A check-count tiebreak rule for reducing draws in chess: preliminary Fairy-Stockfish experiments

[fenor94]

I have been testing a simple draw-reduction rule for chess:

• Normal checkmate still wins immediately.
• If the game reaches a natural draw, the player who has given more checks wins.
• If the check count is tied, Black wins.

By a natural draw, I mean a draw that would also be a draw under normal chess rules, such as threefold repetition, stalemate, insufficient material, or the 50-move rule.

The goal is not to replace normal chess strategy with a check-count mini-game. The goal is to keep ordinary chess mostly intact, while giving drawn positions a meaningful tiebreaker.

Experimental setup:

• Engine: Fairy-Stockfish
• Search: 1M nodes per move
• Games: 200 games per experiment unless stated otherwise
• Policy: draw-aware check-count policy

The engine does not chase checks from the opening. Instead, it starts valuing check-count advantage only when the position becomes sufficiently drawish. Therefore, normal chess play remains the main objective until the game is likely to simplify into a drawn ending.

I also used a max-ply cutoff in the experiments. This means that if a game exceeds a fixed maximum number of plies without checkmate or a natural draw, it is stopped by the experiment script and recorded as a cutoff draw. This is not intended as a chess rule. It is only a practical limit to prevent a small number of extremely long engine games from running indefinitely.

Question 1: Does the rule still produce games similar to normal chess?

A first sanity check is whether the rule destroys the normal structure of chess.

In a 200-game draw-aware self-play run:

Code: Select all

Checkmate wins:              96 / 200
Natural draws by tiebreak:   97 / 200
Max-ply cutoff draws:         7 / 200

For comparison, in the closest normal-chess baseline at similar search strength:

Code: Select all

Checkmate wins:              88 / 200
Draws:                      112 / 200

Color balance in the draw-aware run:

Code: Select all

White wins:                  88
Black wins:                 105
Max-ply cutoff draws:         7

If the max-ply cutoff games are also adjudicated by check count, the result becomes approximately:

Code: Select all

White wins:                  90
Black wins:                 110

This suggests that the rule does not eliminate ordinary chess wins. It mostly converts games that would otherwise be drawn into decisive results.

Game length and where the rule matters:

In the draw-aware self-play run:

Code: Select all

Games not ending by 50-move rule:  179 games, average 188.4 plies
Games ending by 50-move rule:       21 games, average 295.3 plies
Normal-chess baseline:             average 169.1 plies

So the rule mainly adds length and strategic pressure in late endings, especially in low-material positions where normal chess often becomes drawn.

The check race mostly starts after the opening:

Code: Select all

Checks in first 40 plies:    2.7%
Checks in first 80 plies:   13.0%
Most checks occur from ply 80 to ply 200.

This matches the intended behavior: the rule mainly changes simplified middlegames and endgames, rather than encouraging early check-chasing.

Question 2: Does material advantage still matter?

For insufficient-material endings, the result looks encouraging.

There were 44 such games:

Code: Select all

One side had more material:  21 games
Materially equal endings:   23 games

In all 21 cases where one side still had more material, the material-advantaged side won.

More broadly, among all 97 natural-draw tiebreak games:

Code: Select all

Material-advantaged side won:   50
Material-advantaged side lost:   5
Material equal:                 42

This suggests that material still matters strongly. Check count acts as a secondary endgame objective when material alone no longer decides the game.

Question 3: Can this rule distinguish different playing strengths?

I also ran a strength-ladder experiment. The stronger engine had slightly more search nodes than the weaker engine:

Code: Select all

600K vs 500K
700K vs 600K
800K vs 700K
900K vs 800K
1000K vs 900K

The same setup was tested under both the check-count tiebreak rule and ordinary chess.

Results under the check-count tiebreak rule:

Code: Select all

600K vs 500K:    142.5 / 200 = 71.2%
700K vs 600K:    143.5 / 200 = 71.8%
800K vs 700K:    141.5 / 200 = 70.8%
900K vs 800K:    145.5 / 200 = 72.8%
1000K vs 900K:   136.5 / 200 = 68.2%

Total:            709.5 / 1000 = 71.0%
Draws:             21 / 1000

Results under ordinary chess:

Code: Select all

600K vs 500K:    130.5 / 200 = 65.2%
700K vs 600K:    128.0 / 200 = 64.0%
800K vs 700K:    123.0 / 200 = 61.5%
900K vs 800K:    123.5 / 200 = 61.8%
1000K vs 900K:   133.0 / 200 = 66.5%

Total:            638.0 / 1000 = 63.8%
Draws:            568 / 1000

The main difference is the draw rate. Ordinary chess leaves many games unresolved, while the check-count tiebreak converts most of them into decisive outcomes.

Preliminary interpretation:

• The rule preserves ordinary checkmate wins.
• It does not seem to encourage crude early check-chasing under the draw-aware policy.
• It makes drawn endings strategically meaningful.
• Material advantage still strongly correlates with winning the tiebreak.
• It appears to distinguish nearby engine strengths better than ordinary chess, mainly by reducing draws.

Open question: How should the resulting endgames be analyzed?

The part I do not yet understand well is the new endgame theory.

Some endings such as N vs N or R vs R can still reach the 50-move rule. In these endings, the check-count lead may alternate between the two sides, and the final winner can depend heavily on the 50-move boundary.

For example, in R vs R endings, it is possible that some initial rook and king placements are objectively better because they allow one side to generate more checks before the 50-move counter expires. So the next step is probably to study check-count tablebases or specialized endgame searches.

[Aleks Peshkov]

I have seen very long ago another anti-draw suggestion: side that forces draw with the last game move wins (or loses, I forgot) 3/4 vs 1/4 point.
Very view draws left for 1/2 vs 1/2 score.

[fenor94]

I think an ideal anti-draw rule should satisfy three principles.

First, it should be independent of time. Otherwise, it cannot be applied uniformly across different formats, such as correspondence chess. Armageddon-style rules, for example, depend essentially on clock time.

Second, it should depend only on threats to the king. Otherwise, the rule may encourage behavior that is too different from ordinary chess. King of the Hill is a good example: it creates a new objective that is not directly tied to checkmate.

Third, it should not create pathological behavior. For instance, a rule that awards the win to the player who forces the last move of the game may encourage one side to avoid a draw indefinitely, even when the board position is no longer meaningfully changing.

For these reasons, the rule “if the game is drawn, the player who has delivered more checks wins” seems to me like the most natural anti-draw rule. A check is the closest measurable side effect of a checkmate: it is still directly about threatening the king, but it is easier to count before the game reaches a decisive result.

[petero2]

...
The engine does not chase checks from the opening. Instead, it starts valuing check-count advantage only when the position becomes sufficiently drawish. Therefore, normal chess play remains the main objective until the game is likely to simplify into a drawn ending.
...

It seems to me this variant could also change opening theory. Consider the following example. In standard chess 11. Nxd8 leads to a draw, but in your variant it gives black a forced win.
[pgn]
[Event "-"]
[Site "-"]
[Date "2026.05.29"]
[Round "-"]
[White "-"]
[Black "-"]
[Result "*"]

1. e4 d6 2. d4 Nf6 3. Nc3 g6 4. f4 Bg7 5. Nf3 c5 6. Bb5+ Bd7 7. e5 Ng4 8.
e6 fxe6 9. Ng5 Bxb5 10. Nxe6 Bxd4 11. Nxd8 Bf2+ 12. Kd2 Be3+ 13. Ke1 Bf2+
*
[/pgn]

[fenor94]

...
The engine does not chase checks from the opening. Instead, it starts valuing check-count advantage only when the position becomes sufficiently drawish. Therefore, normal chess play remains the main objective until the game is likely to simplify into a drawn ending.
...

It seems to me this variant could also change opening theory. Consider the following example. In standard chess 11. Nxd8 leads to a draw, but in your variant it gives black a forced win.
[pgn]
[Event "-"]
[Site "-"]
[Date "2026.05.29"]
[Round "-"]
[White "-"]
[Black "-"]
[Result "*"]

1. e4 d6 2. d4 Nf6 3. Nc3 g6 4. f4 Bg7 5. Nf3 c5 6. Bb5+ Bd7 7. e5 Ng4 8.
e6 fxe6 9. Ng5 Bxb5 10. Nxe6 Bxd4 11. Nxd8 Bf2+ 12. Kd2 Be3+ 13. Ke1 Bf2+
*
[/pgn]

Yes, I agree that this variant can change opening theory. If $11. Nxd8$ is drawn in standard chess but becomes a forced win for Black under the check-count tiebreak rule, then this move would no longer be a valid drawing resource in the variant.

I think this is unavoidable for any no-draw rule: some lines that are drawn in ordinary chess must be reclassified as wins for one side. The important question is whether these changes are strategically natural and limited, rather than whether opening theory changes at all. In this case, White would simply have to avoid entering a perpetual-check line that loses under the new tiebreak rule.