What changed, and when

There has never been a way to link to a card. The album shows you the whole set once you have an account, but the moment a card came up in conversation — in the forum, in a message, anywhere at all — there was nothing to point at. You described it, or you screenshotted it.

The collection now has a public address. /cards lists all 105, grouped by rarity, and each one has a page of its own at a name you can read: /cards/stone-golem, not a catalog number. The page has the full artwork, what the card costs, what it does in the engine's own words, what a deck may hold of it, what it costs to craft, and the line of flavour text printed on it. No account, no sign-in, and it works with a card you do not own — because a card's face is not something you earn, it is something the game prints.

Everything on those pages is generated from the game itself rather than written out beside it. If a card is re-costed tomorrow, its page is re-costed in the same build, and a test fails if it is not. That is the same arrangement the Codex already runs on, and it is the only version of this worth shipping: a card reference that quietly drifts out of date is worse than none.

There is a second reason this exists, and it is worth saying plainly. A hundred and five cards with artwork and rules text were, until today, invisible to anyone who was not already here. Now they are findable. If you searched for one of these cards last week you found nothing at all.

Arcana is in open beta, and from today there is a beta program to go with it. We are running it jointly with a handful of communities and clubs: each one gets a code to give its people, and redeeming that code credits 2,000 coins. At 100 coins a pack that is twenty packs, on top of whatever you already hold.

The reason for the size of it is narrower than generosity. A new account starts with 400 coins, which is four packs, and four packs is enough to have an opinion about opening packs and not much else. The part of the game we most need looked at is the part after that — whether a collection with some depth in it produces decks that are worth building, and whether the cards we have priced for that stage are priced right. You cannot tell us that from a starter deck, so the program hands over enough to get past it.

If you have a code, redeem it at the beta page, or just open the link the community published — it fills the code in for you. Case, spaces and hyphens do not matter, and neither does whether your account is new: players who have been here since the first week are as eligible as anyone arriving today.

Two things it deliberately does not do. It pays once per account, whichever code brought you in and however many others you collect afterwards — the grant is a welcome into the program, not a rate. And it is not simply switched on for everyone who signs in during some window, which would have been the easier thing to build. A public site with an unbounded coin faucet on it is a coin faucet for whoever notices first, and a code with a limit on it is the same offer with a number attached that we can actually stand behind.

Nothing about the ordinary economy changed. Packs cost what they cost, the starting balance is still 400, daily and match rewards are untouched, and nothing here or anywhere else in the game can be bought with money.

Until today there was no way to play someone in particular. The queue pairs you with whoever is waiting, and if nobody is, a bot takes the seat after ten seconds — which is the right answer when you just want a game, and the wrong one when you and a friend both sat down to play each other. Two people who wanted the same match would miss each other by a few seconds and each end up playing the computer.

The play screen now has a Play a friend panel. Create an invitation and it hands you a link. Send it however you already talk to the person, and when they open it they see who invited them and a button that puts you both in the same match at once.

The link is deliberately short-lived. It works for fifteen minutes, and only while you stay on the play screen — the moment you close the tab, queue for a ranked match or start anything else, it stops working. That is not a limitation we could have avoided: a match that began while you were away would run its turn clock without you and hand your opponent a win you never saw. The first person through the door takes the seat, and the link is spent.

A friendly match is a real match in every way that matters — the same rules, the same decks, the same clocks — with one exception: it pays nothing. No coins, no rating, no statistics, no quest progress. That is the trade we chose knowingly. A link is a pairing the two of you picked, so a friendly match that paid out would be an unlimited supply of wins for anyone willing to hold two accounts, and every other reward in the game is bounded by a day, a season or a first clear. Rather than police it, we made it worth nothing but the game itself.

Until now the computer opponent worked down a list. Play the most expensive unit you can afford. Cast a spell if it passes a usefulness test. Attack whatever trades best. It was a competent list, and once you noticed it was a list, the game against it was over.

It no longer picks a move. It plans the turn: it tries sequences of actions through the real rules engine, scores the positions they lead to, and plays the best line it finds. That is why it will now spend four mana on two units instead of one, hold the removal in its hand rather than throw it at a 1/1, swing before it burns the spell that finishes you, and stop early when everything else on offer would only make its position worse. Against the same decks, the new opponent beats the old one about two games in three.

It plans from exactly what you can see. The search runs on a copy of the board in which every card you have not shown it — your hand, both decks — is a blank. It cannot plan around a trick it has not been shown, which is the whole reason it can be baited, and the reason a card held back is still worth holding back.

Above Hard there is now Expert, and it does one thing the tiers below it do not: before committing to a turn, it hands the turn over in its head. Your board untaps, it swings, and the wreckage is what its plan gets judged on. That is a different opponent to play against — it walls up instead of racing, kills the attacker that matters rather than the biggest one, and declines the trade that would open its hero. It is bounded on purpose: a fixed thinking budget, no knowledge of your hand, and a model of your next turn that assumes you play nothing from it. It is meant to be hard to beat, not impossible.

Expert opponents are drawn once your rating passes 1250, so nobody meets them on the way in. Easy, Medium and Hard keep their places, and Easy is still deliberately fallible — it overlooks its options, misjudges the board and sometimes gives up on a turn halfway through, which is what that rung is for.

There is now a forum in the game, at the Round Table. Five categories: Announcements, Card Balance, Feedback, Deck Workshop and General. You can read it as soon as you have an account, and post in it once the tutorial is behind you.

The reason it exists is narrower than "players should be able to talk", though they should. Everything this application has said until today has been one-directional. The page you are reading announces what shipped. The Codex states the rules. The home screen reports what you have done. None of them can hear you, and there are questions we cannot answer from data — which cards are annoying rather than strong, which matchup feels unwinnable rather than is, which mechanic nobody understood. Play statistics will never tell us any of that. Asking is the only instrument there is.

So the forum does not open empty. There are threads waiting in it, posted in the game's own voice, and each one ends in a question we actually want an answer to: whether the tribal rally cards have any middle between dead and unanswerable, whether the symmetric sweeps are priced right or simply never cast, what the first hour of this game looks like to somebody who has not played hundreds of matches. If you have an opinion about any of those, the thread for it is already there.

One thing worth learning before you post. Wrapping a card's name in double square brackets turns it into something the reader can open and look at, without leaving the thread. A balance argument that assumes everybody remembers what a six-mana spell does is an argument nobody can check, and this is the fix for that. The composer has a picker if you would rather not type it.

What the forum deliberately does not have: reactions, votes or scores of any kind. On a balance thread a vote count measures how loud a complaint is, not how good it is, and we would rather read the reply. It also sends no notifications, in the game or out of it — it is somewhere you go, not something that interrupts you.

You can edit a post for half an hour after making it, and withdraw it whenever you like. A withdrawn post leaves a marker where it stood, so the replies underneath it still make sense. There is a limit on how fast and how many threads you can start, which exists to stop a script rather than a person, and you should never meet it while typing.

The Core Set has gone from 60 cards to 105. Forty-five new cards — thirty-two units and thirteen spells — join it across every rarity, and every one of them has its own artwork and lore.

The reason for the width was not just variety. Every unit in the game has carried a tribe since the day tribes were introduced, and almost nothing read it: four cards in the whole set cared which tribe stood beside them, and for everybody else it was a word on the nameplate. Undead had three bodies in the entire catalogue, which is not enough to build anything with even if there had been something to build towards. It now has thirteen, and no tribe has fewer than eleven.

Two mechanics give that width somewhere to go. A rally counts: a unit with one has +1 or +2 attack for each other friendly unit of its tribe, and like every other aura it is recomputed from the board as it stands — it grows as the tribe grows and shrinks the moment one of them falls.

A Summon effect is the other half, and it works the opposite way. It fires once, the instant the unit lands, and what it writes is permanent. One kind reads the board and grows the arriving unit by +1/+1 or +2/+2 for each tribe-mate already standing; the other pays it forward, permanently lifting every other unit of its tribe. Nothing is handed back when those units die, or when the unit that cast it does.

That is a deliberate line. Every ability in the game so far has moved attack only, because an aura that granted health would have to decide what happens to a wounded unit when the aura leaves, and there is no answer to that a player could see coming. A Summon has no such problem — it resolves once and is finished — which is exactly why it is the one place health moves for a reason other than damage. It also makes the mechanic a decision rather than a stat line: the order you play your turn in is now part of the card.

Two new spell effects come with them. One permanently lifts every friendly unit of a named tribe, which is what a wide tribal board was missing. The other damages every unit on the table, yours included — the first symmetric sweep in the game, cheaper for the damage than the one-sided ones and a real decision about who is further ahead on board. It is also the answer to a tribal deck that has been allowed to build all game.

Everything above is stated in full in the Codex, which has a new section on Summon effects, generated from the engine as always.

Until now a good deal of the interface was emoji. The Wheel of Fortune was a fairground ride, Judgement was a postal horn, Gold rank was an Olympic medal, and the daily reward was a birthday present. They were placeholders that had quietly become the look of the game.

Every one of them has been replaced by a mark drawn for Arcana. The ten arcana have engraved line-work of their own — the Hermit's lamp, the Tower struck across the crown, Death's scythe — set in the same weight whether they sit in a list or fill a tarot card. The six rank tiers now share one heraldic shield in six metals, so the ladder reads as a ladder instead of as six unrelated pictures.

This is not only a matter of taste. An emoji is redrawn by every operating system and phone, in a palette we never chose: the same screen genuinely did not look the same on Windows, on a Mac and on Android, and at the size a tarot card wants, an emoji is a blurred sticker. The new marks are vectors that take the colour of whatever they sit in and hold their line at any size.

Nothing about how the game plays has changed. The arcana do exactly what they did, the tiers sit at the same ratings, and no card, deck or rank moved.

Trading has been removed. The Trades screen is no longer in the navigation, no new offers can be made, and any offer still waiting for an answer will not be settled.

It was the one place where cards moved for a price the game did not set. Everything else — pack price, pity, the coins a match pays, and now what a duplicate is worth in dust and what a card costs to craft — is a number we tune and stand behind. Two accounts trading freely made all of those numbers negotiable, and the cost was carried by the players least equipped to argue about what a card is worth.

It was also the part of the game we could not supervise. Trading is how alt accounts feed a main one and how cards end up sold outside the game, and answering that properly needs trade review, reporting and moderation — more machinery than the least-visited screen in the game had earned.

The gap trading was meant to fill is a real one, and it now has a better answer: a duplicate you cannot use becomes Arcane Dust, and dust crafts the specific card you are missing. That is slower than a good trade and steadier than a bad one, and it does not depend on knowing the right person.

Nothing was taken from anyone. No cards moved as part of this, every collection and every deck is exactly as it was, and the record of the trades that did happen is kept.

Opening a pack and finding your third copy of a card you may only run two of was the worst moment the game had. Nothing happened. The card was already yours, the slot was spent, and the pack was worse than it looked.

That pull now arrives as Arcane Dust. The pack still opens with the card in it — you see what you drew — but the copy you cannot put in a deck is converted rather than discarded, at 5 dust for a common up to 400 for a legendary.

Dust crafts a specific card. That is the point of it: packs are random and an album is not, and the last few cards of a set are the ones random draws are worst at finding. A common costs 20 to craft and a legendary 1600 — four times what disenchanting the same rarity pays, so dust finishes a collection rather than replacing the packs that build one.

Only genuinely dead pulls convert. A card you do not own, or own below the deck limit, arrives as a card exactly as before.

Arcana has been a game you could finish an evening with and have no particular reason to open again on Tuesday. This is the change that answers that, and it is five things at once.

Every day now has an ascendant arcana: one card of the spread is the same for everybody, dealt into every match played that day. It replaces a draw of its own phase, so a match keeps its length and its early-to-ending shape — the day has a theme, not a script. The day is UTC and the arcana comes from the date itself, so there is nothing to check and nobody's server decides it.

Three quests are offered each day, one easy, one standard, one hard, drawn from a pool of fifteen. They are measured from what your match actually did rather than from what you clicked, and a match has to reach turn four before it counts towards anything — conceding on turn one was never going to be the fast route. If a quest asks for something you have no interest in doing, one a day can be swapped.

The daily reward is now a seven-day ladder that climbs and then starts again, ending on a pack. Missing a day no longer costs you the streak: one miss a week is absorbed silently, and you are told it happened only after it has already been forgiven — a free skip you know about is a thing you plan around, which is not the point of it.

Ranked play grew tiers, from Bronze to Arcane. A tier is a floor as well as a badge: once reached, a losing run costs rating but cannot drop you back out of it. Alongside the season runs a thirty-level pass, flat at 600 XP a level, paid by matches won and lost and by quests. It hands out card backs, avatars and titles rather than coins — prestige is the one reward that can be minted without touching an economy, and the only kind still worth having once your album is full.

The Codex has two new sections covering all of it, with the real numbers generated from the engine as usual.

Until now the rules lived in three places and none of them was a reference: the landing page named four keywords, the tutorial taught the first three turns once, and a card explained itself. If you wanted to know whether Ward stops spell damage, or how many copies of a legendary a deck may hold, there was nowhere to look.

The Codex answers both, along with the shape of a match, combat, abilities, spells and targeting, the arcana spread, deck building and the collection loop. Every number on it is generated from the game engine and checked against it, so it cannot quietly go stale the way a hand-written rules page does.

This page is the other half. From here on, changes that you would notice — balance, new mechanics, anything that touches decks you already built — get an entry here on the day they ship.

Winning or losing used to be a single word and a coin total. The end-of-match screen now shows what the match actually was: damage dealt to each hero, units played and lost, spells cast, and how the arcana sequence ran.

The numbers come from the same match state the engine produced, so they describe the match you played rather than a client-side tally of what it thought it saw.

A legendary and a common differed by the colour of a one-pixel border, which is not much of a payoff at the end of a pack. Every rarity now carries its own frame — different metal, different corner treatment, different weight — legible at album size and unmistakable at reveal size.

Nothing about a card's stats or its rules changed. It is the same card; it now looks like what it is.

Browsing your album showed art, name, mana and stats — and nothing about what a card did. A unit with an aura was indistinguishable from a vanilla one, and a small spell card was art plus a mana gem.

Small cards now carry a compact rules band: the keyword by name, the aura as a short label, the spell as its effect. The album's card detail spells out the keyword's full text, names the tribe, and says so explicitly when a unit has no special ability. Deck-builder search matches keywords and tribes, so searching "taunt" or "beast" finds cards.

The pack reveal also grew by about two thirds, because the rules box on a freshly opened card was too small to read at the one moment you most want to read it.

You can now create an account or sign in with a Google account. Existing accounts are unaffected, and email and password keep working exactly as before.

Which deck was active was easy to lose track of, and activating one gave no feedback. The active deck is now pinned to the top of the deck list, marked, and activation confirms itself.

Arcana is playable in the browser with no download and no payment. It is also unfinished, and being honest about that is part of the deal: cards will be rebalanced, mechanics will be added and removed, and accounts and collections may be reset before the beta ends.

Where a change would affect decks you have already built or cards you already own, it gets an entry on this page.