Image courtesy of Scryfall.com
Embeddings that see patterns in white aggression and graveyard-hate: a practical look at card clustering
In the world of MTG analytics, embedding spaces are the modern spellbook. They let us translate a card’s identity—its mana cost, its color, its abilities—into a vector that a machine can swim through. The aim? Discover natural groupings of cards that share strategy, theme, or edge-case interactions, even when those cards come from different sets or printings. Today we’ll explore this through a focused lens: the creature Landscaper Colos from Modern Horizons 2 (MH2) and what its design tells us about generating meaningful clusters in a white-centric space 🧙🔥💎⚔️.
Card snapshot: Landscaper Colos in a single breath
- Name: Landscaper Colos
- Set: Modern Horizons 2 (MH2), draft_innovation
- Rarity: Common
- Mana Cost: {5}{W} (CMC 6)
- Type: Creature — Goat Beast
- Power/Toughness: 4/6
- Keywords: Landcycling, Basic landcycling, Typecycling, Cycling
- Oracle text: When this creature enters, put target card from an opponent's graveyard on the bottom of their library. Basic landcycling {1}{W} (Discard this card: Search your library for a basic land card, reveal it, put it into your hand, then shuffle.)
Josu Hernaiz’s artwork gives Landscaper Colos a sturdy, earthy presence—an image that blends brute chassis with a gardener’s patience. The card’s white identity and its late-game body (4/6 for six mana) sit at an interesting intersection: a value creature that doubles as graveyard disruption and a tutor for lands through landcycling. It’s a film reel worth rewinding when you think about how color, cost, and text create a signature in an embedding space 🎨.
What Landscaper Colos teaches about embedding features
When we translate MTG cards into embeddings, we lean on a blend of structured features and semantic text. Landscaper Colos is a clean testbed for a few core ideas:
- Color identity and mana value: White cards cluster around defensive play, enter-the-battlefield effects, and utility. A high-CMC white card with a heavy investment (5 colorless and 1 white) often marks mid- to late-game ramp or stabilizers rather than early pressure. In embeddings, mana cost is a compact proxy for when a card might show up in a game plan—this helps separate Colos from cheaper or more aggressive white creatures.
- Creature type and body mass: Goat Beast is a flavorful mix, but in embeddings it also hints at stat distribution and possible synergies with tribal or creature-based strategies. A 4/6 frame is sturdy; in clustering terms, it nudges the vector toward durable bodies that scale with board presence rather than fragile evasive creatures.
- Abilities that impact the stack or graveyard: The enter-the-battlefield trigger against an opponent’s graveyard makes it a node in graveyard-hate clusters. Embeddings capture this as a “disruption” feature, distinguishing it from cards that draw, ramp, or directly pump power.
- Keyword surface area: Landcycling, Typecycling, and Cycling all act as signals. They reveal a design pattern: convertible resources and situational utility. In many clustering tasks, cards with landcycling cluster near other cycling or cycling-like effects, even if their base color or CMC differs.
- Set and rarity context: MH2’s draft_innovation frame influences how players perceive value in a clustering sense. While rarity (Common) is a label used by collectors, in embeddings it signals frequency of appearance in datasets; common cards tend to populate clusters more densely.
From tokens to topology: how you craft MTG embeddings
There are several practical routes to building a card embedding that can reveal the kinds of relationships Landscaper Colos embodies:
- Feature-engineered vectors: Encode color identity, mana value, power/toughness bands, and presence of keywords as discrete features. Add a binary flag for each keyword (Landcycling, Basic landcycling, Typecycling, Cycling) to give Colos a recognizable signaling pattern in the space.
- Textual semantics: Use a transformer-based embedding for the Oracle text, then combine it with structured features. This helps text-heavy cards align not just by mechanics but by the ideas those mechanics express: interruption, redundancy (multiple cycling options), and tempo shifts via ETB effects.
- Set-aware normalization: Normalize data across sets but retain set-origin bias to observe how MH2 cards sit relative to other white cards from different print runs. It’s a gentle reminder that clustering is as much about data composition as it is about card design.
- Hybrid distance metrics: Leverage a mixed metric—cosine distance for text-derived features, and Euclidean distance for numerical features like mana value and power/toughness. Landscaper Colos tends to pull toward mid- to late-game white utility cards, a pattern your model will catch with the right weighting.
“When this creature enters, put target card from an opponent's graveyard on the bottom of their library.” Landscape-level disruption with a land-focused toolkit makes it a neat bridge card for white’s toolbox—both in play and in a well-tuned embedding space.
A practical clustering scenario featuring the MH2 landscape
Imagine you’re building a dataset of Modern Horizons 2 cards and you want to discover natural groupings that evergreen players care about. One cluster might center on white cards with graveyard interaction and a side path into land support. Landscaper Colos sits right at the crossroads: it’s a mid/high-CMC body with a disruptive ETB and a useful landcycling aura. In a well-trained embedding space, you’d expect to see it near cards like other white creatures with graveyard interaction, or near the family of landcycling/typing cycling cards—cards that hint at a broader white strategy of resource conversion and library manipulation 🎲.
Grouping cards by these features isn’t just academic. It helps deck builders, content creators, and collectors understand meta-flexible archetypes and sculpt their own analyses around recurring design signals. And in the real world, it helps you curate your collection with more than just raw rarity or price—seeing how a card’s text and mana curve sit within a cluster can reveal surprising synergies and overlooked value.
Design, culture, and the joy of the hunt
MTG is a living catalog of ideas, and Modern Horizons 2 embodies that spirit with a mix of reprints, new mechanics, and cross-set echoes. Landscaper Colos, with its clear white identity and a toolkit that rewards planning, is exactly the kind of card that makes embedding enthusiasts smile. The looping mechanics—landcycling, basic landcycling, typecycling, cycling—are subtle yet powerful design signals that you can tease apart in a clustering exercise and then re-assemble into playable decks. It’s little moments like these that remind us why we fell in love with the game in the first place: that blend of strategy, lore, and clever engineering that makes every draft feel like an exploration 🧙🔥🎨.
Speaking of clever engineering, if you’re at your desk setting up for a long night of MTG data dives or warm-up drafts, a little glow goes a long way. This Custom Neon Desk Mouse Pad is a friendly companion for your workspace—perfect for visualizing clusters on a multi-monitor setup or simply brightening your analysis sessions with a splash of neon style ⚔️💎. It’s a small nod to how aesthetics and analytics intersect in the MTG community.
For fellow collectors and developers who want to dive deeper, consider pairing your card experiments with a reliable data source like Scryfall. The card arts, exact wording, and set details become part of the corpus you map, weigh, and cluster. And if you’re looking to bring that same energy into your workspace, the product link below is a natural fit for your desk CCG vibes.
Curious minds can explore Lands, cycles, and the broader MH2 roster in more detail through community hubs like EDHREC, Gatherer, and TCGPlayer—the kind of resources that make card clustering feel less like a solitary puzzle and more like a shared, evolving exploration 🧙🔥🎲.