How We Score

How We Read Your Deck

A decklist (pasted text or a deck-link import from Archidekt, Scryfall, Deckbox, TappedOut or MTGTop8) is parsed into a commander and a 99, then every card name is enriched against live card data — oracle text, mana cost, color identity, type line — so the rest of the pipeline works from real rules text, not just a name.

Every card is checked against curated lists for tutors, ramp, fast mana, extra-turn spells, mass land denial, and Game Changers. Curated lists catch the well-known staples, but new cards and rarer effects would slip through a name-only list, so most categories also run an oracle-text fallback: anything with "search your library" that isn't a land counts as a tutor (fetch lands search your library too, but they're land fixing, not tutoring — lands are excluded from this check on purpose); anything that says "take an extra turn" counts toward extra turns; anything that destroys or saccs all lands counts as mass land denial; any nonland card that adds mana or puts a land onto the battlefield (rocks, dorks, a commander like Gwenna, land searches like Cultivate) counts toward ramp. Game Changers are the one exception— that list is exact and curated only, straight from WotC's official list, with no oracle-text guessing, because a heuristic here previously mislabeled ordinary powerful cards as Game Changers.

WotC Brackets

Brackets 2 to 4 follow Wizards' published rules and nothing else: Game Changers, two-card infinite combos, mass land denial and extra turns. Fast mana and tutors are not part of Wizards' rules (tutors were removed in October 2025), so they never push a deck above Bracket 2. Bracket 5 is the one place we use judgment, because Wizards defines it by intent; the reason shown always lists the signals behind it.

The power score is separate. A deck can be Bracket 3 by the rules and still be very strong. When its power is well above what its bracket usually scores, the results say so, because Wizards asks players to bracket up when a deck plays stronger than its list suggests.

Power Score Formula

The 0–10 score starts as a weighted blend of eight metrics, all calculated from the decklist and live card data:

Each metric is its own 0–10 sub-score before weighting — for example Ramp Density is min(10, ramp% × 32)(a typical deck's ramp is about 17% of its nonland cards), Tutor Count is min(10, tutors × 0.9), and Combo Presence is min(10, combos × 3.2) — scaled so real cEDH counts actually reach the top of the range instead of plateauing around 6.

That weighted total is then adjusted so the 0–10 scale actually spans its full range instead of bunching in the middle. Up to +0.8 is added for oracle-text synergy density and up to +0.4 for interaction density (both explained below), then everything above 6.0 gets stretched — 6 + (raw − 6) × 1.3, so a raw 7.0 becomes 7.4 and a raw 9.0 becomes 9.8 — so genuinely stacked decks can actually reach 9–10 instead of the formula compressing everything into the middle of the scale.

Finally, a soft floor keeps Bracket 4 and 5 decks from reading below 6.8 and 8.6 (before the stretch), and Bracket 1 decks are softly capped at 4.2. Brackets 2 and 3 have no cap: the power score is how a strong Bracket 3 deck shows that it is strong.

Sub-scores for Speed, Consistency, and Saltiness are derived from subsets of the same metrics and displayed as independent meters on the results page.

Deck Profile — Synergy, Interaction & Resilience

Beyond the curated card lists, every card's actual oracle text is scored against 10 synergy axes: blink/ETB, graveyard, sacrifice, tokens, +1/+1 counters, spellslinger, artifacts, lifegain, landfall, and reanimator. Each axis only scores if the deck has both an enabler and a payofffor it — for example the graveyard axis needs both a way to put cards there (mill, discard, surveil) and a way to benefit from that (flashback, escape, delve, threshold) — and the axis score is the smaller of the two counts, so a pile of enablers with nothing to pay off with earns nothing. The axes that clear a minimum threshold (3, or 0 for the always-relevant legendary-synergy axis) show up on your results as the deck's signature synergies, and the sum of all axis scores (capped at 10, scaled ×0.45) becomes the deck's Synergy rating.

Interaction is measured as a density, not a raw count — what share of your nonland cards are removal, counterspells, or board wipes (oracle-text patterns like "destroy target", "counter target spell", or "destroy all"), and what fraction of those are instant-speed. Scoring off density rather than a flat card count means a small, efficient interaction suite in a lean decklist scores fairly against a bigger suite in a bloated one — a deck needs a genuinely dense, mostly-instant-speed interaction package to approach the 10 cap, not just a dozen or so ordinary answers.

Resilience comes from protection effects (hexproof, indestructible, protection from, ward, shroud, phasing, can't be countered), recursion (return a card from your graveyard), and how strong the deck's token-rebuild capacity already scored on the Tokens synergy axis. Speed and Consistency — the same numbers shown as their own meters on the Power Score panel — plug straight into the radar alongside Interaction, Synergy, and Resilience, so the five-axis Deck Profile shown on every analysis is a genuine cross-section of the same underlying math, not a separate guess.

Bracket Path translates all of it into plain language — exactly which cards or thresholds would move your specific deck up or down a bracket (e.g. "cut your 2 Game Changers to drop to Bracket 3", or naming the specific combo pieces to cut), reasoned directly from your deck's actual flagged cards rather than a generic rule of thumb.

Mana Base Simulation

Every deck also gets goldfished — 10,000 simulated games of just drawing and playing lands, mana rocks, and dorks, with no opponent on the other side of the table. Each game shuffles the deck, draws a 7-card opening hand, and mulligans once for free (Commander's free Commander mulligan) if that hand has fewer than 2 or more than 5 lands. From there it plays a land every turn when one is available — preferring an untapped land that fixes a scarce color — and casts any affordable mana rocks or dorks greedily, cheapest first, before checking how much mana was actually available that turn. The simulation is seeded, so the same decklist always produces the same numbers.

Lands that come into play tapped, or only untapped under a condition (checking a type you control, a land count threshold, and similar), are modeled honestly rather than treated as always-untapped — a deck full of tapped duals will show it in the numbers. Mana rocks come online the turn after they're cast; mana dorks the same, accounting for summoning sickness.

From those 10,000 games we report your tipping point — the earliest turn from which the deck reliably (80%+ of games) has enough mana for a spell at its own average mana value — a cast rate for every individual card in the deck (the odds you actually have the mana and the right colors to cast it on its own curve turn), and an overall mana-base letter grade built from the tipping point, colored-source deficits, opening-hand quality, how often early lands enter untapped, and the deck-wide cast rate.

Colored-source requirements are solved individually per color: for every colored pip demand in the deck (a card's color and the turn it wants to be cast), we calculate — using the deck's exact size, not a lookup table — how many sources of that color you'd need for a 90% chance of having enough by that turn once you've hit your land drops (Frank Karsten's assumption), then keep the toughest demand per color as what the deck actually needs. Lands count as sources, and so do mana rocks and dorks when you draw them. Reactive instants (counterspells, removal spells built to be held up rather than slammed the moment they're castable) get a couple of extra turns of grace before they're allowed to set a color's toughest demand, so the mana base isn't built around treating your interaction like it has to resolve on curve. X spells get the same two turns: their mana value counts X as 0, but an X spell is rarely worth casting for 0 (Exsanguinate would drain nobody), so its colors are due two turns later, and its cast rate is checked then, with two mana spent on X. A hybrid symbol is one pip that any source of either of its colors pays, so every Mountain and Plains in a Boros deck pays for Blade Historian's four red-or-white symbols, and Leyline of the Guildpact never asks a five-color deck for four green. A hybrid pair gets its own row only when the deck is short of sources for it. Phyrexian and two-or-color symbols never demand a color, a split card or Room counts as its cheaper half, and an adventure card as the card itself. A modal double-faced card with a land on the back, like Shatterskull Smashing, counts as a land when an opening hand is judged and is played as that land when there's no other land in hand; a card that only transforms into a land, like Treasure Map, is a spell. In a reanimator or cheat deck (four or more cards like Reanimate, Animate Dead or Sneak Attack), creatures of mana value 6 or more are treated as put into play rather than cast, so they don't set a color's demand or get a cast rate; the panel names them. Your recommended land count comes from a regression formula based on the deck's average mana value and how much ramp it runs (mana rocks get partial credit toward the ramp term, since — unlike a land — a rock can be removed or drawn dead late). A lands deck, with four or more cards that sacrifice lands or bring them back (Crucible of Worlds, Splendid Reclamation, Scapeshift), is never told to trim lands: there, extra lands are fuel.

This is a goldfish simulation, so it's an honest floor for consistency, not a ceiling: it does not model opponent interaction, card draw spells, tutors, cost reducers, cycling, or fetch-land thinning. A deck with a pile of card draw or tutors will, in practice, be more consistent than these numbers alone suggest — the simulation tells you what your lands and ramp can do on their own.

Turn Clock

The Turn Clock plays the whole deck, not just its mana: 1,500 seeded games against three opponents at 40 life who never block, remove, counter or race. Each game mulligans, draws, plays lands, ramps, tutors, casts its commander and threats, attacks, drains, burns and poisons, and after every step checks whether the deck can assemble one of its own combos. It reports the turn the first opponent dies and the turn the whole table dies, as a median plus the fastest and slowest tenth of games.

Every card is read from its own rules text once: mana it makes (and whether it keeps making it, so a Mana Vault or Lotus Petal counts once and a Signet nets one), cards it draws, what it tutors for, power, haste, double strike, infect, tokens and token doublers, anthems (a typal one like "Goblins you control get +1/+1" counts only for the share of the deck's creatures that are Goblins), +1/+1 counters that grow the team or the commander every turn, one-turn pumps like Overrun (held until there is a team to pump), equipment, drains, life gain (lifelink, Soul Warden-style triggers, drains that gain life), "whenever you cast" payoffs, cost reducers, extra combats and damage doublers.

Payoffs count once per event, so a landfall card triggers for every land (fetches twice), an enters-the-battlefield drain for every creature, an aristocrat drain like Zulaport Cutthroat or Blood Artist for every creature that dies, and a life-gain payoff like Sanguine Bond or Marauding Blight-Priest for every time you gain life (each lifelink attacker separately). Panharmonicon and Yarok double enter triggers and Teysa Karlov doubles death triggers. Cards that put creatures from your hand onto the battlefield are read by their wording: Kaalia and Ilharg when they attack, Goblin Lackey when it connects, Sneak Attack and Elvish Piper when activated, Ghalta and Tooth and Nail when they resolve. Lands that ask for another land are played only when they can stay: bounce lands like Simic Growth Chamber return one to your hand, Lake of the Dead sacrifices a Swamp, and Rupture Spire needs a mana to keep. Selvala, Explorer Returned's parley turns over a card for every player each turn, with your opponents' cards taken to be nonland as often as yours, for a green mana each, and draws you a card. Extra turns are played out, X spells scale with the mana on hand, mana doublers and extra land drops apply, cascade and discover cast a free spell, and blink, clones, hydras, Shared Animosity and walls that attack with their toughness (Arcades) are read too. Mill wins when two opponents run out of cards, and Approach of the Second Sun wins on its second cast.

Combos come from Commander Spellbook for your exact list, using Spellbook's own mana and zone requirements. A combo that needs a template card ("an artifact that costs one") only counts when the deck actually holds a card that fits it, and a piece that has to be cast as part of the combo, like Thassa's Oracle, is kept in hand rather than played early. Combo decks spend their free mulligan hunting for a tutor or a piece, and fire rituals into a burst of cards (Ad Nauseam, Necropotence, a wheel) to find the last one.

Read it as the deck's speed limit, not a prediction of your next game. Real games run slower, and a few plans aren't modelled: Doomsday-style piles, long chains of one-off loops and cards whose effect depends on what opponents do. The graveyard is played (Entomb into Reanimate, regrowing a missing piece, casting from the graveyard like Muldrotha, mass returns like Living Death), storm decks go off when the count is lethal, and a commander that transforms for mana (Kefka) does so when it can. The panel lists the cards it couldn't read, so you can see how much of your deck the clock is built on. Every time the engine changes we check it against a fixed set of 42 real decks, from precons and casual lists to full tuned and cEDH builds, and the groups must keep their order. We also run it across the 100 most-played commanders on EDHREC to find the plans it doesn't read yet.

Next to the number, a scale shows where the deck sits against our reference decks (precons clear the table around turn 10, tuned decks around 8, cEDH around 7). Pro also shows a typical game that ends on the median turn, played out card by card, and what drives the clock: every nonland card is swapped for a blank and the deck replayed, with the same shuffles, to measure how many turns that card saves. That replay runs in your browser, only when you scroll to it.

Under the clock you can tell us whether your deck usually wins earlier, around then, or later in your games. Answers are anonymous and are what we tune the clock against: across many decks they show whether it ranks decks the way real tables do, and how far real games run behind the goldfish at each bracket.

Combo Detection

Every analyzed deck's commander(s) and 99 are checked against a live combo database covering thousands of known interactions, returning exact combo lines, prerequisites, and results. If that lookup can't be reached, EDHcheck falls back to a curated list of around 30 well-known 2-card infinites (e.g. Thassa's Oracle + Demonic Consultation, Kiki-Jiki + Pestermite) and checks whether every card in a combo appears in the decklist. Whichever source finds more combos is used. Detected combos are shown step-by-step in the Breakdown panel and feed directly into both the Bracket assignment (an early combo can push a deck to Bracket 4 or 5) and the Combo Presence power-score component.

Data Sources

• Scryfall API — Card images, oracle text, mana costs, color identity, and prices (TCGplayer in US dollars, Cardmarket in euros).
• Commander Spellbook— Live infinite-combo detection for the deck's exact commander(s) and 99.
• EDHREC Salt Data — Community salt scores for roughly 65 of the most-flagged annoying cards.
• Archidekt, Scryfall, Deckbox, TappedOut and MTGTop8 — Deck-link imports, read from each site's own export or API. Moxfield links can't be read yet, so export the list from Moxfield and paste it.

Consent tooling (Cookiebot) and ad tooling (Google AdSense) do not change the core deck scoring model. They are independent of bracket and power calculations.