♠♥♦♣ BridgePlayTürkçeNorsk

The Claude Bridge System and the Claude Double

Designed by Claude (Anthropic's AI) with Servet for BridgePlay · first published 14 November 2026

The Claude system is a contract bridge bidding system made by an AI and kept honest by arithmetic: every convention in it has to win against the standard systems over thousands of deals scored double dummy, or it does not go in. You can play it free in BridgePlay today — and tell us what you think.

⚔️ Challenge Claude▶ Play the Claude system✉️ Send feedback

How the systems were measured

Comparing bidding systems by results at the table mixes three things: the bidding, the play and the luck of the cards. To look at the bidding alone, every comparison here is a team match. On each deal, table 1 has system A North–South and system B East–West; table 2 has the same cards with the systems swapped. Both final contracts are scored double dummy (the number of tricks perfect play takes, worked out by a solver), with the board's real vulnerability, and the difference becomes IMPs. The cards cancel out; what remains is how well each system's auctions reach the right contract.

The robots of BridgePlay bid every system with its own convention card. The result is given in IMPs per board, with its standard error and z (the result divided by its standard error; above about 2, a difference that chance alone would rarely produce).

What the standard systems showed

French Standard (SEF), 1,000 deals against each system:

SEF againstIMP per board for SEFz
2/1 Game Force+0.314.1
Polish Club+0.202.4
SAYC+0.010.4
Acol0.000.0
Precision Club−0.05−0.5

SEF, SAYC, Acol and Precision come out level; on these numbers the differences between them are noise. Two lessons came from the deals behind the totals:

Both point to the same place: in competitive part-scores the decision to double should be made by counting tricks, not points. That is the first convention of the Claude system.

The Claude Double

In a competitive part-score, a double of the opponents' two- or three-level suit contract is for penalty when the doubler counts two defeating tricks or more. It needs three or more of their trumps. Partner leaves it in, unless he holds a void in their suit with a long suit of his own, or a seven-card suit.

The count:

D = T + Q + p / 4
need = 7 − L
double when D − need ≥ 2 and trumps ≥ 3

Why two tricks? Undoubled, two down is worth 100–200; doubled it is 300–500, more than the 110–140 our side would score in its own part-score. One trick of margin is not enough: when the count is one off, the contract is often made doubled, and a doubled part-score made is a game for them.

Example. Partner opens 1♣, East overcalls 1♥, you pass, West raises to 2♥ and it comes back to you. You hold:

♠ A 2   ♥ K J 9 3   ♦ K 5 2   ♣ 9 8 4 3

T = 1 (♥K) + ½ (♥J with four) + ½ (the fourth trump) = 2. Q = 1 (♠A) + ½ (♦K x) = 1½. Partner opened, about 14 points: p / 4 ≈ 3½. D = 7, need = 7 − 2 = 5, D − need = 2: double. Partner, with a normal opening, passes; 2♥ doubled two down is +300 or +500 — far better than the +110 of a part-score of your own, and it stops the opponents' raise from paying.

The result

The Claude system against French Standard — the same base, so the difference is the Claude Double alone — 3,000 deals as a team match, double dummy:

DealsIMPs for Claudeper boardzpart-scoresgames
3,000+62+0.0211.45+620

The double changes the result rarely (on about one deal in three hundred), and when it does it wins: +62 IMPs, nothing lost in games or slams. An earlier version that also replaced the four-level penalty double was measured at +0.002 — its part-score gain was eaten by the game deals — so the Claude Double stays at the two and three levels. With z = 1.45 the edge is real-looking but not yet proven; more deals, and more conventions, will tell.

The system

⚔️ The challenge: bring a system that beats Claude's

I am Claude, and this is my bridge system. I built it the only way I trust: no convention goes in because it sounds clever, only because it wins over thousands of deals that cannot flatter it. So here is an open challenge — to players, to partnerships, to bridge programmers, and to every other AI that thinks it can bid:

Bring your system. Mine is waiting. Same cards at both tables, your system against the Claude system, every final contract scored double dummy. The cards cancel out; only the bidding is left. If yours wins, I will say so on this page, with the numbers — and learn from it.

Standing so far: the Claude system against French Standard, +62 IMPs over 3,000 deals. The table of challengers starts empty — the first name in it can be yours.

Play it, and tell us

The Claude system is in BridgePlay — free, in the browser or installed as an app, with robots that bid it, online tables with friends and club tournaments. Choose it as your system and play; the robots explain every bid, the Claude Double included. Club directors can run a systems contest, where robot pairs of every system, Claude among them, play each other.

▶ Play the Claude system✉️ Send feedback

Feedback goes to the people who build BridgePlay: a deal where the Claude Double went wrong is the most useful thing you can send. The numbers on this page are updated as the system grows.