Schelling Points TODO
Concept
A Schelling point is the option that participants naturally converge on when they cannot communicate with each other. Coordination games have multiple equilibria, and the payoff structure alone doesn't determine which equilibrium actually gets chosen — the key insight is that factors outside the game itself, such as salience, simplicity, and cultural context, decide the outcome. For a Schelling point to work, it's not enough for the option to merely stand out — there must be shared knowledge that everyone else also recognizes it as standing out. This is why Schelling points can produce coordination without any enforcement, but also why they collapse easily when the context changes. The persistence of standards, defaults, and old conventions is usually explained by this same coordination effect.
Many agreements that hold up aren't backed by enforced rules but purely by mutual expectation that everyone else will pick the same option — and protocol forks, standards adoption, and oracle voting are exactly this kind of structure.
Code & Formula
# 셸링 포인트 — 순수 조정 게임에는 대칭적인 내시균형이 여러 개 존재하지만,
# 게임 밖의 "현저성(salience)"이 그중 하나를 특별히 눈에 띄게 만들어 조정을 가능케 한다.
locations = ["Grand Central 시계탑", "Times Square 한복판", "무명 주차장 B구역"]
n = len(locations)
# 순수 조정 게임 payoff: 두 참가자가 같은 곳을 고르면 1, 다르면 0 (완전 대칭)
payoff = [[1 if i == j else 0 for j in range(n)] for i in range(n)]
# 대칭 payoff 행렬에서 순수전략 내시균형은 "둘 다 같은 곳을 고르는" 모든 대각선 칸
pure_nash = [(locations[i], locations[i]) for i in range(n) if payoff[i][i] == 1]
print("payoff 만으로 찾은 순수전략 내시균형 (모두 동등):")
for a, b in pure_nash:
print(f" - ({a}, {b})")
# 게임 자체는 이 균형들을 구분하지 못한다. 현실에서는 "얼마나 유명하고 서로 알 만한가"
# 라는 현저성 점수가 선택을 결정한다 — 이것이 셸링 포인트.
salience = {"Grand Central 시계탑": 0.9, "Times Square 한복판": 0.95, "무명 주차장 B구역": 0.05}
focal_point = max(locations, key=lambda loc: salience[loc])
print(f"\n현저성 점수: {salience}")
print(f"셸링 포인트(예측되는 실제 선택): '{focal_point}'")
print("→ payoff 구조는 동일해도, 공유된 현저성이 균형을 하나로 좁힌다.")
Exercise
Pose a coordination question with no objectively correct answer to a few colleagues without letting them communicate, collect the distribution of answers, and explain why responses clustered on a particular option in terms of salience and shared knowledge.
Practical Connection
Which chain gets treated as the 'real' one after a hard fork, and which outcome the majority is expected to vote for in a dispute, are both Schelling point problems — and UMA-style optimistic oracles are built on the assumption that the honest answer is the Schelling point.
If you study this on a given day, add a note link and a ✅ to this line in the source curriculum (docs/knowledge/math-50-curriculum.md) and this spot will lead straight to the note body. You can also write directly on this page — but regenerating overwrites it, so it's safer to keep anything you want to save as markdown under docs/algorithms/.