Backtesting on bars or aggregated quotes hides the mechanics that decide real execution quality: where your order sits in the queue, how much quantity is ahead of it, and exactly which events moved the book before you traded. TickForge replays historical market activity order by order, with nanosecond timestamps and a strict determinism guarantee, so execution research is reproducible down to the individual fill.
| 📖 Market-by-order book | Full MBO book with price-time priority matching. Every order tracked individually, not aggregated into levels. |
| ⏱️ Nanosecond replay | Deterministic event replay with (timestamp, seq) ordering. Same input, same output, every time, on any machine. |
| 📊 Queue position tracking | Know exactly where a designated order sits: rank and quantity ahead at its price level, through its full lifecycle. |
| 🔢 Exact execution stats | Integer-arithmetic fill statistics. Counts, quantities, min/max and volume-weighted average prices with no float drift. |
| 🧪 Property-based testing | Deterministic fuzzer with splitmix64 RNG, independent reference model, and sequence minimizer. 10,600+ generated events, zero production defects found. |
| 📁 CSV ingestion | One documented CSV format parsed into normalized events, with derived market-state views (best bid/ask, level aggregates). |
TickForge is a layered C++20 system. The replay engine, book, and matching engine form the deterministic core. See ARCHITECTURE.md for hot/cold path design and SPEC.md for the full specification.
CSV market data → Event normalization → Deterministic replay
→ MBO order book → Price-time priority matching → Execution statistics
The determinism boundary is a hard architectural rule, not a hope:
- No wall-clock reads inside the replay/book/matching core
- No uncontrolled randomness (seeded splitmix64 where needed)
- Integer arithmetic for prices (ticks) and quantities (lots)
- Reproducibility is asserted by tests, not assumed (see
test_reproducibility.cpp)
Run the same event stream twice, on any machine, and every fill is bit-identical. The reproducibility audit documents exactly what is and isn't covered.
Requirements: CMake 3.22+, a C++20 compiler (GCC 11+ or Clang 14+), Git. GoogleTest is fetched automatically on first configure.
cmake --preset default
cmake --build --preset default
./build/tickforgeOther presets:
cmake --preset release # optimized build
cmake --preset asan # AddressSanitizer + UBSan
ctest --preset asanctest --preset default
# or, from the build directory:
ctest --output-on-failure282 tests across Debug, Release, and ASan+UBSan builds. clang-format clean, clang-tidy zero warnings. CI runs the full matrix on every push.
Per project rule, every performance claim ships with a benchmark and full methodology (dataset, hardware, compiler, events, runtime, events/sec, median and p99 latency). See BENCHMARKING.md. There are no performance claims without a published benchmark.
tickforge/
├── src/
│ ├── tickforge_event/ # Normalized event model (ticks, lots, seq)
│ ├── tickforge_book/ # Market-by-order book
│ ├── tickforge_matching/ # Price-time priority matching engine
│ ├── tickforge_replay/ # Deterministic replay driver
│ ├── tickforge_analytics/ # Queue position + execution statistics
│ └── tickforge_market_data/ # CSV parsing boundary
├── tests/
│ ├── test_reproducibility.cpp
│ └── support/ # Property-testing framework (RNG, minimizer)
├── SPEC.md # Full specification
├── ARCHITECTURE.md # Design and determinism boundary
└── BENCHMARKING.md # Benchmark methodology
See CONTRIBUTING.md. Bug reports and design discussion are welcome via GitHub issues.
See SECURITY.md for the vulnerability reporting policy.
MIT. See LICENSE.