Your Deck Is a Machine, Not a Collection
You just killed Terror Eel with 11 HP left. The reward screen offers three cards. Every instinct you own says a card is free money, because refusing feels like leaving loot on the table. It is not free. It is a loan taken out against every future draw, and the interest is paid in the fights where the card you actually needed was sitting three cards deeper in the pile.
Here is the mental model that fixes it: your deck is a machine, and each card is a part. The starter deck is a lawnmower engine, ten components, humble but coherent. Every pick either adds a cylinder or welds a paperweight to the crankshaft, and the machine does not care which one looked shinier on the reward screen. A collector asks whether a card is good. A builder asks whether the machine needs this part today. Those are different questions with different answers, and in this game only the second one pays out.
The arithmetic behind the metaphor is draw frequency. In a ten-card deck, you see any specific card once per two shuffles; in a twenty-card deck, once per four. Keystone cards feel this hardest. When Catalyst doubles an enemy's Poison but arrives on turn nine of a fight you needed it on turn five, the card did not fail you, the deck did. This is also why our beginner FAQ states the rule bluntly: a small, focused deck is always stronger, and skipping card rewards is often correct.
Two version-specific notes make the discipline more valuable in StS2 than it was in the original. First, the safety nets are gone: no Dead Branch, no boss-relic gambles, and Afflictions, this game's replacement for Curses, weld themselves onto cards already in the deck, and no shop can strip one off, so pollution that gets in tends to stay. Second, the reward pools are deep (the current build ships 577 cards), which means reward screens offer tempting noise constantly, and the skill being tested every single time is refusal.
A worked example of the machine view. Your Ironclad deck has Anger, Pommel Strike and Shrug It Off doing the early work, and a reward offers a second copy of a decent but redundant attack. Collector logic says take it, it is a good card. Builder logic asks what job is currently unfilled: the deck lacks AoE, so the machine needs a Whirlwind, and the correct action is to skip and keep the deck lean until either a Whirlwind-shaped card appears or the fight schedule stops demanding one.
The rest of this guide is the machine shop manual: a priority order for your first five picks, the courage to skip, the uncomfortable truth that rarity and power are barely correlated, five questions to ask any candidate card, a timeline for when damage matters versus scaling, per-character rules of thumb, the legitimate exceptions, and a short list of cards that age well versus cards that rot. Everything assumes v0.111 beta, and every card name in it comes from our verified database.
The First Five Picks: A Priority Ladder That Works for Every Hero
The opening five card rewards set the machine's topology, and beginners treat them as five independent decisions. They are one decision, sequenced. Here is the ladder, job by job, with the current database's best candidates.
Pick one and two: front-loaded damage. Act 1 is a race against enemies between 42 and 72 HP, and the starter kit does not race well. Ironclad wants Anger, a free attack that pads every hand it appears in, and Pommel Strike, which trades a little rate for a card drawn (two, once upgraded). Silent wants Backstab for its 0-cost burst or Blade Dance to stock the hand with Shivs. Regent wants Astral Pulse, turning 3 Stars into a board-wide 14, with Solar Strike hitting for 9 while banking a Star. Necrobinder wants Scourge, 13 Doom that replaces itself with a draw so the tempo cost is zero, plus Reave paying damage now and hiding a Soul in the pile for later. Defect wants Ball Lightning, 7 damage with a Lightning orb left behind, and Cold Snap, which chips while channeling Frost. The test for these picks: does the deck now kill a 60 HP normal fight roughly two turns faster?
Pick three: the act's demand. Look at where you are heading and take the answer. In Overgrowth that is often AoE for Phrog Parasite's 4 spawned Wrigglers, which makes Whirlwind and Astral Pulse premium. In Underdocks it is more often block density or Weak for the Fossil Stalker and Terror Eel grind, which is where Defy (apply Weak, the best-value mitigation in the game) and Shrug It Off (8 Block plus a draw) earn their slots.
Pick four: the lubricant. Draw and energy are the two fluids every engine runs on. Pommel Strike again is fine. Acrobatics (draw 3, discard 1, and a card the v0.103 patch moved to Uncommon for exactly this reason) is the heart of Silent discard decks. Prophesize draws 6, upgraded 9, and our database tags it take on sight; when it appears on Regent, the decision is over. Alignment converts banked Stars into 2 energy, and Adrenaline hands you draw plus energy in one 0-cost package that fits virtually every Silent deck.
Pick five: the first keystone, only if supported. Inflame (2 Strength, 3 when upgraded) is the premium-value core of every Ironclad Strength deck. Noxious Fumes lays 2 Poison on every enemy each turn as the 1-cost engine of Poison decks. Genesis grants 2 Stars at the start of each turn and defines the Star archetype the moment you see it. Haunt turns every Soul played into 6 lost enemy HP and pivots the Necrobinder engine. Defragment is now the only stable Focus source in the game and sits at rare, which means orb Defects take it on sight whenever it appears.
The ladder flexes with offers; the jobs do not. Damage, answer, lubricant, keystone. When three candidates compete for one slot, give it to the job closest to the top of the ladder, and remember the ladder resets partially each act, because The Hive demands AoE the way Glory demands consistency.
The Courage to Press Skip: Deck Thinning Starts at the Reward Screen
The skip button is the most underused component in the interface, and the habit of not pressing it is the first entry in our tracked list of run-killing mistakes: taking every card dilutes your deck. Skipping feels like losing. Reframed correctly, it is the cheapest card removal in the game, available at every reward, priced at zero gold.
Learn to recognize the four skip situations on sight. The duplicate: you already own two copies of a situational common, and a third copy means drawing your keystone later in every future fight. The redundant: the candidate does a job your deck already does twice; a second defensive skill in a deck that blocks comfortably adds shuffle weight, not safety. The off-plan rare: a shiny card whose archetype is not the one you are building, the subject of a later section, but the short version is that a rare you cannot power is worse than a common you can. The too-late card: raw damage commons offered in Act 3, when your problem is scaling and consistency, not damage per card.
The drill that builds the muscle: after every reward screen, before touching anything, say the fight this card wins, out loud if you have to. "Whirlwind wins Phrog Parasite and Entomancer." "Defy turns Byrdonis's Peck turns into chip damage." If the sentence comes out empty, the correct pick is the skip button, and you have just removed a card from your deck for free. If the sentence is honest, take the card without guilt. The drill takes four seconds and doubles as deck tracking, because a deck you can describe in fight-winning sentences is a deck you understand.
Numbers, for the skeptics. A 15-card deck with one keystone draws it once per three shuffles; three junk picks later, the same deck is 18 cards and the keystone arrives one shuffle later on average, which in a game where elites kill you on specific turns is not a rounding error. This is also why removals matter so much: deleting a starter Strike both improves average card quality and compresses the shuffle cycle. The shop economy guide on this site prices the gold side of that trade; the behavioral side starts here, because every skipped reward is a removal you did not have to buy.
One caution so the lesson does not curdle into dogma: skipping is a tool for keeping the machine lean, not a religion. Small decks are strong, but a 12-card deck that loses to Ovicopter because it skipped every AoE all act was not disciplined, it was stubborn. Thin with intent, skip with a reason, and when a genuinely correct card appears, take it even if the deck is already pretty. The target is not the smallest deck; it is the deck where every card has a job and every job has a card.
Rarity Is Not Strength: Evidence From the Card Pool
Rarity in this game measures how often a card is offered, not how good it is. The beginner materials say it quietly (rarer cards show up less often in rewards), and the card pool proves it loudly. Walk the evidence.
Start at the bottom of the rarity ladder. Neutralize is a Basic card, starter-deck stock, never offered as a reward, and our database rates it the strongest starter in the game: free damage with a Weak attached, every single fight, value that runs all the way to the final boss. Astral Pulse is a common that turns 3 Stars into a full-board 14 and carries Star decks through entire acts. Anger is a common 0-cost attack that quietly fills out hands all run. Prophesize sits one rung up at Uncommon, draws 6 cards (9 upgraded), and is tagged take on sight; card-for-card it outperforms most rares in the pool.
Now the rares. Seven Stars spends 7 Stars for 49 damage to all enemies, a genuine finisher, and in a deck with weak Star income it is a dead card you drew instead of a block. Demon Form is excellent, and taking it commits you to the Strength archetype on the spot, deck plan and all. Void Form ends your turn as part of its effect and belongs only in a Void burst engine; our database appends a warning to it, never mix with Forge cards. Corruption makes Skills free but Exhaust them, which without the Exhaust payoffs (Dark Embrace, Feel No Pain) is a 3-cost card that deletes your own toolkit. Wraith Form, elevated to the new Ancient rarity in StS2, is a legitimate defensive splash any deck can use, and it is the exception that proves the pattern: the rarity label told you nothing in either direction.
The pattern underneath: commons are tools, rares are commitments. A common answers a question your next three fights are already asking. A rare asks a question back, namely whether the rest of your deck is built to power it, and if the answer is no, the rare is the worse card despite the color of its frame. Rarity is a tiebreaker, applied after job, timing, and support, never before them.
Upgrades interact with the same logic. Body Slam is an Uncommon, and its upgrade (cost 1 to cost 0) is the difference between a decent attack and the finisher of a Block deck. Catalyst at rare doubles Poison; upgraded, it triples. When you evaluate a pick, you are often evaluating the upgraded version you will eventually campfire into, and upgrade distance is another axis on which the rarity label stays silent.
Practical takeaway for reward screens: read the three cards in job order, not color order. Ask what each one does for the machine, when it comes online, and what it demands from the rest of the deck. Only if two candidates tie on all three does the rarer card win the slot, and even then, grudgingly.
Five Questions to Ask Any Candidate Card
When the reward screen appears and the ladder does not settle it, run these five questions in order. They take about thirty seconds and they catch nearly every bad pick before it welds itself into your draw pile.
Question one: which upcoming fight does this flip? Name the enemy, not the category. Ovicopter spawns 3 eggs that hatch fast, so AoE flips it. Owl Magistrate telegraphs a Verdict for 33 damage plus 4 Vulnerable, so block density flips it. Vantom opens with 9 stacks of Slippery where every hit deals 1 damage, so cheap multi-hit attacks flip it. Mecha Knight shoves 4 Burns into your hand on loop, so discard and exhaust effects flip it. The Queen chains the opening cards of every turn, capping you at a single Bound play, and the loophole is drawing during the turn itself, which is precisely what Acrobatics does. If the candidate flips none of the fights you can actually see coming, it is fighting hypothetical enemies in a hypothetical run.
Question two: does it effectively replace a starter card? A card that outclasses a starter Strike or Defend at the same job raises average deck quality the moment it enters. Shrug It Off (8 Block, 11 upgraded, plus a draw) is the textbook case on Ironclad; the best common defensive pick for any Ironclad deck, per our own notes. If the candidate is not clearly better than what it would displace from your draws, the upgrade it provides is imaginary.
Question three: what does it need around it? Some cards are self-sufficient; others are connectors that demand a circuit. Tactician is 3-cost energy gain with Sly, meaning it plays itself for free when discarded, so it wants discard triggers like Dagger Throw and Acrobatics before it earns a slot. The Smith wants 4 banked Stars. Haunt wants a steady supply of Souls, which means Reave and the other Soul generators need to already exist. Tempest channels X Lightning and wants energy surplus to feed it. Score the candidate as a team, not as an individual.
Question four: what does it do to the cost curve? Count the cards in the deck costing 2 or more. A 3-energy turn supports roughly one expensive card plus two cheap actions; every premium pick past that ratio starts competing with your own hand for play windows. Echo Form is a ceiling card for every Defect deck and it is also a 3-cost Ethereal that can strand you, which is exactly why our database tags it top upgrade priority.
Question five: is it better or worse by Act 3? Whirlwind multiplies with your Strength stack, so it ages forward. Claw buffs every future Claw, compounding. Backstab's flat numbers stop mattering when enemy health passes 200. Bash's damage rots, but its Vulnerable stays relevant as an enabler for every point of Strength you have stacked. A card that must be good now and a card that must be good later are different purchases; buy the one your run's calendar actually contains.
Five questions, honest answers, and the skip button standing by.
Frontload Now, Scale Later: A Pick Timeline for All Three Acts
The same card can be a correct pick on floor 2 and an incorrect pick on floor 14, because the value of damage and the value of scaling move in opposite directions across a run. Timing your picks to that curve is what separates decks that win acts from decks that win fights.
Act 1 pays front-load. The threats are races: Ceremonial Beast gains +2 Strength every turn, Waterfall Giant detonates on a stack counter, and every elite between 61 and 140 HP is asking whether you can end fights before your Block runs out. Front-loaded damage (Anger, Backstab, Ball Lightning, Solar Strike, Reave) is the correct first priority, and AoE follows it immediately if the act variant spawns boards (Phrog Parasite in Overgrowth, the Ovicopter and Entomancer era ahead in The Hive). Defense in Act 1 means enough Block for specific telegraphed turns, not a Block engine; you are buying life support for the race, not armor for a war.
Act 2 pays engines. Elite health jumps to 200 (Infested Prism), normals gain mechanics (Tunneler's undecaying 32 Block, Ovicopter's eggs), and the bosses grade your ability to produce value every turn for ten-plus turns. This is the act where the lubricants and engines come home: Noxious Fumes dripping Poison onto the whole board each turn, Genesis banking 2 Stars every turn, Alignment converting Stars to energy, Haunt taxing every Soul, Defragment finally making orbs hit like the deck imagined. A useful heuristic for The Hive: every pick should either scale, draw, or answer a board. Raw damage picks past mid-Act 2 are usually the deck telling you it is scared rather than the deck having a plan.
Act 3 pays consistency and finishers. Glory's fights are long, mechanical, and punishing to inconsistency: Test Subject #C10 staggers 600 total HP over three phases, The Queen fights your hand itself through Chains of Binding, and the elites include a 300 HP machine and a 234 HP soul battery. What wins here are upgraded keystones, ways to see them reliably, and a finisher that converts accumulated resources into dead enemies: Body Slam once Block is abundant, Catalyst once Poison stacks, Seven Stars once Star income is proven, Reaper Form for the Doom-by-attack hybrid ceiling. New engine pieces in Act 3 are mostly too late; new answers are never too late, and a card that specifically beats the boss pool (exhaust for Burns, mid-turn draw for Bound) is always in season.
The timeline also disciplines the two classic timing errors. Picking scaling first is the polite one: your deck becomes beautiful by Act 3 and dead by floor 6, because Terror Eel charged 30 HP of tuition while Demon Form was still theoretical. Picking front-load forever is the common one: the deck wins every hallway fight in the game and then loses to every boss in the game. The calendar fixes both: damage while the races are close, engines when the races get long, finishers when the resources exist to feed them.
Per-Character Pick Rules and Every Hero Trap
General frameworks get you 80 percent of the way; each hero has a musculature that changes how the last 20 percent bends. Quick reference for v0.111.
Ironclad: Vulnerable is a full archetype here, not a side dish. Bash opens fights by applying it, every point of Strength multiplies through cheap attacks, and Whirlwind converts stacked Strength into AoE. The trap is drafting Strength payoffs without enablers: Demon Form plus Limit Break is the consistent core, but it wants a deck that survives to turn six. Also watch the Exhaust lane: True Grit (7 Block, exhausts a random card, your choice when upgraded) is the starter motor of Exhaust decks and a fine early pick even outside them, because it thins your deck mid-combat for free.
Silent: every discard matters once Sly exists. Tactician plays itself for free when discarded and turns discard effects into energy, which makes Acrobatics (draw 3, discard 1) and Dagger Throw (draw 1, discard 1) the load-bearing commons of the archetype. Poison runs the other direction: Noxious Fumes generates, Catalyst cashes out, and Snecko Skull (apply 1 additional Poison stack whenever you apply Poison) is mandatory for the archetype if you find it. The trap: Shiv packages and Poison packages dilute each other; commit to one generator type early and let the other lane's cards go by.
Regent: one resource, committed early. Star decks bank (Venerate generates, Solar Strike generates while hitting, Genesis generates every turn) and spend (Astral Pulse for AoE, Seven Stars for the 49-damage ceiling, Alignment for energy conversion); Forge decks feed The Smith's 30-damage payoff and permanently empower the Sovereign Blade. Our database says it twice because players need it twice: never mix Star and Forge cards in one deck. The secondary rule is income math: your Star generation per shuffle dictates how many Star-spenders the deck supports, so count generators before taking the third spender. Prophesize is exempt from all rules; draw 6 is draw 6.
Necrobinder: Souls are a budget, and the archetype dies of both poverty and hoarding. Reave shuffles Souls in, Souls are 0-cost draw-2-then-exhaust tokens, and Haunt converts each play into 6 lost enemy HP; the balance warning in our builds data is explicit, balance Soul generation against spending or boss fights stall. Scourge is the main Doom stacker and never costs tempo, Defy is cheap mitigation that fits anywhere, and the Osty axis (Bodyguard grows its HP, Unleash hits for its current HP, Poke commands it free) scales with how well you feed the hand.
Defect: Focus is scarce and precious now. Defragment is the only stable Focus source after the rework, it sits at rare, and every orb deck takes it on sight; plan around not seeing it rather than assuming it. Orb play is active channel-and-evoke: Dualcast exists to double-evoke Dark orbs on burst turns, Glacier is the Frost turtle's core at 2 cost, and Compile Driver draws per different orb type, rewarding mixed boards. The Claw lane is a separate machine entirely: Claw buffs every Claw for the rest of the combat, so the archetype is about volume and recycle, and it kills faster than anything else in the game when it comes together. Echo Form is the ceiling card of every Defect deck and your top campfire upgrade the moment it joins.
When to Break Every Rule in This Guide
Frameworks are defaults, and defaults are for average situations. Three triggers legitimize rule-breaking, and each has a recognizable fingerprint.
Trigger one: a named, scheduled threat. Rules say diversify and stay lean, but when the map shows two more Hive floors and your deck has no AoE against Ovicopter's 3 eggs, a mediocre AoE is the correct pick over an objectively stronger card, because the question is no longer "which card is best" but "which fight am I about to lose." The same logic endorses narrow Act 3 picks: a card that only beats Mecha Knight is a good pick on a branch that runs through Mecha Knight, and a discard outlet that only answers Beckon statuses is premium the run Soul Fysh sits at the top of your Underdocks map. Name the threat, date the threat, buy for the threat.
Trigger two: a scheduled removal or transformation. Reward screens are not evaluated in a vacuum; they are evaluated against the rest of the run's plumbing. If a shop sits three floors ahead with your Strike-removal gold already banked, slightly looser picks now are subsidized by the removal to come. If Doors of Light and Dark is coming (the Door of Dark removes a card), same subsidy. If Amalgamator lies ahead in The Hive, it will delete 2 starters and hand back an upgraded replacement, which means your deck's endpoint is leaner than its current shape suggests. And when Aroma of Chaos offers to transform a card, the correct play is to have kept a dead pick precisely to feed it, a sentence that sounds like heresy until it wins you a boss.
Trigger three: desperation, honestly labeled. At 25 HP in Act 2 with a thin deck and a brutal schedule, taking the third copy of a good common is right, because the schedule will kill you before dilution ever could. Desperation picks should look different from ordinary picks: defensive skew, immediate effects over engines, anything that converts into survived turns this floor. The label matters because desperation warps judgment both ways; it is the state in which players either grab everything (panic) or skip everything (freeze). Break the rules in one direction, on purpose, once.
What never licenses rule-breaking: color of the frame, fear of an empty slot on the reward screen, and the collector's itch. Those three motivations account for most of the diluted decks in any loss archive.
A closing note on the environment: this is Early Access, v0.111 beta, and the pool moves under your feet. The v0.110 patch redesigned three Silent poison cards outright, v0.107 touched 84 cards in one balance pass, and Defragment's rarity itself changed during the Focus rework. Frameworks survive patches; specific tier lists do not. When a card in your client disagrees with anything written here, trust the client, then re-run the five questions on the new text.
Cards That Age Well, Cards That Rot, and the Final Rule
Every card has a shelf life, and reading it correctly is the quietest advantage in drafting. Two lists, current as of v0.111 beta, and the reasoning behind each entry.
Cards that age well. Pommel Strike, because damage-plus-draw stays proportionate forever and the upgraded double draw only gets better as card quality rises. Shrug It Off, 8 Block and a draw, a defensive staple from floor 1 to the final boss. Adrenaline, 0-cost draw plus energy, fits virtually every Silent deck in every act. Prophesize, 6 cards drawn, 9 upgraded, valuable in exact proportion to how good the rest of your deck is. Alignment, because energy conversion never expires and Star decks only get richer. Claw, which buffs every future Claw and therefore compounds with each copy. Body Slam, worthless in a deck that does not block, a finisher in one that does, and the upgrade to 0 cost is the sharpest single-card jump in the pool. Noxious Fumes and Genesis, engines that pay every turn of every remaining fight. Defragment, the only stable Focus source in the game. Haunt, whose value tracks your Soul supply, which you keep feeding.
Cards that rot. Backstab: excellent floor 2, expensive floor 20, as flat damage always is. Poke stays a fine free Osty command but its numbers stop deciding fights late. Any raw-damage common offered in Act 3 is selling you a past act's product. Tactician without discard density is a 3-cost card that mostly gets held. Seven Stars in a deck with thin Star income is a finisher with no ammunition. Tempest without energy surplus channels a sad little storm. The pattern across all of them: cards that depend on a condition their deck stopped meeting, or cards whose numbers were tuned for enemy HP curves you have outleveled.
Between the two lists sits the upgrade variable, and it can rescue a rotting card or extend a maturing one. Catalyst doubles Poison and triples it when upgraded, so its shelf life is really the upgraded version's shelf life. Bash's damage rots but its Vulnerable never does, which is why Bash keeps a slot in Strength decks that long outgrew its numbers: enablers age differently than damage.
The final rule, the one that subsumes the whole guide: every card must earn its slot by shortening your scariest fight or extending your survival, on the schedule you are actually running. Not the schedule you hoped for when you picked it, the schedule visible on the map right now. A machine built by that rule stays coherent even when the reward screens are generous, especially when they are. Draft with it in Act 1, audit with it before every boss, and the reward screen stops being a slot machine and becomes what it actually is: the workshop where you decide which ten minutes of your next hour of combat you get to design.