Deck analysis

Deck analysis

The header figures for the maindeck. The sideboard is parsed so its cards resolve, but it is never analysed: sideboard cards do not affect a land count or a source count.

What each figure is

• Cards — maindeck cards, counting copies. A legal Constructed deck reads 60 here; anything else means the export was trimmed or the format is not the one this tool models.
• Lands — cards whose front face type line contains "Land". A modal double-faced card with a land back counts as a spell here, which is what Karsten's regression expects: it has its own term for those.
• Spells — everything else.
• Average mana value — the total mana value of the spells divided by the number of spells. Lands are excluded, which matters: including them would drag the average toward zero and cost you two or three lands in the recommendation.

Using it

The share link is the whole analysis: anyone who opens it recomputes the same numbers from the same decklist. Save as PDF prints through the browser, and every collapsed section is opened first so nothing is missing from the file.

Maindeck and sideboard

Which cards are the deck and which are the sideboard. Everything else on this page is computed from the maindeck alone, so a card on the wrong side is wrong everywhere: it changes the deck size every probability divides by, the average mana value, and every source count.

How the split is found

• Declared — the export carries a Deck or Sideboard header and says so itself. Nothing is guessed.
• Guessed — there is no header and the list holds more than 60 cards, so it is a deck and its sideboard run together. Exports written this way separate the two with a blank line, and that blank line is taken as the split.
• Unresolved — more than 60 cards, no header and no blank line. There is nothing to read the split off, so every card is treated as maindeck and the numbers above are those of an oversized deck until you move the sideboard out.

Correcting it

Each arrow moves one copy. A card can sit on both sides — a fourth copy in the sideboard is ordinary — which is why copies move one at a time rather than the whole card at once.

Moving a copy writes the Deck and Sideboard headers into the stored decklist, so the split is settled from then on and everyone opening the link sees the same one. The cards themselves are never changed.

Mana sources

Which colours each land counts for. Every colour figure elsewhere on the page is built from what is settled here, so a wrong answer here is wrong everywhere.

How a land is counted

• A land producing colour C is a source of C.
• A nonland with mana value 2 or less that produces C is also a source of C. It is listed separately, because it dies to removal and a land does not.
• A fetchland is a source of every colour it can actually find among the lands in this deck — fetching a Mountain is worth nothing if the deck runs none.
• A land producing only colourless is a source of no colour. It still counts toward the land total and still pays generic costs.
• A land that always enters tapped still counts, but raises a warning against turn 1 and turn 2 spells, where the delay is the whole problem. A tapped clause with a condition — shocklands, fastlands, checklands — does not count as slow.

Why some lands ask

A land is highlighted when Scryfall's data cannot settle it:

• Scryfall reports no mana for it at all, because its type is chosen as it enters.
• It produces three or more colours, which usually means a restriction the card text carries and the data does not.
• Its text says the mana may be spent only on certain spells — Cavern of Souls lists five colours but pays only for creature spells.

Using it

Tick the colours the land is really worth in this deck, then save. The answer is stored against this deck alone, since what Cavern of Souls is worth depends on the creature types around it. Every other land sits in the collapsed list and can be corrected too. Reset to detected throws away every confirmation and goes back to what Scryfall says.

There is no way to say "this land counts for creature spells only". If a restricted land does not really support the deck, untick its colours.

Card types

What the deck is made of, and how often each part shows up. The curve says when you can act; this says with what.

Composition

Every maindeck card counted once, copies included. A card with several types is counted under its most characteristic one, so an artifact creature is a creature and never both: counting it twice would make the shares add to more than the deck.

Only the front face is read. A modal double-faced card with a land back is counted as the spell it is cast as, which is the same choice the land count makes.

Opening odds

For the groups worth opening on, the hypergeometric chance of holding that many in a seven-card hand, and of having seen at least one by turn 3 on the play. No mulligan is modelled, so a real hand is a little better than the "none" column suggests -- you would throw the worst of them back.

• Creatures — cards typed Creature, artifact creatures included.
• Permanents — nonland cards that stay on the battlefield: creatures, artifacts, enchantments, planeswalkers, battles.
• Instants and sorceries — the rest of the spells.
• Lands — the same count the land section works from. The "none" column here is the chance of a hand you would have to mulligan.

Reading it

The figure to watch is the none column for whatever the deck needs to function. An aggressive deck that opens without a creature roughly a quarter of the time is a deck that loses those games to its own composition, not to variance it can play around. A control deck can afford it; a creature deck cannot.

Karsten's articles do not prescribe a split between types the way they prescribe lands and sources. This section is the same maths applied to the question, not a rule from him: treat the percentages as facts about the deck and the thresholds as your own.

Mana curve

Whether the deck can actually use its early turns. A curve that looks fine as a histogram can still fail in play if the land count cannot support it.

The histogram

One bar per mana value, counting copies, spells only. Mana value 7 and above share the last bar, since the distinction stops mattering there.

Curving out

The share of games that play a one-drop on turn 1, a two-drop on turn 2 and a three-drop on turn 3. There is no closed form for this — it depends on draws and on land drops at once — so it is estimated by simulating 10,000 games.

What the simulation does each game:

• Shuffles the deck as lands and spells grouped by mana value.
• Draws seven and keeps the first hand holding between two and five lands, taking at most two mulligans. On a mulligan it bottoms one card per mulligan taken, dropping the most expensive spells first.
• Plays a land every turn if it holds one, then casts a spell of that turn's cost if it has enough lands in play.

The seed is fixed, so the same deck always reports the same number: a change you see is a change you made, not simulation noise. The per-turn table below breaks the same simulation down — the share of games that made each individual drop, which is always at least as high as making all three.

Reading it

A low figure with a healthy curve usually means too few lands, or too many that enter tapped. A deck with no spell at one of those costs cannot curve out at all, and the page says so rather than reporting a bare 0%.

How many copies to play

How often a card shows up, by copy count. The answer to "is three enough?" and to "is the 61st card free?".

The maths

Each cell is the hypergeometric probability of having seen at least one copy once you have seen that many cards, drawing without replacement:

with N the deck size, K the number of copies, and n the cards seen. Cards seen heads each column because it is what the formula takes: seven is the opening hand, and each turn after adds a draw. The turn that reaches a column is the smaller line under its number, and it shifts by one on the draw — the same fifteen cards arrive a turn earlier. On the play in the land section moves it.

Reading it

The jump from one copy to two is much larger than the jump from three to four: copies have diminishing returns, which is why a four-of is not automatically right. Compare a row against how deep into the game you actually need the card — a four-of you want in the opening seven is a different question from a four-of you want by the fifteenth card.

This table is not conditioned on mulligans, unlike the coloured source targets. It is the raw draw probability.

Why 60 cards

Every figure here scales with deck size. Open the hypergeometric calculator, set the deck size to 61 and compare: the extra card lowers the chance of drawing your best cards, and nothing in the deck gets better to compensate.

Decklist

The Arena export exactly as it was pasted. The analysis is recomputed from this text every time the page loads, so the two can never drift apart.

Only the Deck section is analysed. Sideboard cards are resolved but excluded from every calculation, a Companion sets the companion term in the land formula, and the name in the About block is what labels this deck in the recent analyses list.

To change the deck, paste the new list as a fresh analysis. An analysis is a record of one decklist at one moment, so its cards are never edited in place. Moving a copy between the maindeck and the sideboard is the one thing that rewrites this text: it adds the headers the export left out, and which cards are the deck is part of reading a list rather than a change to it.