Goodbye Power Levels, Hello CRISPI

Overview

Meet the CRISPI Score

Today, DeckCheck officially retires power levels and introduces CRISPI. Crispy like a crispy potato chip. Shoutout to user Legobumb for coming up with the playful name. It's a clear, open rating that tells you exactly how your deck performs and, more importantly, why.

If you read the announcement post, you already know the story: power levels were competing with brackets, and the number was opaque. The CRISPI Score is the tuning tool you asked for.

This post is the full breakdown: what each attribute measures, how scores are assigned, and how to move them.

Transparency

July 2026 Update: New Engine, Recalibrated Scale

A quick, honest note on how these scores are produced. Until today, CRISPI ratings were assigned end-to-end by AI. That got us surprisingly far, but the reasoning was a black box and the same deck could scan a little differently twice.

As of today, CRISPI runs on a new, largely code-based engine. Consistency and Interaction are pure counting — deterministic code applying the exact rubrics on this page to verified card data. Speed and Resilience are computed the same way, with exactly two judgements left to the AI, because they genuinely require judgement: the fundamental turn, and how commander-dependent the deck is. Same deck in, same score out.

This system is young, and I want to be upfront about that. The engine is only as accurate as the card data underneath it, and across 30,000+ cards there will be edge cases — a card that should count as interaction but doesn't yet, a combo priced wrong. The upside of code over vibes: every fix is permanent. When something scores wrong, I correct the data and every deck scores right from then on. If your deck's score looks off, tell me in the Discord — or add a note to your scan explaining what the deck does, and the analyst will re-evaluate with that context. Expect steady tuning over the coming weeks.

July 12 recalibration: we re-derived the middle of the scale against real win-rate data (59 precons with tracked tables, plus tuned decks and cEDH anchors). The short version: raw counts — how many threats, how much recursion, how many interactive cards — turn out to be nearly identical between precons and far stronger decks, so counts now position a deck within its class and the qualitative separators (tutor access, instant-speed interaction, combo assembly, board-level protection) are what move a deck between classes. Precons now center on a CRISPI score of about 5 instead of the low 6s. If your deck's score dropped a point without a card changing, this is why — the scale moved, not your deck.

The System

How the CRISPI Score Works

CRISPI stands for Consistency Resilience Interaction Speed Performance Index. Every deck is rated on the four attributes, each scored 1–10 in quarter-point steps — the whole number is the rubric row, the decimal is where you sit inside it. The Performance Index (the CRISPI Score) is the average of the four, snapped to the same quarter-point grid (exact midpoints round up) and displayed to two decimal places.

CRISPI Score = ( Speed + Consistency + Interaction + Resilience ) / 4

The Black Box is Gone: The best part? CRISPI is completely open-source logic. It is no longer an AI "black box." If you wanted to, you could sit down with your decklist, a calculator, and the rubrics below to assign your own exact CRISPI score. Our AI doesn't use secret vibes or hidden math, it just does the counting and calculating for you in seconds instead of hours.

If your deck scores Speed 7, Consistency 6, Interaction 7, and Resilience 5, your CRISPI is 6.25 — and because you can see each component, you know exactly which stat to target.

Methodology

Ceiling-Anchored Scoring

Every CRISPI score is anchored to the top of the format: cEDH — the most optimized version of the same ruleset everyone plays. Tournament-proven decks define what 8, 9, and 10 look like in concrete, verifiable terms, and every deck below that ceiling grades on the same scale. When the system says your Interaction is a 6, that number sits on the same scale that rates a fully optimized Blue Farm list at 10.

The Composite

Putting It All Together

Once the four attributes are scored, the CRISPI Score is their simple average. But it's worth understanding what the score is and isn't good for.

What CRISPI is good at

• Sorting and filtering—find decks in your power range quickly.
• Tracking changes—did your card swaps improve the deck? Re-analyze and compare.
• Comparing within a bracket—both decks are Bracket 3, but which is more tuned? The CRISPI Score tells you.

What CRISPI is not good at

• Predicting head-to-head outcomes. A deck scoring 9/3/3/9 (fast but fragile) is nothing like a deck scoring 6/7/7/6 (balanced). Both produce a 6.25, but they'd have a very different game against each other.

For pregame conversations, think about bracket (first) + attribute profile (second) + CRISPI Score (third).

Brackets

How Brackets Are More Accurate With CRISPI

DeckCheck's bracket calculation already enforces the official rules — Game Changers, combos, mass land destruction, extra turns, tutors. However, those rules can't catch if a deck complies with the speed restrictions of each bracket. CRISPI guardrails catch those decks using two numbers—the Speed rating and the overall CRISPI Score—and either one on its own can bump up a deck into it's proper bracket.

Here's the idea behind the Speed floors. A Speed rating comes from a goldfish game: how fast the deck wins when nobody interferes. Real four-player games run slower than that — opponents fight back, and you have to eliminate three of them — so a deck's real win usually lands a turn or two after its goldfish turn. Meanwhile, the official rules say how long games in each bracket are expected to last: Bracket 1 games run to turn 10 or later, Bracket 2 to turn 9 or later, Bracket 3 to turn 7 or later, and Bracket 4 to turn 5 or later. Each floor asks one question: even after slowing the goldfish win down by those extra turns, does this deck still end games earlier than its bracket allows? If yes, it gets bumped up:

• Bracket 5 Floor: Speed 9+ OR CRISPI 8.5+. A Speed 9 deck goldfishes a win by turn 3. Even slowed down by a real game, it's still winning before turn 5 — too fast for Bracket 4.
• Bracket 4 Floor: Speed 8+ OR CRISPI 7.0+ OR Consistency 7.5+ AND Interaction 7.5+. A Speed 8 deck goldfishes a win by turn 4. In a real game that's still a win before turn 7 — too fast for Bracket 3. The Consistency + Interaction pairing catches a different deck: one that isn't fast, but runs a premium tutor suite alongside premium instant-speed interaction — the signature of a deck built to compete, whatever its Game Changer count.
• Bracket 3 Floor: Speed 6+ OR CRISPI 5.0+. A Speed 6 deck goldfishes a win by turn 6. In a real game that's still a win before turn 9 — too fast for Bracket 2.
• Bracket 2 Floor: Speed 5+ OR CRISPI 3.5+. A Speed 5 deck goldfishes a win by turn 7. In a real game that's still a win before turn 10 — too fast for Bracket 1.

One last detail: half-step Speed ratings get the benefit of the doubt. A rating like 8.5 means the deck's typical win lands somewhere between turn 3 and turn 4 — sometimes faster, sometimes slower. For bracket purposes, we count the deck as the slower of those two turns. So an 8.5 counts as a turn-4 win: fast enough for the Bracket 4 floor, but not for Bracket 5's. Likewise, a 7.5 counts as a turn-5 win and stays below the Bracket 4 floor. A deck only trips a floor when even the slow end of its rating is too fast for the bracket.

The CRISPI floors follow the score bands the scale is built around: precons — the definitional Bracket 2 experience — center on a CRISPI score of about 5, upgraded and tuned decks own 5–7, maxed-out non-meta decks 7–8.5, and cEDH lives above 8.5. The strongest recent precons deliberately land at the very bottom of Bracket 3 — that is true of them at real tables, too. The Bracket 4 floor's third input deserves its own explanation. Bracket 4's official intent is “Lethal & Consistent” — efficiency as the primary deckbuilding metric — and its minimum-turn rule is a floor, not a speed requirement: an optimized deck that happens to win on turn 7 is still a Bracket 4 deck. So a deck pairing a premium tutor suite (Consistency 7.5+) with a premium instant-speed interaction suite (Interaction 7.5+) gets the Bracket 4 floor even when its overall CRISPI sits below 7.0. That pairing is how a competitively built deck reads when it holds itself to three Game Changers — and it is exactly the deck the “there should be a bracket between 3 and 4” complaint is about. Speed and Resilience deliberately stay out of that test: fast decks are already caught by the Speed floors, and a grindy-but-fair battlecruiser deck with good rebuild belongs in Bracket 3, whatever its Resilience score. And because every input is visible to you, if your deck gets bumped you can see exactly why: too fast, too strong overall, or built on a competitive core.

These checks only bump up, never down (since no one finds issue with the person playing a weak deck that does nothing all game). A low-power deck that contains four Game Changers is still Bracket 4 because of the card restrictions (though, DeckCheck will still flag the deck making note of it likely being able to play in a lower bracket).

Reference

Complete Scoring Rubrics

Consistency — Full Rubric

What it measures: how reliably the deck finds its key cards and executes its game plan through the randomness of a 100-card singleton format. Tutors, card draw, redundancy, and your commander's own utility all feed it.

One honest caveat: Consistency is the one attribute that's somewhat detached from raw power. It measures how reliably your deck does its thing — not whether that thing is any good. A deck can execute a bad plan with perfect consistency and earn a high score doing it. Read Consistency as “how well does this deck follow its own script”; the other three attributes tell you whether the script actually wins games.

How the columns combine: the draw table above and the tutor ladder below are joint requirements — a score is earned only when BOTH totals qualify, so whichever column is weaker holds the score down. The tutor column now spans the whole scale, because search access is what actually separates deck classes: precons carry 8–12 tutor points, tuned decks 20–50, cEDH decks 60+. A tutorless deck reads 3.5 on the tutor column no matter how strong its draw suite is — raw draw totals do not separate classes; guaranteed access does. Tutor ladder (a total ON an anchor reads that score; totals between anchors interpolate linearly): 0 pts → 3.5 · 12 → 4.5 · 20 → 5.5 · 24 → 6.25 · 32 → 7 · 44 → 8 · 56 → 9 · 68+ → 10.

Tutor Density is scored on a scaled point system:

• 6 points — Premium true tutors, CMC ≤2, unrestricted (Demonic Tutor, Vampiric Tutor, Imperial Seal, Enlightened Tutor, Mystical Tutor, Worldly Tutor, Gamble)
• 6 points — Repeatable tutor engines: a permanent that tutors every turn, unconditionally (Survival of the Fittest, Birthing Pod, Fauna Shaman) — premium regardless of CMC or how narrow the target, because the engine finds a piece every single turn. Combat-conditioned tutor engines (attack triggers — Zur the Enchanter, Armored Skyhunter) score 4 instead: the engine only runs if the body survives and connects.
• 4 points — Standard true tutors, CMC 3–4 or restricted (Grim Tutor, Wishclaw Talisman, Fabricate, Eldritch Evolution, Finale of Devastation, Green Sun's Zenith)
• 4 points — Combo-enablers-that-tutor (Demonic Consultation, Tainted Pact, Doomsday)
• 2 points — Narrow or expensive true tutors (CMC 5+)
• Graveyard-destination tutors (Entomb, Buried Alive) score their tier only when the deck has a real recursion package to bring the card back — three or more recursion cards, or a recursion commander like Muldrotha or Meren. Without one they are dead cardboard and score 0.
• +5 bonus — If your commander is a tutor, add +5 to the tutor total
• Premium gate (rows 9–10): the 9 and 10 rows additionally require at least 2 premium-tier tutors (the 6-point tiers; a tutor commander counts as one). Volume alone is not an elite search suite — ten standard tutors make a very good column, and it caps at 8.

Draw Engine Sources are scored on a scaled point system:

• 6 points — Burst draw (Ad Nauseam, Wheel of Fortune, Necropotence)
• 5 points — Premium asymmetric engines (Rhystic Study, Mystic Remora, Esper Sentinel, Sylvan Library)
• 4 points — Standard repeatable engines. Holding the monarch counts — the crown draws you a card every turn.
• Conditional “repeatable” draw is demoted: a draw stapled to a one-shot event or to combat is not an engine. Enter-the-battlefield draw (Prime Speaker Zegana) and sacrifice-itself draw (Commander's Sphere) score 2 (one-shot — it's one card, once). Combat-conditioned draw of ANY kind scores 3: the card's own attack (Cold-Eyed Selkie), your creatures connecting (Toski), or an opponent's combat choices (Breena, Mangara — political trickles an opponent can simply decline to feed). Engines that run off your deck's normal game actions keep the full 4 — Tatyova's landfall and Beast Whisperer's cast trigger fire every turn in their decks — and premium opponent-inevitable engines (Rhystic Study, Consecrated Sphinx: casting and drawing happen every turn) keep their tier.
• 3 points — Card selection / filtering (Brainstorm, Ponder, Preordain, Sensei's Divining Top, loot and connive effects). Selection points count toward the draw total only up to 30 (ten pieces) — filtering re-sorts the cards you were already drawing; the twentieth loot effect does not find you more of them.
• 2 points — One-shot draw (Night's Whisper, Read the Bones)
• 2 points — Symmetric engines (Howling Mine, Temple Bell, Dictate of Kruphix): three opponents drink first, so a symmetric engine is worth a one-shot, not an engine. Wheels stay burst — the caster builds to exploit the refill.
• Cards that trigger on your opponents' draws (Sheoldred, the Apocalypse; Orcish Bowmasters; Underworld Dreams) are punishers, not draw sources — they score as Interaction, never here.
• +3 bonus — If your commander is a draw engine, add +3 to the draw total

Commander Combo Bonus: If your commander is a combo outlet (it participates in at least one combo your deck can actually assemble), add +4 to both the tutor total and the draw total.

Command-Zone Engine: a commander that IS the engine — a repeatable battlefield-tutor (Sisay, Weatherlight Captain) or a repeatable draw/selection engine (Kinnan's activated top-five dig) — lifts its matching column by up to two rows, to at most a 9. The other column still binds.

Tribal/Synergy Bonus (graduated): redundant copies of one effect are virtual tutors for it. 10+ cards filling a single critical role add +10 to the tutor total; 6–9 cards add +5; each additional 10+ role whose carriers are mostly new cards adds +5, up to two extras. The cards must be interchangeable — a pile of different win conditions is variety, not redundancy — and copies count (25x Persistent Petitioners is one role filled 25 times). Interaction suites and tutor/draw packages never qualify. Redundancy lifts the tutor column at most to a 7; the rows above require real tutors. The cap is about tutorless piles, so it is waived for decks that already run a real premium search suite — two or more premium-tier tutors (a tutor commander counts as one): redundancy stacks on top of genuine search, it never substitutes for it.

Mana Reliability modifier: a deck that can't cast its spells on curve loses consistency no matter how good its draw suite is. Count effective sources (lands + 0.75 per rock or dork): 6+ below your curve-adjusted land target = −1; 10+ below = −2. A color carrying 20%+ of your pips on fewer than 10 producers = −1. Capped at −2, calibrated so only genuinely degenerate mana bases are touched. Tapped lands are a Speed cost, not a Consistency one.

Resilience — Full Rubric

What it measures: how well the deck forces a win through disruption — bouncing back from broken-up combos, rebuilding after wipes, and keeping a path to victory open. This is where glass cannons get caught: maximum Speed with minimal Resilience produces the moderate CRISPI those decks earn at real tables.

Combo Layering counts distinct win conditions — only combos that actually produce a win or an infinite engine. Lines sharing a single point of failure count as 1.5, not 2. “Efficient” has teeth: a line costing more than 6 total mana to assemble and fire, needing an X cost to matter, or spanning 4+ cards counts as half a line. And lines are assembly-adjusted: a combo the deck cannot reliably FIND is not a full win layer. Line credit scales with tutor access — full credit at 24+ tutor points (or a battlefield-tutor commander, which is guaranteed access by definition), ×0.75 at 12–23, ×0.5 below 12. Assembly-discounted totals read below the rows above: 0.75 lines read a 5, 0.5 lines a 3.5. A tutorless deck that happens to ship a two-card combo draws into it by luck — that is a bonus, not resilience.

What counts as a combo line: only combos that actually produce a win or an infinite engine — a registered two-card value interaction is a synergy, not a win condition. And “efficient” has teeth: a line whose pieces cost more than 6 mana total to assemble and fire, that needs an X cost to matter, or that spans four or more cards counts as HALF a line. Clunky backups are worth something, but they are not the reliable layers this column measures. And the row-7 entry means what it says: 1 primary + 1 backup. A single line with no other real win path — no combat plan, no stax or control inevitability reading at least a 5 — is a one-trick deck and caps at 6. A precon that happens to ship one two-card combo is not a resilient combo deck.

Reading the combat rows — recursion points: repeatable and format-defining pieces (Sun Titan, Underworld Breach) are 2 each; mass recursion that rebuilds the whole board in one card (Replenish, Living Death) is 2; exile-zone recursion 1.5; repeatable token engines (Bitterblossom) 1.5; repeatable flicker engines (Conjurer's Closet) 1.5; one-shot graveyard spells 1. Heavy draw is recovery: 32+ draw points add +1 recursion point, 40+ add +2 (draw never counts as an engine). The rows demand STRUCTURE, not just totals: row 9 means 12+ points including four rebuild ENGINES; row 8 means 10+ with three; row 6 means 8+ with two; rows 5 and 4.5 ask for 4+ and 2+ points. Engines are the repeatable/mass/token/flicker pieces above, plus one per 3 sticky self-recurring creatures (persist, undying, Bloodghast class). Protection includes attack deterrents — Ghostly Prison protects the player — plus one piece per 3 self-protecting threats and per 3 deathtouch bodies; rows 8–9 count real protection cards only (the per-3 bonuses support the rows below them).

The threat base: a threat is anything the table must answer. Creatures and crewed Vehicles count on power 4+, evasion, or deathtouch; planeswalkers, manlands, repeatable token engines, theft and clone permanents, and drain engines (Blood Artist class) always count; so does your commander whenever it is a body or planeswalker. Go-wide boards count in aggregate: with 12+ small creature copies, every lord, anthem, and token doubler is a threat — the multiplier is what turns the swarm lethal. Threat quality: a vanilla beatstick — big power with no evasion, no self-protection, and no value engine attached — weighs HALF toward the requirements. Colossal Dreadmaw is not Sheoldred.

Protection quality — board-level effects are their own requirement: “rebuilds through wipes” requires wipe answers, and the combat rows now ask for them directly — 1 board-level protection effect at row 6, 2 at row 8, 3 at row 9. Board-level means a player grant (Teferi's Protection), a team grant (Heroic Intervention, Boros Charm), a fog, phasing, or mass blink (Ghostway class). One board-level effect is a card, two is a plan, three is deep protection. Boots answer a Doom Blade, not a Wrath. Protected combo turns: a stack-protection suite of 5+ pieces (counters, free interaction, Silence effects) adds +0.5 to the combo channel.

The command zone is counted: your commander classifies through the same lenses as the other 99 — a recursion commander (Muldrotha, Meren) is a guaranteed rebuild engine, an interactive commander is interaction. The Commander Dependency penalty below nets against this credit. A commander whose activated ability puts pieces directly onto the battlefield (Sisay class) adds +0.5 combo lines when the deck registers a real line.

The stax/prison path: 4 stax pieces read as a 4, 6 pieces as a 5, 8+ as a 6 — capped at 6, because density alone doesn't prove a lock; claims above the middle of the scale need combo lines or the combat rows. Symmetric locks (Winter Orb class) count only when your own deck dodges them (dense rock acceleration or an ultra-low curve); otherwise they're interaction, not inevitability.

The Voltron path: equipment survives creature wipes — the commander recasts and re-equips. Six or more equipment with 3+ protection reads as a 6; ten or more with 5+ reads as a 7. Caps at 7.

The answer-density path: draw-go control is inevitability of its own kind — 13+ interaction pieces with 5+ counterspells and 32+ draw points reads as a 6; 16+/8+/40+ reads as a 7. Caps at 7.

Engine Exposure modifier (penalty-only, capped at −2): a deck that concentrates its nonland engine behind one hoser class folds to a single resolved hate piece — graveyard dependence to Rest in Peace, artifact piles to Vandalblast, enchantment piles to mass enchantment removal. 45%+ of nonland cards in one class with fewer than two ways to answer an artifact or enchantment = −2; answerable, or 30%+ and unanswerable = −1. Worst class only; creature concentration is exempt — and cards that fight as creature threats (creatures, crewed Vehicles, your commander) never count toward the artifact or enchantment share. A deck whose “artifact pile” is its Eldrazi beaters is a creature deck, not a Vandalblast casualty: the penalty measures engine dependence, not what your threats are made of.

Commander Dependency Penalty: after determining your base score, judge how the deck plays when the commander is answered. “Moderate” is the format default — most Commander decks are built around their commander; that is the nature of the format. The other tiers catch the tails.

• None (no penalty): the deck genuinely runs fine commander-less — 80%+ capacity with the commander never cast (a goodstuff pile, a combo deck whose lines live entirely in the 99).
• Moderate (−1): the format default — the commander matters, losing it hurts, the deck still executes its plan (50–80% capacity).
• High (−2): the deck is brought to its knees — a single well-timed counter or removal spell on the commander sets the whole deck back 2+ turns, or the game plan simply stops functioning until it returns (< 50% capacity: the engine, win condition, or mana base lives in the command zone).

Combat-based resilience (redundant threats + protection + recursion) is a supported path to scores 7–9. Decks do not need infinite combos to score highly on Resilience.

Interaction — Full Rubric

What it measures: how well the deck disrupts opponents and protects its own plan — count, efficiency, timing, and coverage across threat types. Toward the cEDH ceiling, interaction shifts to the stack: counterspells, free interaction, and instant-speed answers are what carry a suite to 8–10 — and free spells include free protection (Flawless Maneuver, Deflecting Swat) and free removal (Deadly Rollick, Snuff Out).

How the columns combine: the two ladders above (totals between anchors interpolate linearly) plus answer-scope coverage are joint requirements — all must qualify, and the weakest binds. The raw count is deliberately generous at the top: a compact 26-piece cEDH suite reads 10 on count, and so does a 26-piece precon pile — the stack/timing column is what separates them, and it is load-bearing at every score, not just the top rows. WHEN you can answer is the real difference between deck classes: a suite with zero counterspells, zero free spells and a handful of instants reads ~5 however dense it is; a suite that genuinely operates at instant speed reads 9+.

Answer-scope coverage (also gates rows 8+): count which of creatures, artifacts, and enchantments your suite can actually touch. All three — or a real counterspell suite of 4+ counters — permits every row; creatures plus one other class permits up to 8; creature-only permits up to 7. A twenty-piece all-burn suite can never answer a resolved Rhystic Study; an uncovered class is your suite's real hole.

Total Interaction is an unweighted count of all interactive cards: removal (including repeatable edict engines like Grave Pact), counterspells, stax and hoser punishers (Orcish Bowmasters, Hullbreacher, Aven Mindcensor), protection (including mass blink and totem armor), board wipes, attack deterrents, goad, graveyard hate, theft, targeted hand attack, and spot land interaction. The command zone counts. Quality and timing are priced separately, by the Stack/Timing Points below:

Stack/Timing Points:

• 2 points — every counterspell (effective counterspells included: Veil of Summer, Deflecting Swat, Pyroblast class).
• 2 points — every free interaction spell (0-mana / alternate-cost reactive spells: Force of Will, Fierce Guardianship, Deflecting Swat).
• 2 points — every turn-protection effect (Silence, Grand Abolisher class — protecting your own turn).
• 1 point — every interaction piece usable at INSTANT speed (instant type or flash). This stacks with the classes above: a counterspell is an instant, so it prices at 3 total; a free counter at 5. Premium instant-speed removal (Swords to Plowshares, Path to Exile, Assassin's Trophy) earns its stack presence here. Sorcery-speed removal — however hard its scope — earns 0: it can never answer a combo turn. Bounce is tempo, not removal.
• 1 point — every premium HARD-scope board wipe: exile all, each-player-sacrifices, or −X/−X toughness at CMC ≤ 4 (Toxic Deluge, Culling Ritual class — they answer the boards instants can't). Destroy wipes (Wrath of God class) earn 0 stack points; they count in Total Interaction only. Wrath of God is not Toxic Deluge.

What counts as a counterspell: anything you cast on the stack to stop a spell or fight over your own — including “effective counterspells” like Veil of Summer, Autumn's Veil, Imp's Mischief, Bolt Bend, Deflecting Swat, and Pyroblast.

Post-Calculation Modifier:

• Board Wipe Cap: If the deck relies on 3 or more symmetric board wipes, the final Interaction score cannot exceed 7. One-sided wipes never feed this cap — that includes effectively one-sided sweepers (an overloaded Cyclonic Rift, Toxic Deluge for 2) and edict-style effects that only ever touch your opponents' boards ("each opponent sacrifices…" — Vona's Hunger, Promise of Loyalty). Symmetry is judged in context: a wipe that is symmetric by rules is effectively one-sided when your own board dodges it — a creature-only Wrath in a deck running 8 or fewer creatures, or a conditional wipe whose condition spares your theme (Winds of Rath in an enchantment-heavy deck).
• Where did Density Reconciliation go? Retired — the stack/timing column now prices instant-speed density directly, at every score, exactly once. One mechanism instead of a column plus a penalty charging the same deficiency twice.

Speed — Full Rubric

What it measures: how quickly the deck can realistically take a player out of the game, measured by its fundamental turn — the turn it actually COMPLETES an elimination or wins outright (not merely threatens one) in at least 50% of games with normal draws and no disruption. Goldfish a dozen hands turn by turn (lands, ramp, commander, the win line assembled) and take the turn a player is actually out. The fundamental turn is the entire Speed rating — there is no separate AMV or fast-mana score. Tie-break conservatively: torn between two turns, take the slower one. For stax, a full lock counts as the win turn — a lone tax piece is disruption, not a lock.

One turn is worth a full point where the game is fastest — the difference between winning on turn 3 and turn 4 is enormous, while turn 10 versus turn 12 barely matters — so the fast rows are single turns and the tail rows blend. When a deck's fundamental turn genuinely straddles two adjacent turns (eliminates a player on turn 4 on curve, turn 5 through an average brick), Speed lands on the half-step between them: a 7.5 means “eliminates a player on turn 4 or 5, depending on the draw.” A half-step is a real read of variance, never a hedge — if the 50% turn is one turn, the score is that row; if you are torn, take the slower turn.

Elimination scope: the fundamental turn is when the deck actually takes its FIRST player out of the game (or wins outright), completed rather than merely set up. Simultaneous wins — combo, X-drain, overrun into a wide board, mass poison — take the whole table at once: count the turn they fire. Sequential decks (Voltron 21-commander-damage) count the turn the first opponent actually dies: one-shotting a player on turn 6 is a turn-6 clock, even though cleaning up the remaining opponents takes several more turns. The math still has to add up: a knockout is 40 damage, 21 commander damage from the same commander, or 10 poison — a 12-damage Voltron swing is only half a knockout against a single opponent. And the 50% bar applies to a line's pieces, not its mana: a line that hinges on one specific 2–4 copy effect with no tutor that finds that piece is a high roll, not the fundamental turn — the score times the median game.

Alternative clocks count: poison is lethal at 10, not 40 — a swarm plus Triumph of the Hordes kills the table the turn it resolves. Each-opponent mill is a real clock. A deck that cheats its threats into play (Sneak Attack, Kaalia, reanimation) is faster than its curve, full stop.

Speed/Consistency Coupling: If your calculated Speed score is 9 or 10, check your Consistency score. If Consistency is 7 or lower, Speed is capped at a maximum of 8.