Citrate Technical Paper
A complete protocol specification for a BlockDAG network that fuses GhostDAG consensus (k=18) with BFT finality to achieve 10,000+ TPS at 0.5-second block times with ~12-second finality. The Layered Virtual Machine executes AI inference workloads natively alongside EVM-compatible smart contracts, with the InferenceOracle precompile at address 0x1001 exposing on-chain model invocation as a first-class operation. SALT tokenomics enforce a fixed supply with gas pricing that reflects both compute and inference costs.
Key Contributions
- GhostDAG + BFT consensus mechanism
- Layered Virtual Machine (LVM) architecture
- Model Context Protocol (MCP) integration
- SALT tokenomics and gas model
Explainers
What is GhostDAG?
A DAG-based consensus protocol where parallel blocks are ordered rather than discarded, achieving high throughput without sacrificing security. At k=18, the network tolerates up to 18 parallel blocks per round while maintaining a consistent total ordering for transaction finality.
How does LVM work?
The Layered Virtual Machine separates execution into tiers: a base EVM-compatible layer for smart contracts, and a dedicated inference layer where AI models run inside metered sandboxes. The InferenceOracle precompile at 0x1001 bridges both layers, letting any contract call a hosted model and receive results within the same transaction.
What is Model Context Protocol?
MCP standardizes how AI models receive context, parameters, and return results across the network. It defines a universal interface so any model — regardless of architecture or hosting node — can be invoked with consistent serialization, authentication, and metering.