you’re learning.
- They should work against a variety of your opponent’s plans. One of the downsides of using
custom model games to prepare is that because they’re longer than standard opening lines, you often
have to make do with fewer of them. As a result, you should try to learn ideas that have
general application - they work in as wide a variety of lines as possible. It can be fun to
learn some plan that completely busts a very particular line of your opponent’s play, but if
it works only against that line and you’ll have to learn something else against every other
line of play it can be relatively inefficient as opposed to learning a plan that works against
most everything.
- They should reach positions you like to play. You probably have some idea of positions you
like to play, and you should honor that. These games are intended to be inspirational as
much as they are intended to be instructive. You’re trying to be the best version of yourself,
and that’s how the Hero side should play. Don’t let a small difference in Stockfish eval push
you into positions you don’t like.
- They should perform well in actual games. I haven’t talked about game databases so far, but in
selecting opening moves they can be a useful indicator. You should prefer moves that have
produced good results for the players who have played them - often that data reveals aspects of
the move that might be invisible to Stockfish. For example, a move that isn’t quite as good
objectively may lead to a position that is easy for a human to play. Of course, game databases
stop being useful as a line gets longer, but you should use them for earlier moves as an indicator
of what you should play.
For the Villain’s moves:
- They should be moves you’re likely to see. Game databases should be used liberally here,
both online and OTB games, and you should filter those databases for the rating range of players
you’re likely to play. The principle here is simple - you’re not going to be able to prepare for
every position, so select for the ones you’re most likely to see over the board. If you’re going
to be playing 1400 players, don’t prepare for moves favored by top GMs - prepare for 1400 players.
It’s here where off-the-shelf courses often fail - they’re built for GMs and IMs, but they don’t
cover the positions you’ll reach against lower-rated players.
- They should be good moves. In general, you want to prepare lines against challenging,
well-played moves, as expressed by Stockfish eval and game database performance. It is usually the
case that when an opponent deviates, you can use the ideas from your plans against optimal play to
find a good option against a sub-optimal idea. It is more rarely the case that the reverse is true
- that knowing optimal lines against sub-optimal moves will leave you well informed to find good
moves against good play.
- They should cover a range of different plans. It can sometimes be the case that an opponent
might have a few radically different ways to treat a given position, that will require specific
knowledge on your part to handle each one. Even if one is much more popular than the others, it
can be useful to spend at least some of your training time on the less popular plans because they
are so different. You’re trying to use your model games to give you ideas in relevant positions,
but if there are specific positions you might reach that will require specific ideas then you should
try to learn them. On the other hand, if you’re building a new line and it’s going to be extremely
similar to a line you already have constructed, then maybe you shouldn’t spend the time to build and
learn it.
In addition to selecting each move well, one of the key considerations is where to stop. Sometimes a
game had a natural stopping point - it ended in mate or an easily winnable material advantage - but often
I had to decide where exactly to cut the game off in an equal-ish position. There is some judgment that
goes into this decision: longer games can show more ideas, but take longer to memorize and play over and
so take valuable training time. In general I found that the key aspect was if the position was still
showing ideas I found interesting or needed to better understand. Sometimes this meant cutting them off
in the late middlegame, sometimes deep into the ending.
I’m planning to give examples here in the blog, where I build model games for specific openings using
these principles. That may better show how these principles can be applied in practice, but hopefully
the ideas behind these principles give you guidance to start on your own.