a market mechanism for prediction markets where the two sides of a bet are geometrically coupled — buying one directly suppresses the other

proposed by Nick Williams and Vitalik Buterin, Ethereum Research, 2020


the core idea

standard prediction markets bound prices to [0,1] because shares settle to fixed payouts of $$C(s_{YES}, s_{NO}) = \lambda \sqrt{s_{YES}^2 + s_{NO}^2}$$. ICBS presents an alternative: settlement rebalances reserves rather than paying fixed amounts. prices are not bounded to [0,1]. instead, the ratio of reserves encodes the market's probability forecast.

in traditional bonding curves, buying token A doesn't affect token B's price. ICBS couples them: buying YES pushes NO's price down, and vice versa. this creates genuine opposition between beliefs rather than independent liquidity pools.


the cost function

$$C(s_{YES}, s_{NO}) = \lambda \sqrt{s_{YES}^2 + s_{NO}^2}$$

where $s_{YES}$ and $s_{NO}$ are token supplies and $\lambda$ is a fixed scaling constant set at deployment.

geometrically: this is the Euclidean distance from the origin in the $(s_{YES}, s_{NO})$ plane. iso-cost curves are circles — every point at distance $r$ from the origin costs $\lambda \cdot r$. trading moves outward from the origin.

$\lambda$ is fixed at deployment by the initial deposit $D$:

$$\lambda = \frac{D}{\sqrt{s_{YES}^2 + s_{NO}^2}}$$

for a 50/50 split at initial price $1, a \$100 deposit creates $s_{YES} = s_{NO} = 50$ tokens, giving $\lambda = 100/\sqrt{50^2 + 50^2} \approx 1.414$. markets of different sizes have identical percentage-based price dynamics — enabling cross-market comparison.


prices and inverse coupling

prices emerge as partial derivatives of the cost function:

$$p_{YES} = \frac{\partial C}{\partial s_{YES}} = \lambda \cdot \frac{s_{YES}}{\sqrt{s_{YES}^2 + s_{NO}^2}}$$

$$p_{NO} = \frac{\partial C}{\partial s_{NO}} = \lambda \cdot \frac{s_{NO}}{\sqrt{s_{YES}^2 + s_{NO}^2}}$$

each token's price increases with its own supply but is suppressed by the opposing side:

$$\frac{\partial p_{YES}}{\partial s_{NO}} = -\lambda \cdot \frac{s_{YES} \cdot s_{NO}}{(s_{YES}^2 + s_{NO}^2)^{3/2}} < 0$$

buying NO directly lowers YES's price. this is the inverse coupling that gives ICBS its name — and that makes it the correct market structure for an epistemic system where TRUE and FALSE are genuine opposites, not independent assets.


the invariant: TVL = cost function

virtual reserves are defined as $r = s \cdot p$:

$$r_{YES} = \lambda \cdot \frac{s_{YES}^2}{\sqrt{s_{YES}^2 + s_{NO}^2}}, \quad r_{NO} = \lambda \cdot \frac{s_{NO}^2}{\sqrt{s_{YES}^2 + s_{NO}^2}}$$

total value locked:

$$TVL = r_{YES} + r_{NO} = \lambda\sqrt{s_{YES}^2 + s_{NO}^2} = C(s_{YES}, s_{NO})$$

TVL always equals the cost function — the on-manifold property. this ensures solvency: total claimable value always matches what the vault holds. reserves can rebalance at settlement without minting or burning tokens.

the market's current probability forecast:

$$q = \frac{r_{YES}}{r_{YES} + r_{NO}}$$


settlement

at resolution, actual outcome $x \in \{0, 1\}$ determines settlement factors:

$$f_{YES} = \frac{x}{q}, \quad f_{NO} = \frac{1-x}{1-q}$$

if the event happens ($x = 1$): YES holders gain ($f_{YES} > 1$), NO holders lose ($f_{NO} < 1$). if it doesn't ($x = 0$): NO holders gain, YES holders lose. reserves rebalance directly:

$$r'_{YES} = r_{YES} \cdot f_{YES}, \quad r'_{NO} = r_{NO} \cdot f_{NO}$$

total vault balance is preserved: $r'_{YES} + r'_{NO} = r_{YES} + r_{NO}$. capital flows from incorrect predictions to correct ones without external capital injection.

settlement uses square-root scaling of the supply parameter $\sigma$ (converting display to virtual supply). scaling $\sigma$ by $\sqrt{f}$ makes virtual supplies scale by $\sqrt{f}$, making reserves (proportional to supply$^2$ via the norm) scale by exactly $f$.


key properties

self-scaling liquidity. buying moves supply further from the origin. TVL $= \lambda\sqrt{s_{YES}^2 + s_{NO}^2}$, so trading volume automatically grows liquidity. no external LPs needed. markets bootstrap from minimal deposits and scale organically. this differs fundamentally from LMSR, where the subsidy parameter $b$ caps liquidity.

early conviction rewards. prices range from 0 to $\lambda$:

$$\lim_{s_{YES} \to \infty,\, s_{NO} \text{ fixed}} p_{YES} = \lambda$$

early traders who buy near zero can see prices approach $\lambda$, yielding arbitrarily large returns. unlike LMSR's fixed [0,1] bounds, ICBS rewards early conviction rather than just tracking consensus. this aligns incentives toward surfacing private knowledge early.

geometric simplicity. only square roots — no fractional powers, no exponentials. the mechanism is computationally tractable and the geometry is intuitive.


ICBS vs LMSR

ICBS LMSR
price bounds [0, λ] [0, 1]
liquidity self-scaling (trading grows TVL) capped by subsidy parameter b
external LPs none needed none needed
settlement reserve rebalancing fixed MATH_PLACEHOLDER_391 payouts
early conviction rewarded (prices can approach λ) not specially rewarded
probability encoding ratio of reserves direct price
loss bound none (market maker takes risk) b·ln(2) per market

connection to cyber

the inverse coupling property is the market analog of inhibition: buying FALSE directly suppresses the effective weight of YES in the market, exactly as negative weights suppress activations in neural networks. the geometry makes this explicit — the two sides move on a circle, so amplifying one necessarily suppresses the other.

the self-scaling liquidity property solves the bootstrapping problem for the cybergraph: every cyberlink that attracts market activity automatically deepens its own liquidity. the most-contested edges (the epistemically important ones) become the most liquid, yielding the most accurate prices. this is the Lindy effect on the market structure.

the settlement factors $f_{YES} = x/q$ and $f_{NO} = (1-x)/(1-q)$ are inverse probability weights — the same mathematical structure that appears in importance sampling, in the serum scoring formula, and in the KL divergence terms that measure information gain. this is not coincidental: all three are instances of proper scoring rules applied to belief elicitation.

the on-manifold property (TVL = cost function) ensures the market remains solvent as cyberlinks accumulate, without requiring external capital injection. the cybergraph itself is the liquidity — structural knowledge (cyberlinks) bootstraps epistemic knowledge (market prices).

see veritas for how ICBS fits into the full truth-discovery protocol. see inhibition for the connection to inhibitory weights in the tri-kernel. see serum for the scoring layer that sits above the market mechanism. see market for the broader design.

Local Graph