QuanuX Spreader Architecture

QuanuX Spreader Architecture

[!IMPORTANT] IMMORTAL GUIDANCE FOR AGENTS: When generating or modifying QuanuX-Spreader strategies, you must strictly adhere to the architecture described below. You must perfectly understand the separation of concerns between the Execution Node (a dumb networking stub), the Spreader Engine (the 59ns dual-thread C++ application), and the Strategy (the math formula).

1. The Separation of Concerns

A. The Execution Node (The Remote Shell Stub)

The Execution Node is typically a bare-metal server co-located at an exchange.

• What it does: It manages direct market access (DMA) connections to broker servers and handles NATS/ZMQ cluster telemetry.
• What it DOES NOT do: It possesses zero trading logic. It does not even contain the 59ns execution loop. It sits live and awaits a fully compiled, weaponized trading application to be deployed onto it.

B. The QuanuX-Spreader Engine (The Application)

This is the ultra-high-performance C++ executable. It contains the 59-nanosecond, Lock-Free Dual-Thread architecture (described in Section 2).

C. The Strategy (The Math)

The mathematical formulas and specific instrument relationships (e.g., the spread ratio between CLZ4 and CLF5). These define what the Spreader Engine is calculating on its threads. If the strategy code includes the OnixS SDK and API keys, the engine natively utilizes the OnixS FIX implementation; otherwise, it falls back to QuickFIX (both using identical calling conventions).

2. The 59ns Dual-Thread, Single-Core Architecture

Inside the assembled QuanuX-Spreader application, operations occur on a single pinned CPU core utilizing two highly specialized threads communicating via an L1 cache-aligned quanux::SPSCQueue.

• Thread 1: The Innode Data Pipe (Producer)

  • Ingests raw binary market data from the Execution Node's DMA connection.
  • Instantly formats the payload into a strictly cache-aligned 64-byte quanux::MarketTick.
  • Executes the injected Strategy math directly on this aligned memory (e.g., calculating Locke differentials).
  • If the math triggers, pushes an Event struct to the lock-free queue.
  • Lazily flushes the payload to DuckDB (this is a sideshow, never blocking the math).

• Thread 2: Strategy Order Entry & FIX Layer (Consumer)

  • Spin-waits (_mm_pause()) on the queue in nanoseconds.
  • Pulls the Event.
  • Manages the FIX / OnixS protocol. It holds the direct connection to the broker.
  • The Feedback Loop: Because it executes on the same core, it retains direct reverse-access to the ongoing flow of Thread 1. When Leg A fills via FIX, Thread 2 can aggressively inject the Leg B offset directly against the incoming data stream in sub-microseconds.

3. The Deployment Pipeline

Agents generating spreader strategies must understand the sequence of pushing an idea to production.

1. Generation: AI or analysts draft the complex multi-leg spread formulas and configurations, storing them in server/spreader-strategies/.
2. Assembly & Compilation (quanuxctl spreader package): The quanuxctl CLI injects the specific strategy formula into the QuanuX-Spreader engine codebase. The system compiles the combined components into a single, highly-optimized executable binary.
3. Deployment (quanuxctl spreader deploy): The compiled QuanuX-Spreader application is transmitted over the network and deployed directly onto the waiting, remote Execution Node stub.
4. Live Execution: The Execution Node binds its raw DMA pipes to the newly deployed application, and the 59ns loop begins evaluating live tick data.

4. The 59ns Happy Path: Patterns & Constraints

To execute at 59 nanoseconds, strategy generation and code generation must adhere to strict hardware-level constraints. The engine enforces a "Zero-Drift Policy."

A. The Zero-Jitter Memory Arena

The engine fundamentally treats the OS heap as non-existent during live execution.

• No Allocations: Agents must never instantiate std::make_shared or call new in the Thread 1 or Thread 2 hot paths.
• The Pool: The engine pre-allocates cache-aligned quanux::MemoryPool arrays (fixed-size, power-of-two blocks) during the init() phase. Ticks and events are fetched using bitwise AND pointer math & (Capacity - 1) to eliminate modulo calculation latency and entirely avoid OS page faults.

B. O(1) LOCF Price Matrices

Standard dynamic maps or string-based hashes (std::map<string, double>) are absolutely banned in the hot path.

• Direct Indexing: To execute multi-leg Last Observation Carried Forward (LOCF) spread calculations, the engine uses a cache-aligned PriceMatrix. Strategies must look up leg prices via direct array indexing using a uint32_t instrument_id. This executes in exactly 1 CPU cycle.
• Dirty Bits: Leg updates are flagged using atomic sequence counters (std::atomic<uint32_t> update_seq) initialized with memory_order_release. This allows Thread 2 (FIX Order Entry) to instantly verify state changes via the core's L1 cache without ever locking or traversing the entire queue history.

C. SHM/DMA Handshake Constraints

To prevent subtle memory corruption or L1 cache line eviction during the DMA handshake between Thread 1 (Math) and Thread 2 (Order Execution), the StrategyState payload is hard-capped.

• The 64-Byte Guard: The engine.hpp explicitly enforces a structural size limit: static_assert(sizeof(StrategyState) <= 64).
• Refusal to Forge: If an AI agent attempts to design an overly bloated strategy state that exceeds this single cache line size, the CMake quanuxctl spreader package command will instantly fail the build. Keep your strategy matrix math tightly bundled!

D. Compiler-as-a-Service (CMakePresets.json)

The QuanuX compilation pipeline leverages CMakePresets.json to turn the build environment into a deterministic API. Any fully packaged spreader binary is automatically run through the spreader configuration preset, guaranteeing it builds with -march=native, -O3, and Link-Time Optimization (-flto). This instructs the compiler to aggressively fuse the agent-generated strategy math directly alongside the 59ns loop logic.

5. The Cython Forge & Cryptographic Deployment

Strategy injection uses Cython rather than naive string templating (like Jinja2).

• Strict C-Types: The Foundry parses the JSON IR and transpiles it into strategy_injected.hpp wrapped in extern "C++". This guarantees the StrategyState structure uses strict C-types (double, uint32_t) and avoids 64-byte L1 cache spillage.
• Crucible Equality: It simultaneously generates a strategy_wrapper.pyx file. This allows the exact same C++ calculate() block to be invoked natively by Python during Crucible backtesting, ensuring 100% mathematical parity between the simulation and live deployment.
• Forge Signatures (.sig): When quanuxctl spreader package runs, it cryptographicly hashes the final machine-code binary (e.g., quanux-spreader-063e07ea) and outputs a .sig manifest.
• Deployment Gate: The quanuxctl spreader deploy command reads the binary hash and validates it against the .sig manifest. If an agent or human attempts to alter the executable after packaging, the deployment will fail with CRITICAL SHM ERROR: Binary checksum mismatch.

6. Trial & Error Insights (Cautionary Tales)

When generating C++ for the 59ns Spreader, agents must be aware of these strict compilation traps that were uncovered during development:

1. The -Werror Unused Variable Trap: C++ compilation is extremely strict. Handlers (like main(int argc, char** argv) or init(StrategyState& state)) that don't utilize their parameters will fail the -Werror optimization build. Fix: Always cast unused variables to void: (void)argc;.
2. The C vs C++ typedef Collision: When importing C-libraries like nats.h into modern C++ headers, do not use naive forward declarations like struct natsConnection;. The C library declares typedef struct __natsConnection natsConnection;. Fix: Use identical explicit typedef structures to avoid compiler redefinition with different types errors.
3. CMake Build Directory Pathing: Because the QuanuX root CMakeLists.txt delegates to add_subdirectory(QuanuX-Spreader/cpp), the resulting quanux_spreader binary is placed deep inside the mirrored build directory (build/QuanuX-Spreader/cpp/quanux_spreader), NOT at the root of build/. Agents scripting CLI automation must accurately resolve this nested build target path.