The Copernican Revolution

1.1For most of recorded history the working model of the sky put the Earth at the center, with the Sun and the Moon and the planets carried around it on circles. That model was not stupid and it was not wrong about anything a naked eye could check. Its standard form was compiled in the second century by an astronomer working in Alexandria, Claudius Ptolemy, and it told you where a planet would be on a given night well enough that calendars were run on it for something like fourteen hundred years. Measured by how long it stayed in use it is one of the most successful predictive instruments anyone has built, and any account of what happened to it has to start by granting that.

1.2The hard part of the job is something anyone can go outside and watch. Follow a bright planet across a season and it creeps steadily in one direction against the fixed stars, and then over a few weeks it slows, stops, reverses, traces a small loop backwards, and carries on as it was going before. A body riding one simple circle around a fixed Earth cannot do that. The repair was to give the planet a second, smaller circle, and to let the center of that small circle ride around the big one, so the planet loops as it travels, a wheel turning on the rim of a wheel. The small circle is an epicycle. It worked. It produced the backwards loop, and its size and its speed could be tuned until the prediction matched what was actually seen.

1.3What the model did was cost more every time observation got sharper. A fresh disagreement between the tables and the sky took another circle, or a nudge to a center, or an adjustment to a speed. Every one of those additions was locally correct, and every one was the cheapest available answer to the specific discrepancy in front of the specific astronomer, which is the condition under which accumulating cost stays invisible. Nobody chose the complexity. Everybody chose the next repair, and the complexity was the sum of the repairs. Lehman named the general form of this in software in 1980: a system in real use is under continuous pressure to change, and its complexity increases unless work is done specifically to reduce it, and nothing in the pressure to change supplies that work.

1.4Copernicus moved the center to the Sun, and the version of the story usually told is that this was simpler and more accurate. It was neither. Kuhn is direct about it in The Copernican Revolution: the system Copernicus actually published still used epicycles, needed as many circles as the one it replaced and arguably more, and predicted no better. What changed was which facts needed machinery. The backwards loop stopped requiring a device of its own on every planet, because if the Earth is moving too then we overtake the outer planets on the inside of the track, and while we pass them they appear to slide backwards against the stars. The observation that had cost a mechanism now cost nothing. The planets also acquired a definite order and definite relative distances, which the older model could not supply at all.

1.5So the word revolution in this piece carries Kuhn's sense and only Kuhn's sense. It means a change of center, after which a set of facts that had needed separate accounting falls out of the model for free. It is not a claim to be right, and it is not a claim to be better. Those are different arguments and this piece is not making them.

1.6The application is direct. Anyone who has played one of these games long enough has had the same experience. The game is still good and it is not the game you learned. Nothing broke. There is no set you can point at, and no decision anybody made that was obviously bad, and every release was defensible on the terms available to the people making it.

1.7That experience is usually attributed to nostalgia, and nostalgia is a claim about the player. But players who have been there long enough report the same change, and they report it arriving in the same order, in two games run by two companies that share nothing but a format. A change with that shape is not in the player. It is in the thing, and a change in the thing can be described.

2.1The complaint is usually phrased as power creep, and power creep is the symptom rather than the thing. Cards did get stronger. What people mean, when you push on it, is that the play changed shape, and three things changed shape together. They are separable and it is worth separating them.

2.2 Effects used to be atomic, and atomic effects compose with the board. Jinzo, printed in Pharaoh's Servant on 2002-10-20, is a Level 6 body at 2400 attack, and its entire text is eighteen words: "Trap Cards, and their effects on the field, cannot be activated. Negate all Trap effects on the field." It names no other card. It requires no other card. It is a verb you carry into a situation, and the situation is the board in front of you.

Endymion, the Mighty Master of Magic, printed in Structure Deck: Order of the Spellcasters on 2019-04-19, is a Level 7 body at 2800 attack with 180 words across two text boxes and four distinct effects. That is ten times Jinzo's text, sixteen and a half years later. The words Spell Counter appear in it seven times. Every one of the four effects is keyed to that resource, and the resource is produced by the cards it was printed alongside.

The comparison is chosen to be fair in both directions. Endymion is the card its Structure Deck was built around and Jinzo was the Secret Rare of the booster it came in, so both are the body a product was sold on. And the modern card is the more impressive piece of design. It is more expressive, it does more, it rewards more knowledge. The claim is narrower than "it is worse." The claim is that expressiveness which composes internally moves the decision off the table and into the deck list, because what Endymion composes with is the cards named on it, and what Jinzo composes with is whatever the opponent happens to have set.

2.3 Engines used to be small, and small engines were board facing. A small engine was two or three cards that made each other slightly better, sitting face up in front of both players, where both could see them and both could act on them. A large engine is hand facing and deck facing. The sequencing that decides the game happens before anything durable is on the table, and the windows the opponent gets are windows onto a sequence and not onto a board. Answering it takes a card held in advance for that purpose, and not a read on what is in front of you. The loss is not that combinations are bad. The loss is specific and it is worth stating exactly: what the opponent participates in moves, from the board in front of them to the card they held in case of this.

2.4 Utility was a card type and it stopped being one. A card that is useful in most matchups and excellent in none held a slot in both games for years. In a sufficiently deep pool it loses that slot to a card that is decisive in the one matchup it was printed for, because the pool is now deep enough that the specific card exists, and in Yu-Gi-Oh's case deep enough that the search which finds it exists too. Nobody decided utility should die. It was arithmetic.

2.5All three shifts bought real things. Deckbuilding has a higher skill ceiling than it did. Cards are more expressive. Games are faster and more dramatic, and the competitive scene has genuine depth that the 2002 game did not have. Most of the people playing prefer the current version and they are not wrong to. Rosewater, who has been designing Magic for three decades, has written for years about complexity as a resource to be spent deliberately rather than an error to be avoided, and that is the correct frame. The claim here is narrow. Something specific was traded away, the trade was structural rather than chosen, and there is an audience that has been waiting for someone to trade it back.

3.1A collectible game is funded by the sale of the next set, and it goes on releasing sets for as long as it is alive. That is arithmetic about a business and not a claim about anybody's intentions.

3.2Whether anyone at Wizards of the Coast or at Konami has ever been told that a set must change which decks are viable is not settled by anything in the public record, and I looked. The loop does not need that step. It runs on the release record and on one fact about how cards get picked for decks.

3.3In a constructed format with no copy restriction, a card strictly worse than a card already legal does not take the slot. So the part of the pool that matters grows only where a new card is distinguishable from an old one, and rate is the most reliable direction, because rate is the one thing a player can evaluate without knowing anything else about the format. Nobody takes that decision. It is arithmetic on a pool that keeps being added to and never gets smaller except by intervention.

3.4Written as a loop, running once per release:

1. 1.The pool grows.
2. 2.A larger pool enlarges the interaction surface, meaning the set of distinguishable situations a card has to be evaluated against.
3. 3.Coverage is the ratio of what gets tested to what could be tested, so coverage falls unless testing grows as fast as the surface does, and the public record does not say whether it has.
4. 4.Bans and errata and rotation are the tools for what the sample missed, and each intervention changes the game that the next set is designed against.

3.5The output of the loop has a name and the name is drift. Nobody chose the drift. Everybody chose the sets, one release at a time, and no single release changed the game, and ten years on the same cards are being used to play a game the people who learned the first one would have to learn again. Drift is invisible from inside because there is no moment at which it happens.

3.6Magic and Yu-Gi-Oh are two of the very few designs where drift is visible at all, because measuring drift needs a long baseline and almost nothing in the genre has one. The loop is a consequence of surviving. Magic has released continuously since 1993, the Pokémon Trading Card Game since 1996, and Yu-Gi-Oh's OCG since 1999, and after those three the list gets short quickly, and a game that died in year four never had the problem.

3.7Rotation and bans are the honest response to the loop and they work. A format with rotation is managing the loop deliberately with a good tool. The tool is not the problem. Managing a loop is not the same as leaving it, and the cost of the management is that a deck a player built stops being legal on a schedule.

3.8The loop and the surface multiply. Adding zones and a third stat and movement expands the interaction surface once, deliberately, for reasons that are good. The loop expands it again, every release, whether or not anybody wants it to. A design carrying both is growing the surface on purpose and growing it on a schedule it does not set, and those two multiply instead of adding. Every recent design carrying zones and movement is exposed to that, and Court of Jol is a recent design carrying zones and movement.

4.1A card is balanced against the set of situations it can be evaluated in and not against other cards, and that set is a much larger object than a card list. Whether a card is too strong has no answer in the abstract, and under pressure it resolves to whether there is a board state in which the card's rate is unanswerable. The unit of balance work is the situation. The set of situations is the interaction surface named in 3.4, it is what playtesting samples from, and the loop grows it on a schedule nobody at the table controls.

4.2Three things multiply the interaction surface, and the multiplication is the argument.

Stats multiply comparisons. One combat number is one question per matchup and the question has an answer. Three combat numbers make a body a point in a volume, better along one axis and worse along another, and which of the three gets read on which turn is the subject of The stat wall. That is good design. It is also a comparison table with three dimensions in it.

Zones multiply contexts. A card that can be in one of n zones is n cards for evaluation purposes, because its text gets read against a different context in every zone.

Movement multiplies transitions. Zones without movement are a partition, and the design question is which zone a card is in. Zones with movement are a graph, and the design question is which sequences of zones are reachable, which is a bigger question than which zone.

4.3The interaction surface is the design space. Those are two names for one number, so every argument that adding a dimension opened the design space is at the same time an argument that the thing which has to be playtested got bigger. Rosewater's "New World Order" (Making Magic, 5 December 2011) separates complexity into comprehension complexity and board complexity and strategic complexity, and treats all three as resources a design spends on purpose and not as errors it fails to avoid. The interaction surface is where the second and the third of those land.

4.4Nobody adds a dimension by accident, and the price is hard to see at the moment of adding. The benefit is stateable before the dimension is added, in one sentence about what becomes possible. The cost is not stateable until a card exists that exploits it, and by then it reads as a problem with the card. This project is not exempt from that.

5.1Magic looks like the largest play surface in the genre. The comprehensive rules list its zones at 400.1: library and hand and battlefield and graveyard and stack and exile and command, plus an ante zone that only old cards use. The obvious reading is that thirty years of depth came from having a zone for everything.

5.2That list is a set of accounts and not a surface in the sense 4.1 means, and the distinction is the argument. The battlefield is unordered. It has no positions and no adjacency and no distance, and nothing traverses it, because a zone change is an assignment and not a journey, with no state in which a card is between two zones and no cost per step. Every other zone is an account and not a place. A library and a hand and a graveyard are one player's, the stack is a pile that resolves backwards and empties, and exile holds what has been taken out of play. The one zone where both players' permanents stand together is the battlefield, and the battlefield is the one with no geometry.

5.3What Magic has instead of geometry is a conversion layer, land to mana to cards to board to damage, which Nothing costs anything takes apart as an economy. Every card in it is an exchange rate between two of those stocks, and an exchange rate is one-dimensional and therefore comparable. Three mana for a 4/4 is a rate and a rate can be ranked. That is what efficiency means, and it only means anything on a flat surface.

5.4The property that buys the longevity is that on a flat surface, mistakes stay local. Magic has banned a great many cards across more than thirty years. What matters about the bans is that a ban in Magic usually works, because removing the card leaves the rest of the format standing unchanged, and it works because a broken exchange rate is a property of one card and there is no table geometry for it to reach across. The battlefield has no positions, so no Magic card's problem can be where it is standing.

5.5With contested space, mistakes do not stay local. A card that is too strong in one position is a fact about the position as much as about the card, and the fixes available are a ban that leaves the position broken for whatever gets printed there next, or a change to the rules. Magic put its complexity in the conversion layer and left the surface flat. Whether that was foreseen is not something the public record settles, and it does not matter. The property is part of why the design survived more than thirty years of the loop in 3.4, and no design in the genre has survived more of it.

6.1Yu-Gi-Oh took the conversion layer out. There is no mana and no land drop and no pitch cost, and there is no currency the costs cards do carry are denominated in, so a card in hand is playable or it is not. That is the most under-credited structural decision in the genre, and it was made for a game that launched in Japan in 1999 and in North America in 2002. What replaces the currency is a flow and not a stock, one Normal Summon or Set per turn under Konami's official rulebook, and the permission does not accumulate. The distinction between the two, and what scarcity does under each, is Nothing costs anything's.

6.2Court of Jol's economy is that answer, kept. Nothing costs anything and there is one summon a turn, and neither half of that is an invention. The chain is the other thing worth taking. Both players get real windows, constantly, and a response resolves before the thing it was a response to, and what that system needs in order to matter is a turn whose decisive part sits where a response can reach it. That is what the 2002 game had, and it is the property 2.3 said large engines moved off the table. Court of Jol's response windows are trying to preserve it and not to improve on it.

6.3Yu-Gi-Oh has run the loop in 3.4 harder and for longer than almost any design in the genre, and the reason is the thing 6.1 credited it with. With no resource system there is no mana curve, so a stronger card is immediately a faster card, and nothing structural sits between a new card and the speed of the game. That is the cost of the decision and not a design failure, and the decision bought something real enough that this project copied it. Court of Jol inherited the exposure along with the economy, and what it puts between a new card and the speed of the game is a ceiling on the stats and not a curve on the cost, the conserved statline that The board is not the scoreboard derives. Whether a ceiling holds where a curve did not is not settled, and the evidence so far sits in Math and Simulations.

7.1"We added zones and a third stat and it opened the design space" is true and it is insufficient. The name of the test is section 1's: an epicycle was a repair that was correct on its own, cheap at the time, and expensive only in aggregate and only later. A new dimension on the table is where that bill starts, so the questions are worth asking while the dimension is still a proposal. Three of them get run on a dimension before it is added and not after, and each has a failure signature that can be checked against a rules document.

7.2 Enumeration. The first question is whether the complete interaction table for the new dimension can be written on one page, and whether that page still fits after four more sets. If it fits, the dimension bought expressive range. If it does not, the dimension bought a playtest obligation with no upper bound, and the loop in 3.4 compounds that obligation with every release touching the dimension. The failure signature is that the size of the table is a function of the card pool and not of the rules.

7.3 Rule or cost. The second question is whether the new dimension is capped by a rule or priced by a cost. A cost is defeatable. Anything with a price can be discounted or ramped or generated or cheated, and in a growing pool something that does one of those eventually gets printed, which is the result Nothing costs anything reaches for prices in general. A rule cannot be discounted. "Three zones, hard" is a rule and "extra zones cost 2" is an invitation. The failure signature is that the cap is printed on cards instead of in the rulebook.

7.4 Convertibility. The third question is whether the new dimension can be converted into the others, and whether that conversion compounds. If position converts to tempo and tempo buys more position, the design has a positive feedback loop in its geometry, and a loop in the geometry has no card to ban because it is not in a card. Whether worth compounds at all is the question Draw is not an engine asks of cards, and this is the same question asked of the table. The failure signature is that a player ahead on the new dimension finds it easier to get further ahead on it.

7.5A dimension that fails a question is priced by the failure and not disqualified by it, and the price is frequently worth paying. What the test does is put the price on the page while the dimension is still a proposal, instead of leaving it to arrive the way 4.4 says it arrives now. Running it on extra zones, on a third stat, on movement, and then on this design is What a zone costs (forthcoming).