Ask a game-night table about Scarlet Witch and someone will call her the best hero in the box; ask a tournament bracket and she finishes mid-pack — and the math says both of them are right. Her everyday attack connects on 98% of turns, yet the engine favors her in just 10 of 43 pairings, because nothing in her kit prevents a single point of incoming damage and her flashiest payoffs are rationed by one die face. This guide lives inside that gap.
Your opening turns answer one question: warps or spells? Hex Bolt is the default — a 98% line paying about 6.7 damage a turn is real pressure from the very first roll. But when two spells show up early, take Darkhold instead: it's a 70% line that pays a card, a Combat Point, a Crackle, and a Probability Manipulation all in one swing, and those tokens seed every big turn to come. Never chase crowns early — Bewitch is 24%, Pure Chaos 10% — and since nothing on your board stops incoming damage, a slow start isn't patience, it's just behind. One consolation: Double Trouble pays tokens when you're attacked, so early exchanges hurt less than they look.
The middle game is one decision on repeat: cash the board or grow it? Crackle answers with arithmetic — it adds 1 damage per other status effect you hold, so spending at three-plus statuses buys the full +3, and two Crackles in one turn buy +6. Grow until the count is right, then cash on a hit that's already landing. Pick your line by the defender: Hex Bolt by default; Mind Blast into strong defensive rolls, because 5-plus-a-coin-flip-for-8 is pure damage that ignores their roll entirely — but Crackle is an attack modifier and cannot touch pure damage, so a Mind Blast turn is a hold-your-Crackles turn (it even hands you a fresh one for later). Jinx (49%) is the value swing when the second chaos shows up on its own: about 9 expected damage, plus bonus dice that can pay Probability Manipulation or inflict Reality Warp free. Keep Reality Warp on a cadence — Spellcraft's large straight, Jinx's spell rider, or Conjure in a main phase — so their biggest symbol turn is always one die short.
There is no haymaker coming, so stop waiting for one: she closes with strung-together 6s and 8s wearing Crackle riders, under a standing Reality Warp tax. Two pieces of tech decide close games. First, Conjure's copy mode: copying one of the opponent's helpful status effects — they keep theirs, you gain a duplicate — is strongest when their hand is empty, because they can cancel the copy by discarding a card; time it for the turn they can't pay. Second, ultimate discipline: five crowns is a 1.3% event you never plan around, but if four crowns land on an early attempt, that is the moment for a dice-altering card to buy the fifth — and the same rule guards you in reverse: an ultimate, theirs or yours, is stopped at the dice before activation or not at all. And remember to the end: you prevent nothing, so a tight race rewards spending every token and card on speed rather than insurance.
Her die is three warps, one chaos, one spell, one crown — every reroll is a coin flip for a warp, and that one fact writes the playbook. Keep any three warps and roll the rest at the Hex Bolt tiers: each spare die is 50% to push 4 toward 6 toward 8, which is how a 50% single-roll line becomes 98% across three attempts. When the opening roll scatters, Spellcraft's small straight is your pivot — 59% with best play for 7 damage plus a Probability Manipulation, with the large straight paying 9 and a Reality Warp when it arrives on its own. Mind Blast (54%) wants one of each symbol: hold the crown, spell, and chaos, and let the warp-heavy die find the last piece. Jinx (49%) is only a commit when the second chaos is already showing — chaos is a single face, and hunting a specific one-in-six twice is how turns die. Crowns get no rerolls at all: 24% Bewitch and 10% Pure Chaos are gifts, not plans. Spend Probability Manipulation after the dust settles, as repair — a 4 shifts into a warp or a spell, a 6 shifts down into a spell — but it moves only even dice, and the 5 sitting one step under the crown is odd, so it will never make you a crown. Budget accordingly.
Scarlet Witch's damage plan runs on tokens, not Combat Points — her attacks generate their own fuel — so her CP question is really a card question. Darkhold is her private economy: a 70% two-spell line that pays a Combat Point AND a card AND two tokens in a single ability, an income turn that still advances the board. Bank those points toward dice-manipulation cards: they are the only route to extra crowns beyond raw luck, and altering dice before activation is the only counter to an incoming ultimate. Her own upgrade cards are not in the verified data yet, so no upgrade-path claims here — DiceSlayer will verify them on camera. Until then, one standing rule: upgrades can never be sold back — think hard before sinking early points into permanent value while a racer sits across the table.
You draw one card a turn, hold at most six, and every card sells for exactly 1 Combat Point — so sell without sentiment and shape the hand around her two real gaps. Hold dice manipulation above everything: roll-phase effects that set or re-roll dice are how a four-crown near-miss becomes Pure Chaos or the ultimate — and dice altered before activation are the only way to stop an ultimate, which matters double for a hero with no other answer. Hold one strong damage-prevention card if you draw one: outside the odd Conjure copy, cards are the only defense she will ever have. Watch the opponent's hand before committing a Conjure copy — they cancel it by discarding a card, so a copy into a full hand usually just feeds their discard pile. Specific card names stay out of this guide until her deck is in the verified data; DiceSlayer will verify the list on camera.
Deny her the board, and Crackle starves: every status effect she holds is a waiting point of bonus damage, so force tokens to be spent now rather than stockpiled. Keep one sellable card in hand at all times — that single card cancels Conjure's copy mode and deletes the better half of her once-per-turn value engine. Treat Reality Warp as a planning tax, not a disaster: the borrowed die keeps its number, so straights barely notice — re-plan a symbol-count turn for four dice before you roll. Attack with undefendable lines where you have them: Double Trouble only rolls against defendable damage, so those hits never feed her token income. And chip relentlessly — she prevents nothing, so there is no wrong turn to attack her.
Every turn is a window, because the engine models her damage prevention at exactly zero — there is no wall to wait out. The sharpest moments: right after she cashes her Crackles — the rest of her board stays, but her damage spike is spent until she banks new ones; the early turns before Darkhold and Double Trouble have stocked anything; and any race you can run faster than 6.7 damage a turn — her baseline pace, though stocked Crackles can spike an 8 into 14, so count her tokens first. Her five-crown ultimate is a 1.3% event — respect it only when four crowns appear early, and answer at the dice, never after.
Her reputation is louder than her results — a table favorite some casual lists rank first, while the engine favors her in only 10 of 43 pairings and never above 54.8. A ban spent on Scarlet Witch is usually a ban spent on reputation. The exception: if your three are engine-builders in the Alchemist, Artificer, or Necromancer mold, she holds a real edge against each and the ban starts paying for itself. If you brought one of her answers, keep the ban — Ninja, Barbarian, and Pyromancer beat her on pace, and Iceman beats her worst of all.
A meta call, not a staple. Zero modeled prevention, a 12-point burst ceiling, and 10 favored matchups out of 43 make Scarlet Witch a specialist, and her clearest edges are the slow engine-builders — Alchemist, Artificer, Necromancer — with the rest of her winning scores scattered thin at 51-and-change. If she makes your three, pair her with a fast closer who covers the racer class that beats her, and read her famous split honestly: the kit that tops casual tables pays a real tax when every round is a race, and the engine sides with the brackets.
Her drift problem is subtle: 6.7 expected damage a turn on a narrow spread (standard deviation about 1.6) is a steady clock, but stack-first play — holding back while the token board grows — stretches games, and 3/1/0 scoring punishes stretched games. Default to the Hex Bolt tempo script and cash Crackles on good-enough turns rather than perfect ones. Above all, treat crown chasing as clock poison: a Bewitch hunt that costs two Hex Bolt turns trades about 13 expected damage for 9, and in Swiss it can trade a win for a draw.