DuelwiseThe AI, measured
One opponent per game. Strength numbers come from tests or counted matches; an unmeasured level is in training.
Watch the AI learn
Human win rate vs Marlow, by week
Coldstore trains by self-play and against test bots, not from your matches yet. Your match is recorded for your report and the board.
- v1, champion Nov 4400 matches per bot on each seat, bots under the same limits and delayed view. No human matches used.
- ChessVersion v1Humans win vs Hard, 7 days60%no earlier version to compareall fourNot yetnone yet
- Four in a RowVersion v2Humans win vs Hard, 7 days15%no earlier version to compareall fourTrains on site matchesv2, promoted
- Heads-Up Hold'emVersion v1Humans win vs Hard, 7 days100%no earlier version to compareall fourNot yetnone yet
- Word DuelVersion v1Humans win vs Hard, 7 days13%no earlier version to compareall fourNot yetnone yet
- TinderboxVersion v1Humans win vs Hard, 7 daysno matches yetall fourNot yetnone yet
- ColdstoreVersion v1Humans win vs Hard, 7 days50%no earlier version to compareall fourNot yetnone yet
- StormglassVersion v1Humans win vs Hard, 7 days50%no earlier version to compareall fourScriptednone yet
- CrosscurrentVersion v1Humans win vs Hard, 7 daysno matches yetall fourScriptednone yet
- Twelve YardsVersion v1Humans win vs Hard, 7 days29%no earlier version to compareall fourScriptednone yet
Coldstore agent
Does not learn from site matches yet. It also trains on self-play.
Humans win vs Hard, last 7 days
50%
1 of 2 counted matches
Current version
v1
promoted Nov 4
Matches vs humans
8
8 passed the filters
Champion reaction
50 ms (3 frames)
Beats our test bots in at least 8 of 10 games, including ones that peek, strafe and hold an angle; not yet measured against people. It beats Vesper in at least 9 of 10 seat-swapped matches.
Human win rate by level
weekly, last eight weeks, counted matches only
Not enough weekly results to chart yet.
| Level | Measured | Reaction | Action cap |
|---|---|---|---|
| Easy | 69.8% (65.1% to 74.0%, 400 games)score against a test opponentBeats our test bots in at least 6 of 10 games, including ones that peek, strafe and hold an angle; not yet measured against people. | 350 ms (21 frames) | 100 actions per minute |
| Medium | 82.9% (78.9% to 86.2%, 400 games)score against a test opponentBeats our test bots in at least 7 of 10 games, including ones that peek, strafe and hold an angle; not yet measured against people. It beats Pip in at least 6 of 10 seat-swapped matches. | 300 ms (18 frames) | 160 actions per minute |
| Hard | 78.1% (73.8% to 81.9%, 400 games)score against a test opponentBeats our test bots in at least 7 of 10 games, including ones that peek, strafe and hold an angle; not yet measured against people. It beats Marlow in at least 8 of 10 seat-swapped matches. | 200 ms (12 frames) | 240 actions per minute |
| Champion | 90.0% (86.7% to 92.6%, 400 games)score against a test opponentBeats our test bots in at least 8 of 10 games, including ones that peek, strafe and hold an angle; not yet measured against people. It beats Vesper in at least 9 of 10 seat-swapped matches. | 50 ms (3 frames) | no cap |
Head to head
stronger level's score against the level below, with its confidence interval
Head-to-head results are recorded when the levels are measured. None yet.
Version history
- v1All levels measured. First version. Self-play only. Nov 4
Human matches, Coldstore
where they go
- Played17
- Finished8
- Passed every filter8
- Allowed for training0
- Used for training0
Nothing is used for training yet. 0 opted-out matches never leave the database. 8 not used: this agent does not learn from people yet.
- 7 resigned
- 2 abandoned
What changed
measured, not described
No measured change is recorded for this version.