Turn 12, Topdecking Defends: What a Fat Deck Actually Costs
Picture the loss before you learn anything from it. You are playing Poison Silent into Knowledge Demon. The boss heals itself and gains permanent Strength every round it stays alive, your Noxious Fumes went down on turn five instead of turn one, and Catalyst, the card that doubles your Poison stacks and triples them when upgraded, is still somewhere in the bottom third of a 34-card deck. Nothing in your deck is bad. You took strong cards at every reward screen. The problem is arithmetic, not card quality, and the arithmetic says your deck has become a machine for delaying your own best cards.
The model that fixes this: a deck is not a collection, it is a probability distribution over five-card hands. Your win condition is drawing the right five cards on the turns that matter. Every card you add that does not improve those odds makes every future hand worse on average, and every card you delete makes every future hand better. That is the entire theory of thinning, and the game's own FAQ endorses the conclusion in one line: a small, focused deck is always stronger, and only take cards serving your build.
Run the failure with real numbers. Noxious Fumes (1-cost Power, 2 Poison to ALL enemies at the start of each turn, 3 upgraded) is the engine of Poison Silent. At the base five cards per turn, a 20-card deck is fully drawn by the end of turn four, so the worst-case Fumes turn is turn four. In a 34-card deck the worst case is turn seven. On a ten-turn fight, the gap between Fumes on turn one and Fumes on turn four is three start-of-turn applications, six Poison per enemy that you never banked, before counting anything the boss healed back. The card did not get worse. The deck made it arrive late, and late scaling is strictly smaller scaling.
The same tax lands on finishers and on answers. Catalyst is worth exactly one specific turn, the turn your stacks peak. Body Slam deals damage equal to your Block, so it wants the turn your Block peaks. Adrenaline (0 cost, draw plus energy) wants the burst turn. Alignment wants the turn you convert banked Stars. In a fat deck these cards surface on the wrong turn at a rate you can feel across a run, and every wrong turn converts directly into HP lost or damage not dealt.
So before any technique, adopt the reframe that makes the techniques work. When you open the deck screen, stop asking which cards are good. Ask instead: when this deck shuffles, what are the odds my best card is inside my first ten, and what did I do today to improve them. Skipping a card reward answers it. Deleting a starter answers it. Taking a fun fifth rare does not. The rest of this guide is the math behind that instinct, the priority order for cuts, the shop and event economics of removal, and the point at which thinning stops paying.
Opening-Hand Odds and Cycle Timers: The Actual Numbers
You draw five cards a turn in this game, before any draw modifiers. Everything else follows from that constant. Four numbers matter: opening-hand odds, the odds a key card shows up by a deadline turn, full-cycle time, and how many cycles you get per fight.
Opening-hand odds first, with one copy of a key card. In a 20-card deck, five random cards give you a 25 percent chance of it being there (5 out of 20). In a 35-card deck that drops to 14 percent. With two copies, the numbers jump: two copies in a 20-card deck means a 45 percent chance at least one lands in the opening five, while the same two copies in a 30-card deck give only 31 percent. This is the quiet power of duplication in thin decks and the quiet lie of duplication in fat ones: the second copy of Pommel Strike buys real consistency at 18 cards and almost nothing at 32.
Full-cycle timers are the starker number. A 20-card deck is completely seen by the end of turn four at base draw. A 30-card deck takes to turn six, a 35-card deck to turn seven. The practical reading: in a thin deck, every card you own is guaranteed to appear in any fight that reaches that turn count, so your one Catalyst or one Echo Form has a floor under it. In a fat deck there is no floor, only odds, and long boss fights are exactly where you cannot afford to miss. A typical elite runs six to nine turns and a boss eight to twelve, so a 20-card deck cycles twice to three times per fight while a 35-card deck cycles once, maybe one and a half. Two cycles means two chances to draw Demon Form and two chances to redraw it after playing it. One cycle means one chance, and if it buried, it buried.
One reshuffle nuance worth internalizing: when your draw pile empties, the discard reshuffles, so on average any specific card sits half a deck away at any moment. Half of 20 is ten cards, two turns. Half of 34 is seventeen cards, over three turns. That is the whole difference between decks that feel smooth and decks that feel soggy, expressed in turns.
A caution about draw cards solving this for you: they help, and some are excellent, Prophesize drawing 6 (9 upgraded) is flagged as take-on-sight in our database, and Acrobatics drawing 3 (4 upgraded) is the Sly engine's core. But every draw card is itself a card occupying deck space and itself subject to being drawn late. Draw density raises your odds; it never changes the base rate at which your engine pieces surface. The only action that changes the base rate is shrinking the deck. Use draw to make a 20-card deck feel like 14, not to make a 34-card deck feel acceptable.
Repricing the Starters: Not Every Strike Deserves the Knife
Generic advice says delete your starters. StS2 makes that advice wrong in both directions, because the starters in this game are wildly unequal, and some of them are cards you would happily pick from a reward screen. Price each one before you cut.
Start with the outliers the database itself calls out. Neutralize (Silent, 0 cost, 3 damage plus 1 Weak) is described as the strongest starter card in the game: free damage plus the best-value mitigation debuff, every fight, all run. Falling Star (Regent, 0 cost, 8 damage plus 1 Weak and 1 Vulnerable) is called the strongest starter Strike in the game, and it is better than most commons. These two are not removal targets, ever. Removing a Neutralize to thin a Silent deck is deleting a free Weak application you will miss in every elite fight from Byrdonis to Soul Nexus. Bash is not filler either: it applies Vulnerable, which is a full archetype in StS2 and the opener of the Ironclad Vulnerable engine, especially once Paper Phrog pushes Vulnerable from +50 percent to +75 percent attack damage. Unleash (Necrobinder) scales with Osty's current HP, so it grows with your Bodyguards all run. Zap and Dualcast are the literal foundation of every orb deck.
So the actual removal order is: generic Strikes first, because 1 energy for flat early-game damage is the worst card in any deck past floor 8, then redundant Defends once your block density is online from real cards, and only then anything else. Even that order has one big exception. The Strength Scaling build notes describe Hellraiser, a card that auto-plays every Strike you draw for free. In a Strike-engine Ironclad, Strikes are ammunition, not filler, and their removal value collapses: the engine converts each one into free damage. The same logic protects 0-cost cards that replace themselves: Necrobinder Souls are 0-cost, draw 2, then exhaust. They thin themselves out of the problem mid-combat, so they never count as bloat even though they enter your deck.
Two compression tricks change the math. The Amalgamator event removes 2 Defends and grants an Ultimate Defend, or removes 2 Strikes and grants an Ultimate Strike, which is two starter deletions plus an upgraded replacement in one event, and the community consensus is almost always take it. And the Doors of Light and Dark event offers a Door of Dark that removes 1 card versus a Door of Light that upgrades 2 random cards; targeted removal beats random upgrades in nearly every deck state, which tells you how the game itself prices a deletion.
The summary rule: before every removal, ask what the card does at floor 15, not floor 3. Neutralize still applies Weak at floor 15. A vanilla Strike is still 6 damage for 1 energy at floor 15. Cut by future value, not by origin.
The Reward Screen Is a Test: Five Rules for Pressing Skip
The most common deck-building mistake in this game is written into the official mistakes list: taking every card, because it dilutes your deck, and skipping card rewards is often correct. Skipping feels like falling behind, so most players need actual rules before they will do it. Here are five, each with the failure it prevents.
Rule one, name the slot. Before taking any card, state what job it fills in this specific deck: the AoE slot, the block slot, the draw engine, the finisher. Anger (0 cost, solid damage) is a fine card that fills no slot in a Demon Form deck that already swings for 30. If you cannot finish the sentence, this card does X for my deck, the pick is a skip regardless of rarity. Rarity is quality per card; deck strength is quality per draw, and only the second one wins fights.
Rule two, the third-copy rule. The first copy of an answer is power, the second copy is consistency, the third copy is bloat. Two Shrug It Offs in an 18-card deck is a 45 percent chance of one in the opening hand. A third Shrug It Off adds 6 block of availability while diluting every other draw. The exception is engine pieces that die to removal pressure or feed exhaust plans, but as a default, stop at two.
Rule three, the act pool can overrule the plan. If you rolled Overgrowth, Phrog Parasite dies on turn one of its second phase to area damage, so an off-plan Whirlwind is a correct pick for a Strength deck. If you rolled Underdocks, Phantasmal Gardener punishes multi-hit, and single-target damage is premium. Answer cards for the elite pool in front of you outrank archetypal purity, and one flexible answer taken early is worth more than the fourth engine card taken late.
Rule four, zero-cost is not free space. Backstab, Poke, Anger: 0-cost cards cost no energy but still occupy a draw slot and still dilute your odds of seeing your finishers. Take a 0-cost card when it replaces a worse starter in the damage curve or feeds an engine (Claw chains, Soul generators, Shiv packages). Do not take it because the price tag says zero.
Rule five, tighten after a defining rare. The turn you take Genesis or Demon Form or Corruption, your deck gains a thesis. Post-thesis, only cards that serve the thesis enter: Star generators after Genesis, Strength support and Vulnerable after Demon Form. The five card rewards after a defining rare are where decks quietly die, because every tempting off-thesis rare looks free and none of them are.
Track the result, not the feeling. A 14-card deck that skipped twice beats an 18-card deck that took both fillers in every fight for the rest of the act, and the gap compounds every floor you climb.
Removal Economics: Shop Prices, the Inflation Tax and Free Deletions
The single best gold sink in this game is not a card, it is the removal service: the mistakes list ranks removing a starter Strike or Defend ahead of buying cards, and the FAQ repeats it. The question is when to pay, how much, and in what order to spend the free deletions the game hands out.
Shop timing first. A removal on floor 7 improves every draw for the remaining 80 percent of the run; the same removal on floor 30 improves almost nothing. Pay early, pay on the first shop you see with gold to spare, and treat a second Act 1 removal as a luxury that competes with a key relic purchase. Two budget facts from the patch history shape the math: v0.102 turned Ascension 6 into Inflation, where removal service starts at 100 gold and climbs 50 per use, and v0.103 cut shop relic prices by 25 gold, which changed the relative price of relics versus deletions but not the priority order, because a deleted Strike pays out every fight while a mediocre relic pays out sometimes. Under Inflation, the discipline sharpens: at 150 and 200 gold per deletion, you cannot delete your way out of a bloated deck, so the skip discipline in the previous section becomes the real economy.
The event pool contains most of the free deletions, and knowing them changes routing. The Amalgamator converts 2 starters into 1 Ultimate replacement, the best compression deal in the game, near-always take it. Doors of Light and Dark: the Door of Dark removes 1 card, and community consensus prefers it over 2 random upgrades, because targeted beats random. Dense Vegetation lets you push through for HP to remove 1 card, correct when healthy per the event guidance. Wellspring's Bathe option removes 1 card at the price of a temporary curse, worth it when you are short on removal options, which on Inflation ascensions is always. Notice that these events make event nodes partially function as removal services, which matters when you are choosing between a shop path and an event path with a Strike problem.
Priority order for any deletion, in sequence: worst generic Strikes, then surplus Defends once real block cards exist, then off-archetype filler you took before the deck had a thesis, then dead rares. Yes, dead rares: a Void Form in a committed Forge Regent deck is a 3-cost dead draw every shuffle, and keeping it because it is rare is exactly the trophy-collecting instinct that bloat kills. One hard limit to remember: Afflictions are not removal targets at all. They replace Curses in StS2, they attach to cards you already own, and they cannot be removed at shops. Prevention is the only cure, so treat Affliction-priced event rewards and Calling Bell's gamble as more expensive than they look whenever your deck is thin, since a polluted card in a 15-card deck is one fifteenth of every hand you draw for the rest of the run.
Remove or Transform: Astrolabe Math and When a Fresh Roll Beats a Smaller Deck
Removal is not the only way to fix a bad card. Transform effects roll a fresh card from your character's pool in its place, and choosing between the two is a real decision with a clean rule behind it. Removal raises the quality of every remaining draw by subtraction. Transform trades a known bad card for an unknown new one, and the unknown has a price and a payoff.
Remove when the deck is big and the target is a known quantity. A vanilla Strike at floor 10 is a known-bad card: deleting it is guaranteed value, a certain improvement to the average draw. When your deck sits above 20 cards, certainty beats variance, and the transform roll might hand you another card you do not want, at which point you have spent a transform to stand still. Transform when the deck is already small and the problem is a card that is dead in context rather than weak in general. A 15-card Regent deck holding one Forge card in a committed Star build does not need fewer cards, it needs that slot to stop being dead, and a fresh roll from a pool containing Astral Pulse, Solar Strike and Prophesize has decent odds of producing something playable. Transform is also strictly better when your character pool is stacked: late-act Regent and Silent pools are deep enough that the floor of a transform is acceptable, while a floor-4 Ironclad pool has more misses in it.
The Astrolabe (Ancient relic) is the flagship of this logic: transform 3 cards and upgrade the result of each on pickup. It is effectively three deletions that each pay a bonus, an instant deck-quality spike, and it is strongest exactly when you still have three or more starters lingering, which for most runs means taking it in Act 1 or Act 2 rather than holding out. Aroma of Chaos offers the single-card version with a choice: transform 1 or upgrade 1. The event guidance is the correct heuristic, transform if you have a dead card, otherwise upgrade, because an upgrade on a live engine card (Inflame going from 2 to 3 Strength) is guaranteed while a transform on a live card is unnecessary risk.
The campfire version of this decision is the most common one: upgrade or something else. The FAQ rule still holds, if the next fight will not kill you, upgrade the key card, and some upgrades are worth more than a removal anywhere on the curve. Defragment going from 1 to 2 Focus doubles a Frost orb's block contribution for the rest of every fight. Body Slam going from 1 cost to 0 changes the card from situational to automatic. When a single upgrade multiplies a card's output, it outranks deleting a Strike, and the sequencing rule falls out naturally: upgrade the multiplying card first, then resume deletions.
One practical note: transform effects and removals both count against the same underlying resource, your stock of bad cards. If an Astrolabe or an Aroma of Chaos handled your worst three, redirect shop gold from removal service toward relic purchases, because the shop's marginal value flipped the moment your deck stopped carrying known-bad cards.
How Thin Is Too Thin: Healthy Deck Sizes by Engine
There is no universal healthy deck size, and anyone quoting one number for all decks is selling a slogan. There are healthy ranges per engine type, derived from what each engine needs to see per turn and what fights demand. Four archetypes from the current build illustrate the spread.
Interchangeable-card engines sit smallest: 15 to 20 cards. A Strength Ironclad before Corruption, or a Frost turtle Defect, mostly wants any attack plus any block on curve, and every card in the deck is roughly as good as every other. Consistency here comes cheap, and thinning pays double, because the fifth Strength card adds nothing while the removal of a Strike improves every draw. Distinct-piece engines sit at 18 to 24: a Sly Discard deck needs Sly payoffs, discard outlets (Acrobatics, Dagger Throw) and energy (Tactician converting discards into free energy) in working proportion, and a Star Burst deck needs its generators and its spenders in a ratio the build notes spell out: your Star income per shuffle dictates how many Star-spenders the deck supports. These decks cannot go ultra-thin without breaking the ratio, because removing a generator to shrink the deck starves the spenders.
Self-thinning engines can draft wider, 20 to 26, because combat itself trims the deck. Exhaust Block runs Corruption making skills free but Exhausted, plus Dark Embrace drawing off each Exhaust and Feel No Pain blocking off it: cards leave during the fight, so a 24-card deck plays like a 15-card deck by turn six. Necrobinder is the same story, since Souls exhaust after drawing 2. For these decks, bloat is partially self-correcting, and skip discipline can relax slightly in favor of more engine redundancy. Above roughly 30 cards, any deck needs an explicit written reason (mass exhaust fuel, Soul generation rate, event-churn strategy), and without one you are collecting, not building.
Now the other edge: over-thinning is real, and this game punishes it three ways. First, status pressure. Mecha Knight stuffs 4 Burns into your hand every loop, Haunted Ship shoves 2 Wounds per cycle, Soul Fysh seeds Beckon cards that cost 6 unblockable damage each if left unplayed. In a 12-card deck, four Burns are a third of a shuffle polluting every hand; in a 22-card deck they are a fifth, and the difference is whether your engine hands still function. Second, block density: you want roughly one block card per five deck cards, and cutting Defends past that floor makes Terror Eel and Soul Nexus unwinnable no matter how sharp your damage is. Third, expensive topdecks: Demon Form costs 3, Echo Form costs 3, Corruption costs 3. Ultra-thin decks redraw the same expensive card and cannot deploy it faster than the energy curve allows. The healthy deck is the one where every shuffle produces a playable hand at every point in the fight, and the moment you can name a fight where your deck stops doing that, you have found the direction to adjust, and which direction is a function of the fight you just lost.
What StS2 Changed: No Dead Branch, Afflictions and the New Removal Meta
If your thinning instincts were tuned on the first game, several of them need recalibration, because the v0.111 build quietly removed or rebuilt the systems they depended on. The mistakes list says it directly: playing like it is StS1 misleads, there are no boss relics, no permanent Focus, no Dead Branch.
Dead Branch first, because it anchored an entire philosophy. In StS1, exhaust decks could draft huge and let Dead Branch refill every exhaust with a random card, so bloat had an exit. That relic does not exist here, and permanent Focus scaling is gone from Defect too, replaced by active channel and evoke play. Consequence: exhaust is a thinning tool inside combat only. Corruption, True Grit and Soul tokens genuinely shrink your effective deck mid-fight, but nothing converts exhaust into infinite fuel, so the drafting discipline behind the engine still matters.
Afflictions are the bigger structural change. They replace Curses, they attach to cards you already own, and the glossary is blunt: they cannot be shop-removed. In StS1 you could buy your way out of a curse; here you cannot buy your way out of anything. This inverts the removal meta from cure to prevention. Affliction-priced event rewards, and the Calling Bell gamble (a unique Affliction for 3 relics), cost far more in a 15-card deck than in a 30-card deck, because a polluted card is a permanent share of every future hand. Map guidance already tells you to steer clear of Affliction events; the thinning math explains why the steer matters more the better your deck gets.
Boss relics are gone too, replaced by Ancients who offer Blessings at act start, and several Ancient-tier relics interact directly with deck quality: Astrolabe transforms and upgrades 3 cards, Yummy Cookie upgrades 4 cards on pickup, Arcane Scroll grants a random rare card. The Blessing system means deck-quality spikes arrive on a different schedule than StS1 boss relics did, usually earlier, which rewards committing to a thesis sooner and thinning around it sooner.
The rhythm around thinning also changed. Floor 6 stopped spawning elites in v0.101, so floors 1 through 6 are a safe gold-and-cards phase that conveniently funds the first shop removal. v0.103 moved Acrobatics from Common to Uncommon, making draw density scarcer for Silent and raising the value of not needing draw in the first place, which is to say the value of a smaller deck. v0.107 swept up to 84 cards in balance, v0.110 redesigned the Silent poison package, and v0.109 added Neow relics with Quest cards. Specific numbers drift patch to patch, but the ratios in this guide are version-proof: five cards per turn, one block card per five deck cards, and the compounding value of deleting a card you will never want to draw.
Five Draft Screens, Answered: Worked Thinning Cases
Case one: Silent, floor 8, deck at 18 cards with two Sly payoffs in hand, reward screen shows Backstab, Acrobatics, Blade Dance. Take Acrobatics. It is the draw-discard core of the Sly engine, it was promoted to Uncommon in v0.103 precisely because it is that good, and Dagger Throw-class outlets are what convert Sly cards from hand-fillers into free plays. Backstab is generic damage that dilutes, Blade Dance builds a Shiv package your deck is not currently committed to.
Case two: Ironclad, floor 9, deck at 26 cards with three vanilla Strikes remaining, shop offers removal service and Membership Card (everything 50 percent off). Buy the removal. Membership Card pays on future purchases, but it produces zero cards and zero draw improvement today, and your deck has a measurable disease: three Strikes are three guaranteed bad draws per cycle. Delete one, and remember the Inflation math if you are past Ascension 6, where each subsequent removal costs 50 more, which argues for front-loading deletions before the tax compounds.
Case three: Regent, floor 14, committed Star build at 21 cards, reward shows Prophesize among three mediocre commons. Take Prophesize and skip the next two screens unless a Star generator appears. Draw 6 (9 upgraded) is flagged as one of the strongest draw cards in the game, and in a 21-card engine deck it effectively shows you a third of your deck in one card, which stabilizes the generator-to-spender ratio the Star engine lives on. This is the exception to bloat: a card that raises the quality of every future turn earns its slot.
Case four: Necrobinder, floor 11, deck at 15 cards after an Amalgamator compressed two Defends, reward offers three mid-tier commons (generic attack, generic block, a second Haunt). Skip all three. A 15-card Doom deck with its engine installed wins by executing at thresholds, and every filler card delays Scourge and End of Days. The second Haunt is tempting, but the first Haunt already converts every Soul into 6 damage, and your Soul count, not your Haunt count, is the bottleneck. This is also the screen where most players break, because skipping three times in a row feels like refusing free money, when it is refusing three future bad draws.
Case five: Defect, floor 16, deck at 24 cards, Echo Form appears. Take it, then skip the next two screens unless a 0-cost orb support card appears. Echo Form is the ceiling card of every Defect deck and the top upgrade priority, and a 24-card deck is not so fat that a 3-cost Ethereal power breaks it, but it is exactly fat enough that adding two more cards after it starts choking the draws that find it. The pattern to internalize across all five cases: defining cards get taken instantly, support gets taken only while the ratio holds, and everything else gets skipped with a clear conscience. Your deck is not a museum of good cards. It is a draw distribution, and every one of these decisions moved the distribution toward the five cards you actually want in hand on the fight's biggest turn.