Routing

Cost Function

C = λ₁·(price impact) + λ₂·(compute cost) + λ₃·(latency) + λ₄·(privacy cost)

• Price impact from pool reserves or maker depth.
• Compute cost approximates Solana CU and tx size.
• Latency based on historical maker/adapter timings.
• Privacy cost adds smearing/tie‑break randomness.

Candidate Generation

• RFQ legs: top‑N makers by past hit‑rate and freshness.
• AMM legs: k‑shortest paths over pool graph with fee/impact weights.
• Hybrid: enumerate splits across RFQ and AMM within slice bounds.

Selection Algorithm

for each candidate route r in R:
  score[r] = C(r)
best = argmin(score)
// tie-break with privacy jitter
if |score - score[best]| < ε for some r:
  choose randomly weighted by exp(-score/τ)
        

Constraint Handling

• Respect minOut, maxSlippage, and maxHops per profile.
• Reject routes requiring unsupported compute/IX combos.
• Ensure legs are atomic under router’s state machine.