Matchup Lab

Facts, not verdicts

The page shows what can be counted, measured or read from the rules, and leaves the judgement to you. It says a deck has eight threats the other can't cleanly answer, a lower curve and a kill two turns sooner; it never says which deck is stronger, who is the beatdown or what to bring in. Where a well known framework turns numbers like these into a call (Mike Flores's "Who's the Beatdown?", Reid Duke's "Role Assignment"), the page names it and leaves the call to the reader.

Three kinds of number

Every number on the page carries one of three labels, and none of them is ever blended with another into a win chance, because no honest method can do that from two lists.

• Simulated games: from games each deck plays alone, 2,500 per deck on the play and 2,500 on the draw (10,000 in all). Each game has its own seeded shuffle, so the same two lists always give the same numbers, and a share is within about 2 points either way (95%).
• Read from card text: from the card facts, card by card, with the Interaction Oracle's judge (its page has the details): whether each answer deals with each threat, and what stops it.
• Recorded matches: real results, with their 95% Wilson range: the two archetypes' matches in Meta Lens's data, and your own Game Diary.

standard error of a share = sqrt(p x (1 - p) / games)

At a glance lists its findings in a fixed order (the clocks, then interaction, answers, the board, the graveyard, combos and the long-game counts, then what was recorded), each with its kind and nothing else.

The games

Each deck plays alone against an empty board, the way King Goldfish does: Frank Karsten's keep rules and the London mulligan (the extra lands, fewest colors first, or the most expensive spells go to the bottom), one land a turn chosen to spend the most mana, then the cards that press forward, most mana first. Mana comes from lands (with fastlands, slowlands, checklands and taplands entering as they would), mana creatures and rocks from the turn after they arrive, and Urza's Tower, Mine and Power Plant making seven between them once all three are out. Spells that cost less for each instant and sorcery in your graveyard, delve, affinity, improvise, convoke, and costs that count your lands, creatures or basic land types get cheaper as the game goes; other reducers aren't followed, and the page says so when a key threat has one. A spell with a land on its back (Fell the Profane) is played as the land here, and still counts as castable for the timing. Phyrexian mana is paid with life, the other half of a split card or adventure (Ice, Stomp) counts as a way to cast it, and a card you can cast for free by pitching another card or paying life (Force of Will, Solitude, Snuff Out, Fireblast) counts as castable once there's another card in hand to pitch.

Damage: creatures attack the turn after they arrive (at once with haste), double strike counts twice, and prowess and its cousins add their bonus for each noncreature spell cast that turn. Burn and drain that can hit a player go at the face; tokens count at their printed size, including ones made every upkeep. Removal, counterspells and discard wait for a target, and mana left over at the end of a turn is spent on them anyway, the way a real game spends them on the other deck's cards: they never touch the clock, but prowess and graveyard costs see them. The kill turn is the first turn the damage reaches the format's starting life (from the format pages' registry: 20 in most formats, 25 in Brawl, Standard Brawl and Competitive Brawl, 30 in Pauper Commander, 40 in Commander, cEDH and PreDH), or infect and toxic reach ten poison counters. A deck that deals poison says that the pump spells an infect deck leans on aren't counted, in the Speed notes and in the clock line at a glance.

A deck that also wins another way says so, and says its clock here covers only attacks and burn: a complete combo (from Commander Spellbook); an alternate win card (a card whose own text wins the game, like Thassa's Oracle or Approach of the Second Sun); cards that put creature, artifact or permanent cards onto the battlefield from a graveyard, hand or library, or take extra turns, when there are two or more big threats to cheat in (a return capped at mana value four or less, like Emptiness or Zoraline, brings back value, not a big threat); six or more cascade cards; a storm card with six or more fast mana cards; eight or more dredge cards; four or more huge threats with ten or more mana cards to ramp into them; eight or more lock pieces that aren't creatures or planeswalkers; a land combo (Scapeshift, or Dark Depths with Thespian's Stage or Vampire Hexmage); or eight or more auras and equipment the games don't count. A deck with none of those, fewer than 14 answer copies (spells with a land on the back left out), lands in no more than 60% of its main deck, and no kill by turn 12 in at least half its games on the play gets one sentence of what was measured: the share of its games with no kill by turn 12, and its threat copies. The page doesn't say how, or whether, it wins. Planeswalkers are a long game, not a faster clock, so they stay in the long-game counts.

The same games record the first turn each card is in hand with the mana to cast it (for a land, the first turn it is in hand), which is where every timing number comes from.

Who answers what

A threat is a creature, planeswalker, battle, vehicle or equipment, or another permanent that wins, pressures or grinds: an engine that draws or makes tokens again and again, a token maker, an anthem, a drain, a lock piece costing two or more. Card draw once (a Bauble), a mana rock or a combo tag alone doesn't make a threat, and removal that stays on the battlefield counts as an answer first. For each answer in one deck against each threat in the other, the judge gives a result for every way the answer works:

• Clean: it works for sure and for the rest of the game. On the battlefield (exile, destroy, damage enough for its toughness, shrink, sacrifice, into the library, steal, shut off), as it's cast (a counterspell that can counter it, or an answer that only reaches a spell, like Subtlety) or from the hand (discard where you choose the card). Color limits are read ("target blue permanent", "if it's red"), and so are a counterspell's own limits.
• Only sometimes: a condition decides (ward, a mana value limit, X damage, their choice of what to sacrifice or discard, a counter that depends on what's on it like Chalice of the Void).
• For a turn only: bounce, tapping and phasing buy time. They are listed apart and never counted as answers.

coverage = threat copies with at least one clean answer / all threat copies

trade = mana value of the threat - mana value of the cheapest clean answer (above zero, the answer is cheaper)

A sweeper is a card with the board wipe role whose text hits many creatures at once: destroy or exile all or each, damage to each creature, creatures that all get minus, each player sacrificing several, or every creature returned or put away together. The role alone isn't enough (Ajani, Nacatl Pariah carries it for his back face's ability).

Timing and the Fundamental Turn

Counted answers are the ones that handle at least a quarter of the other deck's threat copies, or counter or take at least half of its spells (weighted by cost). The timing chart is the share of games where one of them is in hand with the mana for it by each turn. Zvi Mowshowitz's Fundamental Turn is the turn a deck wins, and the pairing below is the number that framework works from: your first counted answer and their kill turn come from separate games, paired. You on the play act on your turn k before their turn k, and on the draw your turn k - 1 is your last before it.

P(in time) = sum over k of P(their kill turn = k) x P(your first answer by turn k, or k - 1 on the draw)

A key threat is the one the page times: the threat with the most copies, weighted up for a mana value of three or more, planeswalkers, engines and top end and down for cards that cycle, can be cast for free or are also answers, among those the other deck has a clean answer to. Its turn is the one it is castable by in half of its deck's games that draw it. When its deck can also put it into play without casting it (Goryo's Vengeance with Atraxa), the line is timed to the cheat card castable soonest from the hand instead, and names that card; extra turns aren't cheat cards, and a card cast by flashback (Dread Return) is timed from the hand only, since the games don't track flashback. When the key threat has ward, the line says what the ward adds: the timing counts each answer at its own cost, without the ward.

Speed, curve and the long game

This section shows the numbers role frameworks work from, and assigns no role. The median empty-board kill turn for each deck on the seats in view; the average mana value of each deck's nonland cards and its curve, copies by mana value; and each deck's lands against Frank Karsten's land regression for its own curve (a spell with a land on its back counts as a land, and cards that cost less as the game goes are left out of the average; the regression is for 60-card decks, so it isn't given for others).

lands = 19.59 + 1.90 x average mana value - 0.28 x cheap draw or ramp spells

A spell with a land on its back counts with the lands and not in the curve or the average, so the curve and the lands add up to the deck. The regression's cheap spells are noncreature card draw and ramp costing 2 or less, as in Karsten's model; a creature that draws or makes mana isn't counted.

The long-game counts: card and token engines (a draw or a token made again and again, not once), planeswalkers, cards that come back, top end (creatures and planeswalkers costing five or more), draw-two spells, threats with no clean answer in the other deck, counted answers, and complete combos from Commander Spellbook. At a glance names the lines where the two counts differ, with both counts, and says the rest are level; it keeps no score across lines, since they don't weigh the same. In the table, bold marks the larger count.

Each complete combo is timed in its own deck's games: the share of games in which every piece has been in hand with the mana for it by turns 4, 6 and 8. Each piece is castable by then, not necessarily on the same turn, and the mana to finish the combo isn't counted. When Commander Spellbook says a piece starts in the graveyard or library (Narcomoeba, Dread Return by flashback, a dredger), an in-hand share doesn't describe the combo, so the page says so and gives none. Beside it are the other deck's clean answers to each piece, counterspells and discard included, as copies and names. What can reach a piece depends on where Commander Spellbook starts it, and the line names those zones whenever a piece starts outside the hand: in hand, counterspells and discard before it resolves and removal once it's out; on the battlefield, any clean answer; in the graveyard, only graveyard hate; in the library or exile, nothing, and the line says so. For a piece whose part is its enters or cast trigger (Thassa's Oracle, cast or reanimated), only what acts before it enters counts; removal that reaches it after it enters is listed apart, since its trigger still happens.

P(combo by turn T) = share of games where, for every piece, its first castable turn is T or earlier

Mike Flores's "Who's the Beatdown?" and Reid Duke's "Role Assignment" turn numbers like these (who kills sooner, whose curve is lower, who has more that keeps going late) into a role for each deck. The page names the frameworks; which role each deck plays is the reader's call.

Mulligans

The chance that the seven cards you see hold at least one of k cards in a deck of n is exact (the hypergeometric distribution). With the London mulligan you see seven again each time, so the chance of seeing one over several looks grows as below, while the hand you keep gets smaller.

P(at least one in seven) = 1 - C(n - k, 7) / C(n, 7); over m looks: 1 - (1 - p)^m

Sideboarding

Four tables lay out the facts behind every nonland card: your sideboard and your main deck, each against their main deck, and their sideboard and their main deck, each against yours. Every row is read against the other deck's game 1 main deck, and says what applies:

• An answer: the threat copies it answers cleanly (with their names), the ones it answers only sometimes, the ones it only buys a turn against, and for a counterspell or discard the share of the other deck's nonland cards it can counter or take. That share is weighted by cost: a card costing 1 or less counts half, and so does one it only sometimes stops. It is also judged against every land and every permanent of theirs that isn't a threat (mana rocks, a Bauble, Void Mirror), and says which it hits. Only an answer that hits nothing at all, threat or not, says it has no targets, and a threat never leads with an empty removal side. With Painter's Servant in either deck, color hate (Pyroblast, Red Elemental Blast) reaches any card while Painter's Servant names its color, and the row says so. Ward the answer would pay is named, and burn that can hit a player says so.
• A card that stops other cards: the copies it stops on each side. Graveyard hate stops the cards that use a graveyard (both sides when it hits every graveyard); a lock or tax is read with the Oracle's pair engine, and a tax or a one-spell limit is listed apart as taxing or limiting (a one-spell limit like High Noon or Archon of Emeria touches every spell of the players it names); a card that names a card is read only for permanents that stay, not for a spell like Cabal Therapy; "Activated abilities of artifacts (creatures, planeswalkers) can't be activated" (Stony Silence, Null Rod, Collector Ouphe, Clarion Conqueror) stops every card of those types with an ability a permanent on the battlefield activates, mana abilities included (not cycling or other abilities used from the hand or graveyard); a card that names one card (Pithing Needle, Phyrexian Revoker, Sorcerous Spyglass, Meddling Mage) stops the other deck's card with the most copies among those it can stop, ties going to the higher mana value, lands included when it may name any card, and none of your own; one that only makes the named card cost more (Anointed Peacekeeper) taxes it instead; and Containment Priest stops what puts a creature onto the battlefield without casting it: reanimation, cheats, flickers, and creatures that return themselves (Bloodghast, Prized Amalgam, Narcomoeba, Ichorid) or come back with persist, undying or unearth.
• A threat: how many of the other deck's answer copies handle it cleanly, and how many of its sweeper copies kill it.
• A card that uses the graveyard: the graveyard hate the other deck runs, main deck and sideboard. Any other work it does (makes tokens, draws cards, makes mana, protects your side and so on), a companion, and for their sideboard how many of the archetype's recorded sideboards carry it, with its average count there. A card the read has nothing on says so.

Rows run in a stated, neutral order: answers first (by the larger of the share of threat copies they handle cleanly and the share of cards they counter or take), then cards that stop other cards (by the share they stop), then threats, then the rest, most copies first. The page makes no changes of its own. Each sideboard row has an in stepper and each main deck row an out stepper, from none up to the copies there are, for your side and theirs. Changes are kept per seat, as real sideboarding differs on the play and the draw, and start over when either deck changes. A tally counts what goes in and out and the main deck size after, and says plainly when in and out don't match. Games 2 and 3 play both decks after the changes (none by default), and the header says what changed. The land note gives each deck's lands before and after (cutting a spell with a land on its back lowers the count) and Frank Karsten's regression for the curve before and after. With a saved deck from the same format, the changes save to its sideboard plans, as long as the deck still holds every card they move. Saving over the plan for the same matchup keeps the note written on it and offers Undo, which puts the old plan back.

A Meta Lens archetype

A Meta Lens archetype's typical list comes only from its lists legal today (Meta Lens's legality check: a card banned since, a restricted card over its limit, a commander banned as commander), from the last 30 days, else 60, 90 or all of them when fewer than 8 legal lists are that recent. When every list holds a card banned since, there is no typical list, and the page says which card; a single list can still be picked as it was played, labeled not legal today. The sideboard shares under Sideboarding facts come from the same lists. One line under the two decks says which lists they are, how many legal lists they come from, the days those cover and how old the newest list is.

What was recorded

Meta Lens keeps the rounds of MTGO events where both players' decks are known. For two archetypes, every match between them counts once, as won, lost or drawn by games; mirrors are left out. A deck that isn't a Meta Lens archetype stands in for the archetype it reads closest to, by cosine similarity of its nonland, nonbasic cards against each archetype's average list over the last 90 days, when the similarity is at least 55%. Your Game Diary counts the matches logged with that deck against an opponent written the same way. Both show the 95% Wilson range, and under 20 decided matches they say it's a small sample.

How it was checked

Misses first. The games can't see decks that win another way: a planeswalker or combo deck (Legacy Tron, for one) shows no kill by turn 12, and the page says its clock covers only attacks and burn. Pump spells, auras, equipment and triggered damage beyond what's listed aren't counted, and neither is reanimation or cheating a creature in, so those decks read slower than they are; interaction is never cast in anger, so no game shows a threat dying.

Two simple decks check the arithmetic. Twenty-four Forests and thirty-six Grizzly Bears kill on turn 6 in 98% of games on the play: casting a bear every turn from turn 2 makes 2 + 4 + 6 + 8 = 20 damage by turn 6, with the land count to spare. Twenty Mountains and forty Lightning Bolts on the draw kill by turn 4 in about 55% of games, which needs three lands on curve (3 + 6 + 9 + 9 = 27). Eight real matchups across Standard, Pioneer, Modern, Pauper and Legacy were read line by line against what their players know about them; the problems that turned up (bounce counted as a real answer, prowess and cost reducers left out, archetypes ranked by the wrong window) are fixed, and the misses above are what's left.

Then an independent attack, a tester trying to break the page, found 35 problems, 8 of them serious: a view or seat change could leave the last pair of decks on screen under the new names; sideboard answers were never judged as answers; mana rocks and Baubles counted as threats; and combo and planeswalker decks looked slow with no warning. The judge itself (shared with the Interaction Oracle) got sharper from it: a counterspell's own limit is read (Spell Snare, Change the Equation, Stern Scolding), a fixed minus like Dismember's is weighed against toughness and a minus paid for in life reaches any, protection from a player (True-Name Nemesis) stops that player's damage, a cast trigger's mana value is no longer read as a limit (Kozilek's Return), a mode chosen three times counts three times (Fiery Confluence), and Engineered Explosives reads as conditional. All 35 were fixed or answered. Four more attacks, reading well over a hundred real matchups in six formats plus pasted lists, saved decks and the games 2 and 3 view, found more: color-limited answers (Pyroblast, Hydroblast, Celestial Purge) read as hitting everything, converge (Prismatic Ending) and flickers (Phelia, Flickerwisp) read as clean removal, Reanimate read as taking a creature, Living End not read as a graveyard deck, combo detection that missed some plans and flagged Death and Taxes, affinity never lowering a cost in the games, false "wins another way" flags (Lifegain, Initiative, Pauper Elves), conditional removal (Amalia, Force of Despair) read as clean, and the judge's color words for "monocolored" (Ultimate Price), ability counters (Stifle), guard riders (Reality Smasher), combat-only removal (Stinkweed Imp) and one mode's mana value limit leaking into another (Brotherhood's End). Those are fixed, and poison counts, spells with a land on the back answer as spells, and list problems (too many copies, a big sideboard, a companion that doesn't fit) are flagged.

Since September 24, 2026, the page makes no calls and shows the facts behind them. The tone tags (for you, against you), the role call (who is the beatdown), the long-game verdict, the mulligan advice and the suggested sideboard swaps are gone. In their place are the clocks, curves and long-game counts side by side, combo odds with the other deck's answers to each piece, and a facts table for every card with steppers to set your own changes. Checking that version found and fixed these: Pithing Needle, Phyrexian Revoker and Sorcerous Spyglass registered nothing (a card that names a card is now read directly), Containment Priest stopped nothing, and locks like Stony Silence and Clarion Conqueror stopped nothing against Affinity (all now read directly); Ajani, Nacatl Pariah counted as a sweeper; Gyruda, Obosh and Keruga were never checked as companions, because their real wording differs from what was matched (all eight companions the page checks are now tested both ways); a threat its deck cheats in was timed to its own cost rather than to its cheat cards; and Legacy Lands, Oops! All Spells and Selesnya Depths weren't flagged as winning a way the games don't show (the flag is wider, and alternate win cards and the Dark Depths land combo are now named).

A second check on the facts found facts that were false or misleading, now fixed: an answer whose only targets were lands or mana rocks said it had no targets (Force of Vigor against the Moxen, Raze against lands), and a threat's row opened with its empty removal side; color hate ignored Painter's Servant; ward never appeared; a combo piece that works from the graveyard was still timed from the hand, and Lightning Bolt counted as an answer to Thassa's Oracle, whose trigger still happens; extra-turn cards counted as cheat cards (Time Walk for Oath's Atraxa); Containment Priest missed Ichorid, and Null Rod counted cycling; Anointed Peacekeeper and Cabal Therapy read wrong; a deck with no kill in these games was said to win some other way even when it has none; the long-game line kept a score across lines that don't weigh the same; and the curve counted spell-lands twice. A sweep of real pairs in six formats, both seats, game 1 and games 2 and 3, finds no verdict words on the page. What's left is listed under limits, and every change is checked again.

Sources

• Mike Flores, "Who's the Beatdown?" (StarCityGames); Reid Duke, "Role Assignment" (January 5, 2015) and "Inevitability" (December 8, 2014), Level One on magic.wizards.com: the frameworks that turn speed, curve and the long game into roles.
• Zvi Mowshowitz's Fundamental Turn, as laid out in "Examining the Fundamental Turn" (CoolStuffInc): the turn a deck wins, and interacting before it.
• Frank Karsten, "How Many Lands Do You Need in Your Deck? An Updated Analysis" (2022): the land regression; and "How Many Sources Do You Need to Consistently Cast Your Spells?": the mana model the games share with Manabase+.
• Magic: The Gathering Comprehensive Rules, rule 103.5 (the London mulligan), 605.1a (what a mana ability is) and 702.108 (prowess).
• Edwin B. Wilson, "Probable Inference, the Law of Succession, and Statistical Inference" (1927): the 95% range.
• Meta Lens decklists and rounds (see its page for sources and dates), card facts and rulings as on the Interaction Oracle's page, and combos from Commander Spellbook.