I have played 902 games as Black in the Italian Game since 2010, and scored 46.3 percent. In 95.1 percent of them I played the same third move. I never chose it. I absorbed it, and then I repeated it for sixteen years.
Two things this is not. It is not a time-management problem: 42 of the 468 losses were on the clock, and the archive wins more of those than it loses (58 to 42). And it is not an endgame problem: 103 losses were over inside fifteen moves, before an endgame existed. What is left is the opening.
| Opponent rating | Games | Score |
|---|---|---|
| 0-1000 | 121 | 52.1% |
| 1000-1200 | 372 | 49.5% |
| 1200-1400 | 359 | 43.7% |
| 1400-1600 | 50 | 28% |
Every game in the archive, laid out by move. Each column is one ply, each block a position, and a block is as tall as the number of games that reached it. Color runs from ember (losing) to blue (winning). At Black's third move, 695 of 731 games take the same branch.
| Line | Games | Score |
|---|---|---|
| 1.e4 | 901 | 46.4% |
| 1.e4 e5 | 897 | 46.5% |
| 1.e4 e5 2.Nf3 | 758 | 46.9% |
| 1.e4 e5 2.Bc4 | 131 | 45% |
| 1.e4 e5 2.Nc3 | 8 | 31.3% |
| 1.e4 e5 2.Nf3 Nc6 | 756 | 47% |
| 1.e4 e5 2.Bc4 Nc6 | 81 | 45.1% |
| 1.e4 e5 2.Bc4 Nf6 | 49 | 45.9% |
| 1.e4 e5 2.Nf3 Nc6 3.Bc4 | 731 | 45.9% |
| 1.e4 e5 2.Bc4 Nc6 3.Nf3 | 74 | 43.9% |
| 1.e4 e5 2.Bc4 Nf6 3.Nf3 | 33 | 40.9% |
| 1.e4 e5 2.Nf3 Nc6 3.Nc3 | 23 | 78.3% |
| 1.e4 e5 2.Bc4 Nf6 3.d3 | 13 | 53.9% |
| 1.e4 e5 2.Nf3 Nc6 3.Bc4 h6 | 695 | 46.6% |
| 1.e4 e5 2.Bc4 Nc6 3.Nf3 h6 | 68 | 43.4% |
| 1.e4 e5 2.Bc4 Nf6 3.Nf3 Nc6 | 33 | 40.9% |
| 1.e4 e5 2.Nf3 Nc6 3.Bc4 Nf6 | 25 | 24% |
| 1.e4 e5 2.Nf3 Nc6 3.Nc3 Nf6 | 18 | 72.2% |
| 1.e4 e5 2.Bc4 Nf6 3.d3 Nc6 | 13 | 53.9% |
| 1.e4 e5 2.Nf3 Nc6 3.Bc4 h6 4.d4 | 208 | 39.9% |
| 1.e4 e5 2.Nf3 Nc6 3.Bc4 h6 4.Nc3 | 138 | 47.1% |
| 1.e4 e5 2.Nf3 Nc6 3.Bc4 h6 4.O-O | 123 | 49.6% |
| 1.e4 e5 2.Nf3 Nc6 3.Bc4 h6 4.d3 | 110 | 51.8% |
| 1.e4 e5 2.Nf3 Nc6 3.Bc4 h6 4.c3 | 79 | 46.2% |
| 1.e4 e5 2.Bc4 Nc6 3.Nf3 h6 4.d4 | 25 | 40% |
| 1.e4 e5 2.Bc4 Nc6 3.Nf3 h6 4.Nc3 | 20 | 47.5% |
| 1.e4 e5 2.Bc4 Nf6 3.Nf3 Nc6 4.Ng5 | 20 | 40% |
| 1.e4 e5 2.Nf3 Nc6 3.Nc3 Nf6 4.Bc4 | 18 | 72.2% |
| 1.e4 e5 2.Nf3 Nc6 3.Bc4 Nf6 4.Ng5 | 15 | 13.3% |
| 1.e4 e5 2.Bc4 Nf6 3.d3 Nc6 4.Nf3 | 13 | 53.9% |
| 1.e4 e5 2.Nf3 Nc6 3.Bc4 h6 4.h3 | 11 | 54.6% |
| 1.e4 e5 2.Nf3 Nc6 3.Bc4 h6 4.a3 | 9 | 66.7% |
| 1.e4 e5 2.Nf3 Nc6 3.Bc4 h6 4.d4 exd4 | 168 | 39% |
| 1.e4 e5 2.Nf3 Nc6 3.Bc4 h6 4.Nc3 Nf6 | 107 | 48.6% |
| 1.e4 e5 2.Nf3 Nc6 3.Bc4 h6 4.O-O Nf6 | 88 | 49.4% |
| 1.e4 e5 2.Nf3 Nc6 3.Bc4 h6 4.d3 Nf6 | 80 | 47.5% |
| 1.e4 e5 2.Nf3 Nc6 3.Bc4 h6 4.c3 Nf6 | 59 | 50.8% |
| 1.e4 e5 2.Bc4 Nf6 3.Nf3 Nc6 4.Ng5 d5 | 20 | 40% |
| 1.e4 e5 2.Nf3 Nc6 3.Bc4 h6 4.O-O Bc5 | 18 | 41.7% |
| 1.e4 e5 2.Nf3 Nc6 3.Nc3 Nf6 4.Bc4 h6 | 18 | 72.2% |
| 1.e4 e5 2.Bc4 Nc6 3.Nf3 h6 4.d4 exd4 | 18 | 50% |
| 1.e4 e5 2.Nf3 Nc6 3.Bc4 h6 4.d3 Bc5 | 17 | 64.7% |
| 1.e4 e5 2.Nf3 Nc6 3.Bc4 h6 4.d4 d6 | 15 | 46.7% |
| 1.e4 e5 2.Bc4 Nc6 3.Nf3 h6 4.Nc3 Nf6 | 15 | 43.3% |
| 1.e4 e5 2.Nf3 Nc6 3.Bc4 Nf6 4.Ng5 d5 | 13 | 15.4% |
| 1.e4 e5 2.Nf3 Nc6 3.Bc4 h6 4.O-O d6 | 12 | 75% |
| 1.e4 e5 2.Nf3 Nc6 3.Bc4 h6 4.d3 d6 | 12 | 66.7% |
| 1.e4 e5 2.Nf3 Nc6 3.Bc4 h6 4.Nc3 d6 | 11 | 27.3% |
| 1.e4 e5 2.Nf3 Nc6 3.Bc4 h6 4.h3 Nf6 | 11 | 54.6% |
| 1.e4 e5 2.Nf3 Nc6 3.Bc4 h6 4.d4 Nxd4 | 10 | 35% |
| 1.e4 e5 2.Nf3 Nc6 3.Bc4 h6 4.c3 Bc5 | 10 | 55% |
| 1.e4 e5 2.Bc4 Nf6 3.d3 Nc6 4.Nf3 h6 | 9 | 66.7% |
| 1.e4 e5 2.Nf3 Nc6 3.Bc4 h6 4.Nc3 Bc5 | 8 | 37.5% |
At each decision point big enough to talk about (25 games or more), is the move played most often the best-supported move available? Four times it is not, and each of those alternatives is a move from this same archive, already played, already scored. Rows are ranked by the lower bound of a Wilson interval, so a short lucky run cannot outrank a long record.
| After | Played | Instead of | Gap | Half-points |
|---|---|---|---|---|
| 1.e4 e5 2.Nf3 Nc6 3.Bc4 h6 4.O-O | 4...Nf6 88 games, 49.4% | 4...d6 12 games, 75% | 25.6% | 22.5 |
| 1.e4 e5 2.Nf3 Nc6 3.Bc4 h6 4.d4 exd4 5.Nxd4 | 5...Nxd4 62 games, 28.2% | 5...Nf6 27 games, 57.4% | 29.2% | 18.09 |
| 1.e4 e5 2.Nf3 Nc6 3.Bc4 h6 4.d3 | 4...Nf6 80 games, 47.5% | 4...Bc5 17 games, 64.7% | 17.2% | 13.76 |
| 1.e4 e5 2.Nf3 Nc6 3.Bc4 h6 4.d4 exd4 5.Nxd4 Nxd4 6.Qxd4 | 6...b6 39 games, 29.5% | 6...Nf6 8 games, 62.5% | 33% | 12.88 |
Half-points is the gap multiplied by the games already played that way. It is what the habit has plausibly cost, not a prediction of what changing it would earn. See what this cannot claim.
Exact sequences reached at least 8 times and scoring at or below 0.22. These need no hedging. The first one is the whole essay in one row.
Fifteen games have reached this position across twelve years. Not one of them scored a point. The move that walks into it, 6...b6, has been played 39 times in total for 29.5%. The alternative already sitting in the same archive, 6...Nf6, has 8 games at 62.5%.
Nothing about this position is hard. It has simply never been looked at, which is the entire point.
| Line | Games | W / D / L | Score |
|---|---|---|---|
| 1.e4 e5 2.Nf3 Nc6 3.Bc4 h6 4.d4 exd4 5.Nxd4 Nxd4 6.Qxd4 b6 7.Qd5 | 15 | 0 / 0 / 15 | 0% |
| 1.e4 e5 2.Nf3 Nc6 3.Bc4 Nf6 4.Ng5 | 15 | 2 / 0 / 13 | 13.3% |
| 1.e4 e5 2.Nf3 Nc6 3.Bc4 h6 4.d4 exd4 5.Nxd4 Bc5 6.Nxc6 bxc6 | 13 | 2 / 1 / 10 | 19.2% |
| 1.e4 e5 2.Nf3 Nc6 3.Bc4 h6 4.Nc3 Nf6 5.d4 Bb4 | 9 | 0 / 0 / 9 | 0% |
A move that scored better is not a move that is better. It may have been played in different years, against different opponents, at longer time controls, or in games that were already going well. The overall score rose from 45.3 percent in 2010 to 2019 to 50.0 percent in 2024 to 2026, so any move played more recently inherits some of that drift.
The alternative move in a habit row sometimes has only a handful of games behind it. Every row is therefore ranked by the lower bound of a Wilson interval rather than by the raw score, so a short lucky run cannot outrank a long record, and both intervals are published beside the point estimates.
A game continues for forty more moves after the opening ends. A score attached to a fifth move is the average outcome of everything that followed it, not a measurement of that move.
The cliff table is the exception. Fifteen games reaching the same position and scoring zero, after a queen move that attacks f7 and b7 at once, is not a sampling artifact.
These checks run in your browser, right now, against the same data the page printed above. If one of them went red, the page would be wrong and would say so.
All 12 checks reconcile.
Release v0.2.0, CC-BY-4.0. Aggregates only: the node table, the two findings tables, the drill deck and the summary. No per-game rows and no opponent names, because the question is about one player's decisions and answering it does not require re-hosting a roster of the people he played.
| File | SHA-256 |
|---|---|
| nodes.csv | db567d2e33df396193b98d6dab1a2e8b1e7064de526a9a83d1a0f4a0ebc3373d |
| leaks.csv | e00dcd92531cc2a56b07a5ce42aef11715d1a9bc909d81ea18549c914f87b3a8 |
| cliffs.csv | dc103cd6501b1fd3f6911784243b06dea02a34c06c4cd05b7804e3cf8bfb7ce8 |
| drills.json | 87f43d3729ff13e77dc0ae1a71957aca3c6d83f3c22bfc954362405d47874540 |
| book.json | 1320cd375492ea3445653083152a878f5b2b734257df4701cffde9eafc25df3e |
| evals.json | 6f790e076484e3e1d8884bc96c6317d6bc31eae45a2b7c0f9c1affa5bb219f03 |
| summary.json | 7090a00241ebce7589d421a323d4aa584644e019d5ea134608021b91917c9e3f |
Browse the release directory, or check the sums. Cite as: Lawrence, J. (2026). Italian as Black: a decision audit of 902 games (v0.2.0). jakelawrence.xyz/research/italian-as-black. Derived from the player's own public Chess.com archive.
| Source | Publisher | Accessed |
|---|---|---|
| Public player archive for JakeAlanLawrence | Chess.com | 2026-09-02 |
| Opening classification (ECO codes and opening names) | Chess.com | 2026-09-02 |
| Probable inference, the law of succession, and statistical inference | Journal of the American Statistical Association | 2026-09-03 |
Updated 2026-09-03. Next review 2026-12. First release. 902 games pulled from the public Chess.com archive on 2026-09-02, replayed through a legality-checked engine, aggregated to ply 16.
Every move sequence in the archive is replayed through a legality-checked rules engine on every build, so a mis-parsed game cannot reach a published figure. The engine evaluations, when they appear, are a frozen measurement rather than a computation: Stockfish is not bit-reproducible across builds, nets and search limits, so they are never recomputed and are labeled as what they are.
I design and ship AI tools, full-stack apps, and data pipelines — end to end, to production. Tell me the problem in a sentence; I'll give you an honest read on fit within a day.
Work with me →