This is a custom-built, low-latency C++ replay engine and mean-reversion trading strategy built for the CSoT'26 Track. It processes the exact STRATEGY_SPEC.md Z-score algorithm with zero heap allocations on the hot path.
Measured on the 10-million row synthetic_large.csv dataset, using __rdtsc hardware counters (calibrated at 2.4956 cycles/ns) for good timing.
- p50 Latency: <= 128 ns
- p99 Latency: <= 512 ns
- Throughput: ~8.42 Million ticks/sec (via Google Benchmark)
Note: The baseline friction of the engine's for loop + virtual function call was measured at 64ns using a Null Strategy. The mathematical cost of the Z-score logic is incredibly tight.
# Compile Release Build with LTO
cmake -B build && cmake --build build -j
# Run the Replay Engine
./build/quant_runner ./build/spec_strategy.so data/synthetic_large.csv
# Run the Micro-benchmark
./build/quant_bench