Connect-4 Opening Database

67,557 legal 8-ply positions — captured after each player has made exactly 4 moves, before anyone has won

67,557

Total Positions

Each with exactly 4 pieces per player on a 6×7 board

65.8%

Player X Win Rate

Goes first — a big structural advantage at this stage

~30%

Bottom Row Occupancy

Any bottom-row cell is occupied about 30% of the time — columns are used evenly

⚠

Why the raw charts look asymmetric — a dataset artefact

Connect-4 is a perfectly symmetric game. To save storage, the dataset stores each position only once in its "left-canonical" form. Only 180 positions have their mirror in the dataset — exactly the self-symmetric ones. The remaining 67,377 have their mirror deliberately omitted. Because X moves first, X's pieces appear strongly left-heavy (1.89× more in cols A–C vs E–G); O's second move can land anywhere so O shows a weaker bias (1.40×). The toggle below reveals the true, symmetric picture.

Outcome Distribution

X (first mover) wins nearly 2 in 3 positions. Draws are rare — under 1 in 10. Unaffected by the symmetry artefact.

Column Popularity — X vs O

View:

Board Heatmap — Player X

% of positions with X in each cell. Hover for value.

Brighter = more frequent

Board Heatmap — Player O

% of positions with O in each cell. Hover for value.

Brighter = more frequent

Row Distribution — how high pieces stack

Gravity dominates — row 1 holds over 5× as many pieces as row 2. Symmetric and unaffected by the artefact.

Centre Column (D) Control vs Game Outcome

In winning positions X averages 0.64 pieces in col D vs O's 0.45. In losing positions those flip: X drops to 0.39 while O commands 0.87. Centre column control is the clearest single predictor — and is unaffected by the canonicalization artefact (col D is its own mirror).