High-Frequency Trading & Betting API
A comprehensive architecture for UK/IE/NZ/AU greyhound markets. Built for ultra-low latency, deterministic reproducibility, and operational safety.
- Target Latency
- <100 ms
- Event Frequency
- 150 / day
- Integrity
- Deterministic
- Risk Control
- Pre-Trade
Core Design Principles
System Blueprint
Interact with the system diagram below to explore the data pipeline, from raw feed ingestion to order execution. The architecture prioritizes a Batch/Stream Hybrid design to handle high frequency updates.
Select Component
Click any block in the diagram to view its technical specifications, data responsibilities, and key constraints.
Canonical Data Model
The backbone of the system. All endpoints and historical replay logs use these exact schemas.
The Race Entity
The `race_id` is globally unique and immutable. However, the race object is versioned. If a runner is replaced or a declared time changes, `race_version` increments.
- Globally Unique ID: Prevents collisions across meetings.
- Versioning: `version` and `changed_at` fields track lifecycle edits.
- Runner Metadata: Includes permanent stats like `age_months` and `trainer_id` for ML features.
// Select a tab to view the schema
SLA & Latency Architecture
Pipeline Latency Targets
* End-to-end latency from Exchange Pub to Client Sub
| Client Type | Rate Limit | SLA |
|---|---|---|
| Streaming (Premium) | Unlimited Subs | ≤ 100 ms |
| Snapshot (REST) | 60 req/sec | ≤ 200 ms |
| Orders | 20 req/sec | ≤ 100 ms Ack |
Deterministic vs. Speed
We prioritize Replayability. Normalization adds ~5ms but allows perfect backtesting parity, crucial for model debugging.
Colocation
Optional Enterprise tier. Reduces latency by ~40ms but increases operational complexity. Only recommended for arb strategies.
