2nd/10_Wiki/Topic_Programming/AI_and_ML/덱 빌딩 (Deck building).md

id, title, category, status, canonical_id, aliases, duplicate_of, source_trust_level, confidence_score, verification_status, tags, raw_sources, last_reinforced, github_commit, tech_stack

id	title	category	status	canonical_id	aliases	duplicate_of	source_trust_level	confidence_score	verification_status	tags	raw_sources	last_reinforced	github_commit	tech_stack
wiki-2026-0508-덱-빌딩-deck-building	덱 빌딩 (Deck building)	10_Wiki/Topics	verified	self	Deck Building Deckbuilding 덱빌딩 TCG Deck Construction	none	A	0.9	applied	game-design tcg ccg strategy balance		2026-05-10	pending	language framework text TCG/CCG game design

덱 빌딩 (Deck building)

매 한 줄

"매 60장 안에 win condition / mana curve / interaction / consistency 의 balance를 압축하는 행위". Magic: The Gathering (1993)에서 시작한 덱 빌딩은 Hearthstone (2014) · Marvel Snap (2022) · Pokémon TCG Pocket (2024)을 거치며 mobile-first 12-card mini deck → 60-card constructed의 spectrum으로 evolve. 2026 현재 LLM-assisted deckbuilder (e.g. Untapped.gg AI Coach)가 meta-aware suggestion을 제공.

매 핵심

매 4 Pillar of Deck Construction

• Win Condition: 매 매치 종료시키는 primary path (combo / aggro burn / control late-game / mill).
• Mana Curve: 매 cost-by-turn distribution — 1-drop ~8, 2-drop ~12, 3-drop ~10, 4+ tapering.
• Consistency: 매 draw engine + tutor (search effect) — variance 감소.
• Interaction: 매 removal + counterspell — 상대 plan 방해.

매 Format별 제약

• Constructed (60+): max 4-of (Hearthstone 2-of), banlist 적용.
• Limited / Draft (40): 매 sealed pool에서 on-the-fly 구성.
• Singleton / Commander (100): 매 1-of, 100-card highlander.
• Marvel Snap (12): 매 ultra-compressed — 매 card가 "tech choice".

매 응용

1. Pokémon TCG Pocket pack-opening + 20-card deck — onboarding-first design.
2. Slay the Spire roguelike deckbuilder — run마다 deck 점진 구성.
3. MTG Arena Brawl — 60-card singleton + commander.
4. Hearthstone Battlegrounds — auto-battler에 deckbuild 제거, hero-pool curation.

💻 패턴

Mana Curve Validator

from collections import Counter

def validate_curve(deck: list[dict]) -> dict:
    """deck = [{'name': str, 'cmc': int, 'count': int}, ...]"""
    curve = Counter()
    total = 0
    for c in deck:
        curve[min(c['cmc'], 7)] += c['count']
        total += c['count']
    pct = {k: v / total for k, v in curve.items()}
    ideal = {1: 0.13, 2: 0.20, 3: 0.17, 4: 0.13, 5: 0.08, 6: 0.05, 7: 0.04}
    deviation = sum(abs(pct.get(k, 0) - v) for k, v in ideal.items())
    return {'curve': dict(curve), 'total': total, 'deviation': deviation}

Hypergeometric Draw Probability

from math import comb

def draw_at_least_one(copies: int, deck_size: int = 60, draws: int = 7) -> float:
    """Opening hand에 specific card 포함 확률."""
    miss = comb(deck_size - copies, draws) / comb(deck_size, draws)
    return 1 - miss

# 매 4-of in 60: ~40% to see in opening hand
print(draw_at_least_one(4))  # 0.3994

Archetype Classifier (LLM-assisted)

import anthropic

client = anthropic.Anthropic()

def classify_archetype(decklist: list[str]) -> str:
    msg = client.messages.create(
        model="claude-opus-4-7",
        max_tokens=200,
        system="You are an MTG meta analyst. Classify decks as Aggro/Midrange/Control/Combo/Tempo.",
        messages=[{"role": "user", "content": "\n".join(decklist)}],
    )
    return msg.content[0].text

Snap-style 12-Card Synergy Score

def synergy_score(deck: list[str], synergy_graph: dict[tuple, float]) -> float:
    score = 0.0
    for i, a in enumerate(deck):
        for b in deck[i+1:]:
            score += synergy_graph.get((a, b), 0) + synergy_graph.get((b, a), 0)
    return score / len(deck)

Sideboard Optimizer

def optimize_sideboard(main: list[str], meta: dict[str, float],
                      cards: dict[str, dict[str, float]]) -> list[str]:
    """meta = {archetype: weight}, cards[c][archetype] = win_rate_delta."""
    scored = []
    for c, deltas in cards.items():
        if c in main: continue
        ev = sum(meta.get(a, 0) * d for a, d in deltas.items())
        scored.append((ev, c))
    scored.sort(reverse=True)
    return [c for _, c in scored[:15]]

Draft Pick EV

def pick_ev(card: str, picks_so_far: list[str], wheels: dict[str, float]) -> float:
    """Card power × signal × wheel probability."""
    base = card_power(card)
    in_color_bonus = 1.3 if shares_color(card, picks_so_far) else 0.8
    wheel_pen = 1.0 - wheels.get(card, 0) * 0.2  # 매 wheel 가능성 → 후순위
    return base * in_color_bonus * wheel_pen

매 결정 기준

상황	Approach
Mobile onboarding	12-card (Snap) or 20-card (Pocket)
Competitive depth	60-card constructed
Social play	Commander 100-card singleton
Roguelike	Procedural deckbuild (StS)
Beginner	Pre-constructed starter

기본값: 60-card constructed with 4-of legality + 15-card sideboard.

🔗 Graph

• 부모: Game_Design
• 변형: Draft_Mode

🤖 LLM 활용

언제: meta classification, sideboard suggestion, decklist parse → archetype, beginner coaching. 언제 X: 매 micro-frame technical play (turn-by-turn) — Monte Carlo simulator가 superior.

❌ 안티패턴

• Random pile: 매 win condition 없는 60장 collection.
• Curve cliff: 매 mana curve의 4+ slot 비대 → opening hand brick.
• No interaction: 매 pure goldfish deck — 상대 plan 무시.
• Over-tutoring: 매 tutor 8장 + 1-of toolbox → consistency illusion, 실제는 expensive.

🧪 검증 / 중복

• Verified (Frank Karsten Mana Math, MTG Pro Tour data 2014-2025).
• 신뢰도 A.

🕓 Changelog

날짜	변경
2026-05-08	Phase 1
2026-05-10	Manual cleanup — TCG/CCG deck construction 4-pillar + format spectrum + 2026 LLM coach 정리