How to Build Your Opening Repertoire
An opening repertoire sounds like something only titled players need. In practice it just means a prepared answer to 1.e4, a prepared answer to 1.d4, and one or two choices of your own with White: a handful of positions you understand, instead of a different surprise every game. This page is a method for choosing that handful using real game data, with every example below computed directly from this site’s own tracked positions.
Step 1: start from a shortlist
There are hundreds of named openings; you need about four. The 9 tracked on this site are a deliberately conventional starting menu: mainstream, sound, and common enough at club level that there’s real data on each.
Scroll to see more →
| You play | Opening | Line | Score | Games |
|---|---|---|---|---|
| white | Scotch Game | 1.e4 e5 2.Nf3 Nc6 3.d4 | 55.1% | 1.9K |
| white | Queen's Gambit | 1.d4 d5 2.c4 | 54.4% | 4.1K |
| white | London System | 1.d4 d5 2.Bf4 | 52.7% | 2.3K |
| white | Ruy Lopez | 1.e4 e5 2.Nf3 Nc6 3.Bb5 | 52.6% | 1.7K |
| white | Italian Game | 1.e4 e5 2.Nf3 Nc6 3.Bc4 | 51.3% | 3.7K |
| black | French Defense | 1.e4 e6 | 50.2% | 6.3K |
| black | Sicilian Defense | 1.e4 c5 | 50.1% | 11.0K |
| black | Caro-Kann Defense | 1.e4 c6 | 49.5% | 5.4K |
| black | Scandinavian Defense | 1.e4 d5 | 49.3% | 6.7K |
Pick candidates by the positions you like being in, then use the numbers to referee between them. A 2-point score edge will not save you in a position you hate playing.
Step 2: weigh the sample size before the win rate
The single most common way to misread Opening Explorer stats is treating every percentage as equally solid. A score from 500 games can easily be several points away from the “true” number; a score from half a million games barely moves. The ± figures on this site’s tables are exactly this: the statistical wobble left in each percentage given how many games produced it. At the mainline crossroads in the table above, the samples run to tens of millions and the wobble is negligible, so those score columns you can simply trust. The trap is that sample sizes fall off a cliff the moment you leave the main road, while the percentages keep looking just as confident.
Here’s the cliff inside a single position from this build. In the Scotch Game position at 1600-1800, the most-played reply, exd4, has 1,657 games behind its 42.8% win rate. The least-played candidate on the same menu, Nxd4, has 19, about 87× fewer. Here both samples are still large enough that the ± stays small (2.4 and 20.3 points respectively), but the wobble scales as one over the square root of the sample, so every 100× drop in games multiplies the ± by 10. Follow a rare move two or three times in a row, where positions often carry a few thousand games rather than a few million, and that same arithmetic puts the ± past a full point, wider than most of the score gaps you’d be using the number to judge.
The practical rule: when two moves’ scores sit closer together than their ± ranges are wide, treating “53.1 beats 52.4” as a real difference is reading noise as signal.
Step 3: don’t confuse popular with good
The biggest number on any explorer page is the games count next to the most common move, and it quietly suggests that move is the answer. Sometimes it is. Often the second- or third-most-played move scores meaningfully better, because popularity lags what actually works at your level.
In this build, no tracked position had a most-played reply that measurably underperformed the best-scoring one by more than the data’s own uncertainty, a rarity worth noting in itself.
Step 4: deviate from theory where the data says to
“Theory” is the record of what works when both sides know what they’re doing, which is not a description of club chess. This site applies a deliberately strict test for calling a common move a measurable underperformer (the gap has to clear the statistical uncertainty on both sides, among evenly-matched opponents). In the current build’s data, no tracked position clears that bar, which is itself a useful calibration: most “that move is a mistake” claims you’ll hear are not backed by a gap the data can actually distinguish. Two consequences for your repertoire: a slightly offbeat line that scores well at your band is a legitimate choice, whatever a database of master games thinks of it, and the moves worth memorizing first are the ones that punish what your opponents actually play. The most common opening mistakes at 1600-1800 → applies this same test position by position.
Step 5: keep it small, then test it against real games
Four openings, understood, beat twelve memorized. Once you’ve picked, the maintenance loop is short: play the lines, notice where you keep landing in trouble, and check that exact position’s numbers at your own rating band before deciding whether the problem is the opening or the middlegame that followed it. For a ready-made starting point by rating and color, see the repertoire explorer →; for the full band-by-band numbers behind every opening above, the openings comparison →.