design-arbitrage

What I do

• •Provide architecture frameworks for latency-sensitive trading systems
• •Document tick-to-trade pipeline design and execution engine patterns
• •Define risk management as a first-class design constraint
• •Cover cross-venue and cross-chain arbitrage patterns

When to use me

Use this skill when designing trading systems, arbitrage bots, or latency-sensitive execution infrastructure. Pair with design-core for foundational design principles and decision frameworks.

Latency Budget Framework

Allocate your total latency budget across pipeline stages. Every microsecond matters -- measure, don't guess.

Rule: Measure end-to-end latency at p99, not p50. The tail kills profits.

Tick-to-Trade Pipeline Architecture

Market Data Feed(s)
  |
  v
Feed Handler (normalize, deduplicate, sequence)
  |
  v
Order Book Reconstruction (L2/L3 book maintenance)
  |
  v
Signal Engine (strategy logic, opportunity detection)
  |
  v
Risk Gate (pre-trade checks, position limits, exposure)
  |
  v
Execution Engine (order routing, smart order routing)
  |
  v
Confirmation Handler (fill tracking, position update)
  |
  v
Post-Trade (reconciliation, PnL, reporting)

Critical path: Feed Handler -> Signal -> Risk -> Execution. Everything else is off the hot path. Never add latency to the critical path for logging, metrics, or persistence.

Market Data Normalization & Distribution

• •Feed handlers -- One per venue. Normalize to internal format at the edge. Binary protocols (FIX/FAST, WebSocket binary) over JSON.
• •Order book reconstruction -- Maintain local order book from incremental updates. Detect gaps and request snapshots. Never trust stale books.
• •Multi-venue aggregation -- Merge books across venues for best bid/offer (BBO). Account for fees, latency, and fill probability.
• •Distribution -- Shared memory or lock-free ring buffers for intra-process. Kernel bypass (DPDK, io_uring) for inter-process.

Execution Engine Patterns

Default choice: Event-driven single-threaded for simplicity. Move to lock-free pipeline only when measured latency demands it.

Smart Order Routing

• •Venue selection -- Route to venue with best price after fees. Factor in historical fill rates and latency.
• •Order splitting -- Split large orders across venues to minimize market impact. Use TWAP/VWAP for size.
• •Retry logic -- Rejected orders retry on alternate venues. Never retry without checking current position and risk limits.

Risk Management as Design Constraint

Risk checks are on the critical path. They must be fast AND correct.

Kill switch is non-negotiable. It must work independently of the trading system. Hardware kill switch preferred. Test it weekly.

Co-location & Infrastructure Decisions

• •Co-locate when latency is the primary competitive advantage
• •Central hub with low-latency links for cross-venue strategies
• •Own node for DEX strategies; connect to block builders / private mempools
• •Cloud in nearest region for moderate latency requirements
• •Hardware -- FPGA for nanosecond-critical feed handling; GPU for parallel signal computation; commodity hardware for everything else
• •Network -- Dedicated NICs, kernel bypass (DPDK), jumbo frames. Measure and minimize jitter, not just average latency.

Cross-venue / Cross-chain Arbitrage Patterns

Atomic execution eliminates leg risk but limits you to single-chain, single-block opportunities. Non-atomic execution accesses more opportunities but requires hedging and position management.

Anti-Patterns