Every rated game played on lichess and chess.com, pulled at build time and taken apart: the rating arc, where the games actually end, what the repertoire drifted into, and a bot tuned to play like its owner.
The bot on the Play tab runs the same search code, and Gaddi's ISMCTS runs right beside it.
19,406 rated and casual games across two platforms, pulled from both public APIs at build time and analysed before anything on this page was written. The finding was not flattering: 40.8% of my decided games ended on a clock, not on a board — 49.3% of every loss was a timeout, against 32.4% of wins won on the opponent's. I don't lose positions nearly as often as I lose time.
Same clock, two curves:and the rating curve moves with it,
| Progress | Wins | Losses | Gap |
|---|---|---|---|
| 0–10% | 99.1% | 98.7% | +0.3% |
| 10–20% | 95.2% | 93.5% | +1.7% |
| 20–30% | 89.3% | 86.0% | +3.4% |
| 30–40% | 82.3% | 77.0% | +5.2% |
| 40–50% | 74.6% | 68.0% | +6.6% |
| 50–60% | 67.1% | 59.3% | +7.8% |
| 60–70% | 58.7% | 50.2% | +8.5% |
| 70–80% | 50.7% | 42.1% | +8.6% |
| 80–90% | 43.7% | 35.5% | +8.2% |
| 90–100% | 36.7% | 28.5% | +8.2% |
Corpus
19,406
games, 2019-02-09 → 2026-10-27. 9,440W / 9,469L / 497D — a losing record by 29.
Time at the board
1,082 h
763 h self-reported by lichess, plus 319 h derived from the wall clock in 4,982 chess.com PGNs — chess.com publishes no play-time figure, so this is two measurements added together, not one metric.
Showing up
83.4%
of days played — 2,351 of 2,818 days in the span, longest unbroken run 298 days. Longest loss streak 14 beats the longest win streak 12.
Repertoire as Black
I opened 2019 with the Scandinavian — at least 36.3% of my games as Black that year. By 2020 it was displaced by the Modern Defense (35.8%), and by 2022 it had dropped out of the top five entirely. It is my most-played reply again today: 36.9% of Black games in 2026.
Both halves are real, but they are two within-platform observations rather than one continuous line. The abandonment happened on lichess and the return happened on chess.com, with the 2023 handoff sitting between them — so “I came back to my first opening” cannot be cleanly separated from “I started fresh on a new site.” It is a handoff, not two accounts running side by side: lichess’s last flicker is 3 games in 2025, which is the only reason its rating history reaches that far. Game counts say when someone was actually playing; rating-history dates do not.
| Year | lichess | chess.com | Most-played reply | Scandinavian (top 5) |
|---|---|---|---|---|
| 2019 | 2,480 | 0 | Scandinavian Defense Mieses Kotrc Variation 26.3% | 36.3% |
| 2020 | 3,262 | 0 | Modern Defense 35.8% | 12.5% |
| 2021 | 4,672 | 19 | Modern Defense 54.5% | — |
| 2022 | 3,365 | 0 | Modern Defense 51.2% | — |
| 2023handoff | 337 | 2,017 | Modern Defense 21.1% | 20.5% |
| 2024 | 0 | 1,043 | Scandinavian Defense Mieses Kotrc Variation 17.2% | 21.3% |
| 2025 | 3 | 590 | Scandinavian Defense Mieses Kotrc Variation 14.5% | 17.9% |
| 2026 | 0 | 1,618 | Scandinavian Defense Mieses Kotrc Variation 29.0% | 36.9% |
14,119 games · joined 2019-02-06 · last game 2025-01-16
Every format on this account reads provisional — rating deviation grew while it sat idle after 2025-01-16, so what it shows is a last rating, not current form.
5,287 games · joined 2021-01-22 · last game 2026-10-27
Over in the Loopdown I give recurring production bugs names and personalities, because a bug you can name is a bug you can hunt. The same three keep turning up over the board — and unlike the ones at work, these came with their own receipts.
49.3%
of every loss, decided by the clock rather than the board. Not an opponent — a deadline.
34.8%
win rate by game nine of one sitting, against 48.5% on game one. He should have stopped at eight.
69 games — a thin tail, shown with its n
29.0%
of games as Black are the Scandinavian again in 2026, after it was displaced almost entirely on the other account. First loves are a repertoire choice.
Generated 2026-10-28 from the lichess and chess.com public APIs by scripts/gen-chess-stats.mjs — every number on this page is build output, not prose. Re-run it and the figures move.
Second pass — four fields the first analysis never read
More clock does not help me
Ten times the thinking time moves the win rate by half a point. Whatever decides these games, it is not how long I get to look at them.
I leave theory on move one
median book exit: ply 2
In the 109 games where I stayed in a named opening to ply 8 or deeper, I won 54.1% — against 48.9% across the 12,990 where I was out by ply 4. The sample is small and I have never acted on it.
Where the game came from
Matchmaking is a coin flip. Arena is a bloodbath. The gap is twenty-one points and the arena sample is small enough to stay a hypothesis.
How they actually end
Across every game, not just decided ones: the clock ends more of them than checkmate and resignation combined.
14,119 lichess games · scripts/gen-chess-deep.mjs