How Manayard's centrality and effectiveness scores work

October 18, 2026

Every archetype page on Manayard shows two scores under each card: centrality and effectiveness, each from 0% to 100%. Here's what those numbers mean, how they're calculated, and why they often disagree with popularity.

What the scores measure

Centrality answers: how directly does this card advance the archetype's strategy?

A 100% centrality card is the card you'd name first when describing what the deck does. A 0% centrality card is technically labeled for the archetype but serves a different strategy — or works against it.

Effectiveness answers: how well does the card execute what it does?

This captures efficiency, repeatability, and reliability. A card that does something powerful once scores lower on effectiveness than a card that does it repeatedly. A card with heavy restrictions scores lower than one that works unconditionally.

Both scores are independent. A card can be highly central but ineffective (a powerful effect that's hard to use), or effective but not central (a well-designed card for a different strategy).

These are not popularity. A card played in thousands of decks can score low on centrality if it doesn't serve the strategy. A card almost nobody plays can score high if its text is perfect for the job.

Step 1: Labeling — determining the population

Before scoring, we determine which cards belong to each archetype. A card gets labeled for an archetype when multiple signals agree:

1. Parsed rules: Our parser reads the card's Oracle text into structured operations. If it finds a `create_token` operation, that's a signal for the tokens archetype. If it finds `sacrifice`, that's a signal for aristocrats. 2. Oracle text phrases: Phrase matching on the raw text (e.g., "create a token", "tokens you control"). 3. Scryfall tags: Scryfall's curated functional tags.

A card is labeled when two or more signals agree, or when one highly specific signal matches. This gives us the population: 2,257 cards for tokens, 1,549 for lifegain, 2,166 for counters, and so on across 38 archetypes.

Labeling is imperfect. Some cards that should be labeled aren't (we're working on parser coverage). But the population is what the scoring compares within.

Step 2: Scoring — reading the full text

Within each archetype's population, a language model reads every card's full Oracle text and assigns centrality and effectiveness scores from 0 to 1.

The model considers:

• Strategic centrality: Does the card's text directly serve the archetype's game plan? A token doubler is more central to tokens than a card that incidentally makes one token. A card that gives your opponents tokens scores near zero — it works against the strategy.
• Effectiveness: How efficiently does it work? Is it repeatable or one-shot? Are there heavy restrictions or costs? A repeatable engine scores higher than a one-shot effect. An unconditional effect scores higher than a heavily gated one.
• No setup penalties: We assume the strategy matters and rate the card's contribution in isolation. If you make the payoff happen, how good is it? Setup costs are not penalized — that's a deck-building question, not a card-assessment question.

The scoring is comparative within each archetype. Cards are judged against the archetype's population, not against a fixed rubric. This is why the same card can score highly in one archetype and poorly in another.

Step 3: The ranking

Cards are ordered by centrality (highest first), then effectiveness. This ordering determines what you see on the archetype pages — the most central cards appear first.

The percentages you see are the raw scores multiplied by 100. A card with centrality 0.95 shows as 95%. These are not percentiles — they're absolute assessments of how well the card serves the strategy.

Where each number on a card comes from

Theme pages scored by Haiku cover the top 100 cards; those scores are display-only. Variant and component pages use a separate Haiku run that deck plans read.

Why these scores differ from EDHREC

EDHREC ranks by how many decks play a card. Our scores rank by how well the card's text serves a specific strategy. These measure different things:

• Orcish Bowmasters is in thousands of decks (high EDHREC rank) but scores modestly on centrality for tokens — it's a good card, not a tokens card.
• Caradora, Heart of Alacria is barely played (EDHREC #11,614) but scores highly for counters — its text is ideal for the strategy.

Neither is wrong. They answer different questions. EDHREC tells you what the crowd does. Centrality tells you what the card does for the strategy.

Limitations

• Labeling gaps: If a card isn't labeled for an archetype, it can't be scored. We're improving parser coverage to catch more cards.
• Text isn't everything: The model reads rules text, not gameplay context. It can't know that a card is better in practice than its text suggests, or that a meta shift changed its value.
• No cross-archetype comparison: A card's centrality in tokens says nothing about its centrality in aristocrats. Each archetype is scored independently.

These scores are a starting point for brewing, not a final answer. Read the card, consider your deck, and use the rationale in each card's detail view to see why it scored where it did.