How to build a chess opening repertoire from the games you already play
Most repertoire advice starts with a choice of opening. Start with the positions you already reach instead: count which families your games actually produce, write down what you intend to play in them, and only then decide what is missing. A repertoire is a list of positions you have committed to, not a list of moves you have read about.
Start from the games you already play
To build a chess opening repertoire from the games you already play, the first step is counting rather than choosing. Take your reviewed games and group them by opening family. You will usually find that a handful of families cover most of your games, that one or two of them carry a disproportionate share of your errors, and that several openings you own a book about have never appeared at all.
Mikhail's statistics page does that grouping for you. Its by-opening rows come from games that went through the engine, and each row carries the family, how many games of yours it accounts for and how many measured errors those games contained, with a link straight into that family in the opening database. That is a repertoire brief written by your own results rather than by a repertoire course.
Read the two numbers separately
Games and errors answer different questions. A family with many games and few errors is working; leave it alone. A family with few games and many errors may be nothing at all - a small number of bad games rather than an opening problem. The one that deserves work is the family you reach often and handle badly, because that is the only combination where the work pays off every week.
Be careful with the boring conclusion, which is often the right one: your worst family is the one you meet twice a month and have never looked at, and the fix is one evening of reading rather than a new opening. A new opening resets your experience to zero in every position at once, which is a large price for a fresh start.
A repertoire is a list of positions
Write it as positions and intentions, not as move orders. "In this structure I play the c-file exchange" is a decision you can apply after a transposition. "1.e4 e5 2.Nf3 Nc6 3.Bb5" is a sequence that stops helping the moment your opponent plays the moves in a different order, which at club level they constantly will.
Mikhail's repertoire store follows the same rule. Membership of a line is decided by the position - piece placement, side to move, castling rights, and an en-passant square only when a capture is really legal - so reaching your line by another route still counts as your line. You store lines for either colour, label them in your own words, and record alternatives you are happy to allow rather than only one permitted move.
Store it before you judge it
This part is a rule about evidence, not about chess. An engine's small preference against your third move is not proof that you played badly in the opening; it is proof that an engine would have chosen otherwise, which it does to almost everyone almost always. Mikhail therefore refuses to treat engine disagreement in the opening as an opening weakness at all.
What creates something measurable is an obligation you took on. Once a line is stored, deviating from your own stored intention is a fact about your preparation, and that is the only opening obligation the system will claim. No stored repertoire, no opening exposure - whatever an engine thinks of your moves. Writing the repertoire down is what makes it possible to be measured against it.
Test it by recall, and let the trace tell you the depth
Once the lines are stored you can be asked about them one position at a time: one board, one question, the expected moves kept on the server, coverage reported as counts of what you have and have not confirmed. Nothing in that loop calls a deviation a failure with an engine - it reports whether you played what you said you would play, which is the only question a repertoire test can honestly settle.
Your game reviews supply the other half. Each review traces the game through the named-line catalog and reports the line, its ECO code, how many plies stayed in the book and the exact move that left it, including which side left first. Extend your written repertoire one move past where it keeps ending, in the families your own counts say you actually play, and the work stays small and permanently relevant.
Questions
How many lines should a repertoire have?
Enough to cover the replies you actually meet, which is a much shorter list than a published repertoire assumes. Add a line when a position keeps arriving and you keep improvising in it, and not before - an unused line is maintenance you pay for and never collect on.
Should I copy a strong player's repertoire?
Only the openings, and even then carefully. The lines a titled player needs are shaped by the opponents they face, who prepare. Yours are shaped by opponents who leave known lines early, so depth matters less for you and understanding the resulting positions matters more.
What if I keep leaving theory on move six?
That is a fact about the reference, not a verdict on your move. Look up the position where the match ended, read one move past it, and write down what you intend there. Leaving a named line early only costs you something when you cannot describe the position it leaves you in.
Does a repertoire matter below 1500?
A written one does, and a memorised one mostly does not. Knowing what you intend to play saves you the same decision every game and makes your positions repeat, which is what makes reviews start to accumulate into something. Twenty moves of a variation buys almost nothing at that level.
Try it on your games
The parts of Mikhail this guide describes are Opening database and Statistics, which need a free account — your games and your counts have to live somewhere.