Open data

Each time a game or a test on this site ends, the page reports who won: the person or the machine, and at which level. Those reports are the shared score on the front page. Here they are as data: two tables, in CSV and in JSON, free to use under CC BY 4.0.

It is a small dataset and an honest one about its limits. It does not measure how good people are at games in general. It records how visitors to one site fared against these particular programs, at the settings described on each game's page.

Download

Totals — one row per game, plus one per level, since the counter began. No dates: the counter never kept them.

totals.csv totals.json

By month — one row per month, game, level and page language, counted from the event log. The log starts on 11 November 2026, the day it was switched on.

monthly.csv monthly.json

Both are built from the live counter on request and cached for an hour. The addresses are stable: link to them, and your copy refreshes itself.

Right now

Columns

game

The key of the game or test, as the site's counter knows it: reversi, checkers, mindreader and so on. The page column says which page it is.

kind

game — a game against an opponent; test — one of the human-vs-machine tests, where a finished run counts as won when the person beat the machine's mark.

level

all for the sum of every level, otherwise normal, hard or expert; minesweeper uses beginner and intermediate, which are board sizes. A level of unknown in the monthly table means the page did not report one.

human_wins, computer_wins, games

Wins of the person, wins of the machine, and their sum. There are no draws: games that can be drawn report only decisive results.

human_share

human_wins divided by games, rounded to four places.

page

The page the game is played on, with its rules, levels and how the machine plays.

month, lang, players

Monthly table only: the month (YYYY-MM), the language of the page (en or ru), and the number of distinct visitors behind the row.

Read before you use it

• The results are reported by browsers. Anyone can send one by hand; the counter only limits how often. It is fine for a games site and not for anything that matters.
• It includes the authors' own test games, which are a large share of the earliest rows.
• Levels are not comparable across games: normal in Nim and normal in Reversi are different opponents. Compare a game with itself.
• The games changed over time: engines were fixed and rules corrected, and the totals mix games played before and after. Each page states its current rules.
• A test counts as one result per finished run. Abandoned games are not counted at all, which flatters whoever tends to finish.

What is not in it

No addresses, no names, no browser details. The server keeps a salted hash per visitor, only so that repeat players can be counted; that hash is never published, and the dataset carries only the count of distinct visitors per row.

License and citation

CC BY 4.0: use it, change it, publish it, sell it — with a credit and a link to this page.

Suggested citation: You vs Computer, “Human vs machine results”, youvscomputer.com/games/data.html, accessed on the date you downloaded it.