Strategy

Sail’s optimization engine manages cross-chain portfolios using a balance of analytic scoring and adaptive learning.

Each agent runs a multi-factor function that scores all active markets by expected yield, liquidity depth, and volatility cost, then reallocates based on that score.

1. Scoring Function

Each protocol receives a normalized score based on yield, liquidity, and volatility:

S_i = (A_i^α * T_i^(βγ) * e^(-λσ * σ_i²)) /
      Σ_j (A_j^α * T_j^(βγ) * e^(-λσ * σ_j²))

where:

• A_i = protocol APY

• T_i = liquidity (TVL)

• σ_i = historical volatility

• λσ = risk adjustment

• α, β, γ = sensitivity parameters that adapt to portfolio size

Larger portfolios automatically receive higher sensitivity to APY and liquidity risk, meaning big accounts favor deeper, lower-volatility venues while smaller ones chase lighter, higher-yield pools.

---

2. Cost Awareness

When the agent moves capital between chains or tokens, it models full cross-chain and swap costs:

c_(i→j) = μ_chain * [c_i ≠ c_j] +
           μ_token * [t_i ≠ t_j] +
           slippage_i(P) +
           gas_(i→j) / P

Each swap or bridge is treated as an “energy barrier.” The optimizer only crosses if the expected yield improvement outweighs that cost.

---

3. Optimization Objective

The agent searches for a new allocation w* that maximizes a regularized utility:

L(w) = wᵀS − λ_c * C(w, w₀)
       − η * ||w − w_t||²
       − ν * H(w)

where:

• C(w, w₀) = total cost to move from previous allocation

• H(w) = entropy term that prevents over-concentration

• λ_c, η, ν = cost and diversification controls

The optimizer uses simulated annealing, gradually cooling search temperature to escape local minima and converge near an optimal allocation. A CNN meta-controller provides priors for (α, β, γ) based on volatility and APY history.

---

4. Learn More

Read the full paper: Cross-Chain Optimization Engine for Autonomous Agents