Turing machine in Bletchley Park
Turing machine in Bletchley Park
Our infrastructure is designed to provide low-latency and reliable access to blockchain mempools we operate on, along with an execution system optimized for high-frequency algorithmic trading.
Take the example of Arbitrum:
we operate a globally distributed network of Arbitrum nodes, strategically positioned to enable near-instant execution.
This geographic distribution allows us to be among the first to receive newly issued transactions, before they fully propagate across the network and are added to the blockchain. This data feeds our LiveFeed¹, which instantly delivers them to our algorithms for computing optimal strategies.
The models are hosted as close as possible to the sequencer, in a location dynamically determined through our latency benchmarks, in order to minimize the time between receiving information and making a decision.
When one of our algorithms identifies an opportunity or needs to execute an action, the transaction is sent to the nodes that our measurements indicate are closest to the sequencer.
We also optimize this process by leveraging MEV² mechanisms such as Timeboost to maximize our inclusion priority.
In parallel, we operate a dedicated data infrastructure for research, backtesting, and model training. We run our own indexers across all blockchains and protocols we operate on, including high-volume environments such as Hyperliquid. This allows us to build a complete, consistent, and usable historical dataset for quantitative development.
This data infrastructure relies on two complementary databases: the first is used for large-scale analytics thanks to its high-performance on massive queries and historical data exploration; the second is chosen for its transactional robustness and versatility, ideal for storing structured data, configuring strategies, and orchestrating our internal services. Together, these components form a coherent architecture, optimized both for low latency in production and for the analytical depth required for advanced quantitative strategy development.
¹LiveFeed refers to our real-time data stream that aggregates and transmits raw mempool and state-transition information from our distributed nodes directly to our models. It is designed to deliver incoming transactions with minimal delay, ensuring the algorithms operate on the earliest possible view of network activity.
²MEV (Maximal Extractable Value) is the additional value that can be captured by influencing how transactions are ordered, included, or excluded within a block. On rollups like Arbitrum, MEV often involves mechanisms that affect sequencing priority—such as Timeboost—to improve execution timing or positioning in the block.
┌───────────────────────────────────────────────────────┐
│ Low-Latency Access Layer │
├───────────────────────────────────────────────────────┤
│ - Globally distributed blockchain nodes │
│ - Strategic geographic placement │
│ - Early mempool transaction capture │
│ - Pre-propagation transaction visibility │
└───────────────┬───────────────────────────────────────┘
│ Raw mempool events
▼
┌───────────────────────┐
│ LiveFeed │
├───────────────────────┤
│ - Real-time ingestion │
│ - Normalization │
│ - Zero-copy routing │
└───────────────┬───────┘
│
▼
┌───────────────────────────────────────────┐
│ Algorithmic Engine │
├───────────────────────────────────────────┤
│ - Strategy computation │
│ - Real-time model inference │
│ - Signal generation │
└───────────────┬───────────────┬───────────┘
│ │
Strategy outputs Model outputs
│ │
▼ ▼
┌───────────┐ ┌────────────┐
│ Strategies│ │ Models │
└─────┬─────┘ └─────┬──────┘
│ │
└──────┬────────┘
▼
┌────────────────────────────────┐
│ Execution Layer │
├────────────────────────────────┤
│ - Sequencer-proximity routing │
│ - Dynamic latency benchmarks │
│ - MEV-aware submission │
│ (TimeBoost, priority paths) │
└────────────────┬───────────────┘
│ Transactions
▼
┌─────────────────────────────────┐
│ Blockchain Sequencers │
└─────────────────────────────────┘
┌───────────────────────┐
│ LiveFeed │
│ (real-time events) │
└───────────┬───────────┘
│
│ 1. raw events (mempool, trades, states…)
▼
┌──────────────────────────────────┐
│ Raw Events DB (Undecoded) │
│ - append-only │
│ - chain-specific binary payload │
└───────────┬──────────────────────┘
│
│ 2. ETL / decoding jobs
▼
┌──────────────────────────────────────┐
│ Decoding & Enrichment Services │
│ - protocol decoding │
│ - normalization (common schema) │
│ - enrichment (labels, features) │
└───────────┬───────────────┬──────────┘
│ │
3.a decoded │ │ 3.b decoded
analytics │ │ transactional
▼ ▼
┌─────────────────────────┐ ┌─────────────────────────┐
│ Analytics DB │ │ Operational / Meta DB │
│ - columnar / OLAP │ │ - OLTP / transactional │
│ - full history │ │ - strategies, configs │
│ - massive scans │ │ - model registry, runs │
└───────────┬─────────────┘ └───────────┬─────────────┘
│ │
│ 4. data access │
▼ ▼
┌────────────────────────────┐ ┌────────────────────────────┐
│ Quant Research & Backtests │ │ Internal Services │
│ - simulations │ │ - orchestration │
│ - pnl & risk analysis │ │ - monitoring & dashboards │
│ - strategy design │ │ - feature delivery to prod │
└────────────────────────────┘ └────────────────────────────┘
▲
│ 5. feedback (new features, labels, configs)
└───────────────────────────────┘