Methodology

Separate ratings, never one combined number

Ratings engine manarank-ratings-2026.1 · Game Changers list wotc-game-changers-2026.08

A deck is never reduced to a single unexplained "power level". Ratings are reported separately because they answer different questions and can disagree with each other.

A. Structural Strength Index

A 0–100 estimate of the decklist as constructed, published with a likely range and a confidence level. Twelve weighted measures: speed, consistency, win conditions, combo access, mana efficiency, interaction, protection, card advantage, resilience, functional redundancy, mana-base reliability and commander dependence (scored inverted — a deck that functions without its commander scores higher). The range widens with low parsing confidence, combo ambiguity, a non-100-card list or a missing commander. It is an estimate from construction, not a verified win rate.

B. Official Commander bracket result

Reported entirely separately from the strength index, against the five optional beta brackets: Exhibition, Core, Upgraded, Optimized and cEDH. Brackets 1 and 2 exclude Game Changers, bracket 3 allows up to three, and brackets 4 and 5 permit an unlimited number.

• Bracket floor — the lowest bracket the deck is permitted to sit in, driven by the Game Changer count first, then by mass land denial, a complete two-card combo line, or extra turns combined with lock pieces.
• Recommended bracket — the floor or the barometer-derived placement, whichever is higher.
• Player-declared bracket — whatever the pilot states; the platform never overwrites it.
• Conflicts — every specific reason the declaration disagrees with the floor, named card by card where a Game Changer is involved.
• Confidence — reported for the bracket placement itself, independent of the strength score's confidence.

Brackets are a tool for matching players with similar game intentions, not a mathematical power ranking. A numeric strength score therefore never sets a bracket, and a bracket never adjusts the strength score.

Standalone Card Score

A 0–100 estimate of a card's general Commander power before any deck context. Ten weighted components are scored 0–100 each; components that do not apply to a card's profile have their weight redistributed, never treated as zero.

• Mana and tempo efficiency15%
• Card and resource advantage12%
• Interaction quality12%
• Immediate impact10%
• Repeatability and scalability10%
• Flexibility and modality8%
• Resilience8%
• Multiplayer scaling8%
• Combo and engine potential10%
• Zone, timing and accessibility utility7%

Penalties are applied separately and capped: setup dependency 0 to −8, narrowness 0 to −6, symmetrical or meaningful drawback 0 to −6, fragility 0 to −5, colour intensity 0 to −3, activation investment 0 to −2, with a −30 total floor. A peer adjustment of at most ±10 compares the card with its type and mana-value band, damped so a weak card is never promoted merely because its peer group is weak.

Deck Fit Score

A 0–100 measurement of how well a card supports one specific deck.

• Commander synergy18%
• Archetype and theme alignment14%
• Card-to-card synergy network16%
• Combo and engine participation12%
• Role need and coverage14%
• Mana-curve fit9%
• Colour and mana-base castability7%
• Redundancy balance5%
• Interaction and resilience balance5%

Contextual Card Score

Contextual = 30% Standalone + 50% Deck Fit + 20% normalized Marginal Contribution − context penalties, clamped to 0–100.

Marginal contribution estimates how much the deck changes when the card is replaced by a neutral baseline of the same role, type and mana-value band, using cached deck aggregates rather than re-running a full analysis once per card. It is deliberately different from Deck Fit: a card can fit the theme perfectly and still add little when the deck already has many similar effects.

Deck Power Score

The deck score is structural, not an average of card scores. Ten weighted components plus explicit deck penalties for illegal configuration, insufficient mana, colour mismatch, missing win conditions, isolated cards and absent commanders.

• Commander alignment and strategy coherence12%
• Mana base and castability13%
• Curve and expected speed10%
• Ramp and resource acceleration9%
• Card advantage and selection9%
• Interaction quality and coverage10%
• Protection, recursion and resilience8%
• Win-condition quality12%
• Consistency, tutors and redundancy8%
• Synergy, engines and combos9%

Bracket estimate

The 1–5 bracket is estimated from detected barometers — combo lines, extra turns, mass land denial, efficient tutors, fast mana, cheap interaction, lock pieces, deterministic wins and average mana value — never from a numeric threshold on the power score. Two decks can score alike numerically and belong in very different pods. The estimate is unofficial; confirm expectations in the pregame conversation.

Confidence

Confidence reflects source completeness, parsing coverage, unusual rules text, variable values and layout complexity. Missing information lowers confidence; it never scores as zero power. High ≥ 80, Medium ≥ 60, otherwise Low.

Target profiles

• Exhibition: lands 37–40, ramp 6–9, draw 6–9, interaction 4–7
• Casual battlecruiser: lands 37–39, ramp 8–11, draw 8–10, interaction 6–9
• Core balanced: lands 36–38, ramp 9–12, draw 9–12, interaction 8–11
• Upgraded: lands 35–37, ramp 10–13, draw 10–13, interaction 9–12
• Optimized: lands 33–36, ramp 11–15, draw 11–14, interaction 10–14
• Competitive: lands 28–33, ramp 12–18, draw 12–16, interaction 12–18

Effect tags

69 deterministic Oracle-text tags are currently defined. Each assignment stores the exact passage that justified it.

Card draw · Cantrip · Loot / rummage · Impulse draw · Wheel · Tutor · Restricted tutor · Scry · Surveil · Self-mill · Opponent mill · Land ramp · Mana rock · Treasure · Cost reduction · Mana doubling · Extra land drop · Creature removal · Permanent removal · Exile effect · Bounce · Counterspell · Board wipe · Graveyard hate · Hand disruption · Edict / sacrifice removal · Hexproof · Ward · Indestructible · Protection · Blink · Recursion · Reanimation · Self-recursion · Uncounterable · Flying · Trample · Menace · Evasive / unblockable · Double strike · Deathtouch · Lifelink · Haste · Goad / forced combat · Additional combat · Damage multiplier · Anthem · Token creation · Token doubling · Counter doubling · Proliferate · +1/+1 counters · Poison / infect · Sacrifice outlet · Death payoff · Enters payoff · Cast payoff · Spell copy engine · Untap engine · Discard outlet · Life-gain payoff · Graveyard payoff · Alternate win condition · Mana / damage outlet · Theft · Clone / copy permanent · Stax / tax · Mana denial · Extra turn

Pod balance: confidence, freshness and uncertainty

Pod engine manarank-pod-2025.1

A pod balance estimate combines three separately reported measurements: power balance (45%), experience balance (40%) and confidence balance (15%). Each is shown on its own so a pod can see why a number is low rather than only that it is.

• Power balance — spread of estimated win chance and bracket45%
• Experience balance — play-pattern clashes and game-length spread40%
• Confidence balance — how much the platform can trust the comparison15%

Confidence levels

Confidence balance starts from the averaged per-card parsing confidence of every deck in the pod (High ≥ 80, Medium ≥ 60, otherwise Low), scaled by 0.72, reduced by 4 points for each deck at Low confidence, and given a +4 allowance once three or more decks are compared. It never raises or lowers a win probability; it only tells you how much weight to put on one.

Data freshness

Card data is read live from Scryfall at the moment a deck is resolved, and the resolution timestamp is shown on the deck dashboard. Resolved names are cached for 30 minutes in the browser session, so a comparison may use card data up to half an hour old. Nothing is precomputed overnight: there is no bulk snapshot, no stored rank percentiles, and no cached combo index behind these numbers yet.

Known uncertainty in every estimate

• Win probabilities come from a softmax over structural strength indices with a deliberately high temperature and a uniform prior. They are damped on purpose: a real Commander pod is high variance, and the model will never claim near-certainty.
• No played game results are recorded yet, so nothing here is calibrated against real outcomes. The match learning engine is a later phase.
• Pilot skill, table politics, threat assessment, mulligan decisions and draw luck are not modelled at all.
• Combo detection is in-deck only; an external combo database is not yet connected, so an unreported line is not proof that none exists.
• Bracket estimates are unofficial, and a bracket spread of one already costs 7 points of power balance — small differences move the score more than they may move a real game.
• Expected game length is a goldfish estimate from curve and speed signals, with no opponent interaction.

Known limitations

• Card data is read live from Scryfall; a full local bulk synchronization with rank percentiles is not yet active, so global ranks and percentiles are not displayed rather than being estimated.
• Combo detection currently derives lines from in-deck enabler/payoff relationships. An external combo database provider is not yet connected, so absence of a line is not proof that none exists.
• Consistency uses exact hypergeometric probability on role counts. It is a goldfish estimate and does not simulate opponents or full rules interactions.
• Ability parsing is regex-driven over Oracle text. Unusual templating lowers confidence and flags the card for review.
• Accounts, cloud-saved decks, share links and the administrator dashboard are queued for the next phase; decks currently save to your browser.

Sources

• Card data and images: Scryfall API (api.scryfall.com).
• Combo data: Commander Spellbook backend (planned provider).
• Commander policy: versioned in-app policy data, administrator editable (planned).