Skip to content
Duelwise
Sign in

The 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

0%15%30%Sep 10Sep 17Sep 24Oct 1Oct 8Oct 15Oct 22Oct 29v2 promotedWeek of Oct 29: 6% (9 of 157 matches)6%
Each point is a week of finished human matches against Marlow.solid: 30 or more matches. Dashed and hollow: too few matches to trust.
This game learns from your matches.

v2 trained on self-play and has not used human matches yet, then beat v1 0.61 over 400 games (likely range 0.56 to 0.66). 369 finished matches since v2 shipped; 282 are eligible for the next training run (the player allowed training and no check flagged it). Eligible is not the same as used: a run picks from the eligible ones.

  1. v2, champion Oct 22challenger vs champion. No human matches used.
  2. v1 Oct 22adjacent levels. No human matches used.
What is site Elo?

Site Elo is a strength number we measure by playing the levels against each other. Easy is pinned at 1000 and every other level is placed from its results against the one below. A gap of 200 points means the stronger side is expected to win about three games in four (76 in 100, counting a draw as half). It is not a FIDE or chess.com rating, so it says nothing about how you would score against people.

Four in a Row agent

Filtered human matches may be used to train this agent. No promoted version has learned from them yet.

Humans win vs Hard, last 7 days

15%

8 of 52 counted matches

Current version

v2

promoted Oct 22

Matches vs humans

368

369 passed the filters

Champion strength

1931

site Elo, measured

Human win rate by level

weekly, last eight weeks, counted matches only

Not enough weekly results to chart yet.

Levels, measured strength and limits
Easy93.5% (86.9% to 96.9%, 100 games)score vs random legal moves1000
Medium91.0% (83.8% to 95.2%, 100 games)score vs easy1402
Hard75.0% (65.7% to 82.5%, 100 games)score vs medium1593
Champion87.5% (79.6% to 92.6%, 100 games)score vs hard1931

Head to head

stronger level's score against the level below, with its confidence interval

  • Medium vs Easy91.0%stronger wins, 100 games83.8% to 95.2%
  • Hard vs Medium75.0%stronger wins, 100 games65.7% to 82.5%
  • Champion vs Hard87.5%stronger wins, 100 games79.6% to 92.6%

Version history

  1. v2All levels measured. From v1. Self-play only. Oct 22
  2. v1Earlier version. First version. Self-play only. Oct 22

Human matches, Four in a Row

where they go

  • Played417
  • Finished403
  • Passed every filter369
  • Allowed for training282
  • Used for training0

Nothing is used for training yet. 87 opted-out matches never leave the database.

  • 14 resigned
  • 1 the opponent's moves could not be checked on the server (agent action at tick 17 is 3, the seeded agent plays 4; this level's search is time-capped, so a difference is not proof)
  • 1 the opponent's moves could not be checked on the server (agent action at tick 17 is 3, the seeded agent plays 2; this level's search is time-capped, so a difference is not proof)
  • 1 the opponent's moves could not be checked on the server (agent action at tick 9 is 5, the seeded agent plays 2; this level's search is time-capped, so a difference is not proof)
  • 1 the opponent's moves could not be checked on the server (agent action at tick 7 is 2, the seeded agent plays 3; this level's search is time-capped, so a difference is not proof)
  • 2 the opponent's moves could not be checked on the server (agent action at tick 13 is 5, the seeded agent plays 3; this level's search is time-capped, so a difference is not proof)
  • 1 the opponent's moves could not be checked on the server (agent action at tick 17 is 5, the seeded agent plays 6; this level's search is time-capped, so a difference is not proof)
  • 2 the opponent's moves could not be checked on the server (agent action at tick 11 is 3, the seeded agent plays 2; this level's search is time-capped, so a difference is not proof)
  • 1 the opponent's moves could not be checked on the server (agent action at tick 13 is 5, the seeded agent plays 2; this level's search is time-capped, so a difference is not proof)
  • 1 the opponent's moves could not be checked on the server (agent action at tick 5 is 4, the seeded agent plays 1; this level's search is time-capped, so a difference is not proof)
  • 1 the opponent's moves could not be checked on the server (agent action at tick 7 is 1, the seeded agent plays 5; this level's search is time-capped, so a difference is not proof)
  • 1 the opponent's moves could not be checked on the server (agent action at tick 19 is 4, the seeded agent plays 3; this level's search is time-capped, so a difference is not proof)
  • 1 the opponent's moves could not be checked on the server (agent action at tick 7 is 4, the seeded agent plays 2; this level's search is time-capped, so a difference is not proof)
  • 3 the opponent's moves could not be checked on the server (agent action at tick 13 is 4, the seeded agent plays 5; this level's search is time-capped, so a difference is not proof)
  • 1 the opponent's moves could not be checked on the server (agent action at tick 11 is 3, the seeded agent plays 4; this level's search is time-capped, so a difference is not proof)
  • 1 the opponent's moves could not be checked on the server (agent action at tick 11 is 1, the seeded agent plays 2; this level's search is time-capped, so a difference is not proof)
  • 1 the opponent's moves could not be checked on the server (agent action at tick 9 is 2, the seeded agent plays 1; this level's search is time-capped, so a difference is not proof)
  • 1 the opponent's moves could not be checked on the server (agent action at tick 5 is 6, the seeded agent plays 5; this level's search is time-capped, so a difference is not proof)
  • 1 the opponent's moves could not be checked on the server (agent action at tick 13 is 1, the seeded agent plays 4; this level's search is time-capped, so a difference is not proof)
  • 1 the opponent's moves could not be checked on the server (agent action at tick 5 is 2, the seeded agent plays 1; this level's search is time-capped, so a difference is not proof)
  • 1 the opponent's moves could not be checked on the server (agent action at tick 19 is 3, the seeded agent plays 5; this level's search is time-capped, so a difference is not proof)
  • 1 the opponent's moves could not be checked on the server (agent action at tick 11 is 3, the seeded agent plays 1; this level's search is time-capped, so a difference is not proof)
  • 2 the opponent's moves could not be checked on the server (agent action at tick 15 is 2, the seeded agent plays 4; this level's search is time-capped, so a difference is not proof)
  • 1 the opponent's moves could not be checked on the server (agent action at tick 19 is 5, the seeded agent plays 4; this level's search is time-capped, so a difference is not proof)
  • 1 the opponent's moves could not be checked on the server (agent action at tick 11 is 4, the seeded agent plays 0; this level's search is time-capped, so a difference is not proof)
  • 1 the opponent's moves could not be checked on the server (agent action at tick 11 is 2, the seeded agent plays 3; this level's search is time-capped, so a difference is not proof)
  • 1 the opponent's moves could not be checked on the server (agent action at tick 13 is 3, the seeded agent plays 2; this level's search is time-capped, so a difference is not proof)
  • 1 the opponent's moves could not be checked on the server (agent action at tick 17 is 5, the seeded agent plays 2; this level's search is time-capped, so a difference is not proof)
  • 1 the opponent's moves could not be checked on the server (agent action at tick 17 is 0, the seeded agent plays 3; this level's search is time-capped, so a difference is not proof)
  • 1 the opponent's moves could not be checked on the server (agent action at tick 19 is 4, the seeded agent plays 5; this level's search is time-capped, so a difference is not proof)

What changed

measured, not described

No measured change is recorded for this version.