Math and Simulations

Every card game teaches you the same first lesson: more is better. More cards in hand, more bodies on the board, more mana untapped. Magic formalised it as card advantage in the nineties — if you have more cards than they do, you have more options than they do, and over a long enough game options become wins. Yu-Gi-Oh players count it the same way, out loud: a card that turns one card into two is a "plus one," and entire decks are built around chaining plus-ones until the opponent has nothing left to answer with.

That lesson is true. It’s also the root of the most common way people lose.

Because if more is always better, the correct play is always the play that gets you more right now. Draw the extra card. Summon the extra body. Commit everything to the board because the board is where the advantage lives. That’s greedy play, and in a game where card advantage is the scoreboard, greed isn’t a character flaw — it’s the default strategy the rules hand you.

Ask what separates a good Magic player from an average one and you get a list of things that are about the cards rather than on them.

So the meta-skills in both games — role assignment, bluffing, reading the room, playing around the punish card — are all ways of being less greedy than the opponent expects, or of making their greed cost more than it pays. They’re real skills. They’re also skills that live around the edge of the cards, not in the middle of the board.

Both games have spent a decade making the cards stronger, and stronger cards change which skills matter.

In Magic, the average quality of a new card is far above what it was, and players of every format say the same thing in different words: games are now decided on the turn the first real threat sticks. Threats that protect themselves forced the printing of cheaper and cheaper removal, which forced threats that dodge removal, and the loop has left tempo with less room to work in than it used to have.

In Yu-Gi-Oh the shift is more dramatic. Going first is a structural advantage. The first turn can run five minutes of one player building a board of negates while the other player holds a hand and hopes it contains the one card that interacts. Players call it solitaire, and they’re not wrong.

None of that is a complaint about those games. They’re deep, they reward work, and their best players are genuinely brilliant. But the depth has migrated. It’s in deck-building, in sequencing, in format knowledge, in the opening hand. Less of it is in looking at a board mid-game with the same resources as your opponent and finding the better move.

Court of Jol is a positional card game, and the design does four things on purpose so that reading the board is the skill.

Greedy play still exists here — take the biggest break in front of you, farm the lane nobody’s contesting, dump every body into the Court. The question was whether the game punishes it, or just permits it. So we measured.

The game ships one bot with two settings. Both run the same reading of the board; Experienced adds a layer of doctrine on top, and Novice is that same bot with the layer switched off. Two shallower settings exist as measurement instruments — nobody plays them, but they give the ladder rungs below the ones you can choose, and without rungs below you cannot tell whether reading further is worth anything.

Then we made them play. 200 games per pairing — 600 for the two settings you can actually choose between — every game played to a result, and every game played twice: same shuffle, same decks in the same seats, same player moving first, swapping only which side is which setting. This game is lopsided before anyone thinks at all: with the same bot on both sides, going first wins 56.8%. Playing each setup from both sides cancels that exactly rather than on average. Sometimes the deal decides it and each side wins its turn with the good end of it — that pair says nothing about the settings. The last column counts only the pairs where reading actually decided it.

That’s the positional claim in one number. "What do I do after that" is a question about movement: which body comes out of Manifest, into which lane, Ready or not, and what that sets up. The opponent’s reply is mostly arithmetic — who breaks whom. Your follow-up is geometry. Geometry is where the win rate lives.

A fair question: with fourteen chosen cards and a seven-card hand, how much of this is the draw?

Less than in most card games, by design. Sixteen of your thirty cards are the same as theirs, so more than half of every deck is known to both players before the shuffle. The variance lives in the fourteen, and the fourteen are support — a Talent that readies a body, a Prayer that shores up a defender, a Prophecy that punishes a lane entry. They change how you execute a position; they don’t hand you one.

The simulations are seeded, so every pairing saw the same spread of opening hands — the thinking side didn’t win because it drew better, it won across the same draws. Better than that: every setup was played from both sides, so each side got the good end of the same shuffle exactly once, and the seven mirror controls confirm that cancellation is exact rather than approximate. What the deal is worth here is measurable and not enormous — going first is 56.8%, and the two decks are 53–47 once both sides know what they are doing. When the better reader takes four decided pairs in five across all of that, the draw is the noise and the reading is the signal.

We tried the cruelest version first: a board wipe. At the start of the handicapped side’s Formation on a given turn, every Cogum it has in a lane goes to the graveyard. Then it plays on. The control column is Novice wiped against Novice, which prices the wipe itself; the other column is Experienced wiped against Novice. The gap between them is what the extra reading buys back.

A board wipe in this game is not a card-advantage swing. It’s a tempo swing, and tempo converts straight into score. Two bodies off the board at turn 7 means two turns of the opponent Influencing undefended lanes for full SPRT. Nobody comes back from that reliably — not greed, not the reader. The old card-game instinct that a wipe "resets" the game is wrong here, because the scoreboard doesn’t reset.

Reading buys back a real share of the deficit, and more of it the earlier the damage lands: 28% becomes 43% at turn 3, 17% becomes 28% at turn 5, and by turn 7 — with the scoreboard already moving — 17% becomes only 21%. Skill doesn’t erase a deficit. It converts more of them, and it converts far more of them while there is still game left to convert in.

The wipe is a bad proxy for "behind", though, because it also destroys the thing the reader uses to come back. So here’s the cleaner test: a handicap on the scoreboard. The other side starts every game already ahead on INF in all three lanes. Full boards, full hands, same draws. The only difference is the score.

Down three points in every lane — a fifth of the way to losing each of them before the first card is played — Experienced against Novice is still a coin flip, 101–99. Hand the same deficit to Novice against Novice and the side behind wins 34%. Down five it is still 45%, against 28% for the control. It takes an eight-point head start in every lane, more than half the distance to a win, before the deficit beats the reading outright — and even then a third of them come back.

Being behind on the scoreboard is not the same as being behind in the game. A bad opening hand, a lost engagement, a turn you misread — the reader absorbs it, because the board is still there and the board is where the game is decided.

Simulations are bots. Here’s the human. Sixteen recorded games between the designer and the in-app AI — 238 turns of play, five games finished, all five won by the human, the rest exported mid-game. The records were made across several card-set versions and the AI tier wasn’t stamped on them, so treat these as a style study, not a win rate.

Read the first two rows together. The human summoned less and moved more. The AI put bodies on the table; the human put them where the game was. Every downstream number follows from that: more Ready entries, twice the instigations, twice the influences, and — the one that decides games — twice as many Influences landing on a lane nobody was defending. The human wasn’t winning engagements harder. The human was arriving where there was no engagement to win.

That is the positional thesis in one table, with a person in it instead of a bot. The third-ply result from section 06 — your own next turn is worth more than their reply — is what it looks like when someone actually plays that way: retreat, re-enter Ready, score into the gap.

Two of the five finished games were won from behind. Exactly the shape the score-handicap table predicts — behind by more than three a lane, against an opponent that has already closed one lane, and the reader still converts. Not every time, but often enough that "behind" isn’t "lost."