4320820ae2
Second half of the Wave 3 val-cost-streams refactor (3a kernel-side
foundation landed at e968f4ded). Atomically migrates the
GpuBacktestEvaluator::new contract; 3 call sites + 6 test call sites
+ 5 WindowMetrics fields + 11 new buffers + launch sequence wiring
all in this commit.
Constructor signature change: GpuBacktestEvaluator::new gains
window_lob_bars: &[Vec<LobBar>] parameter alongside existing
window_prices + window_features. Three production call sites migrated
atomically:
- trainers/dqn/trainer/metrics.rs:651 (val_evaluator construction)
- trainers/dqn/trainer/metrics.rs:1166 (extra_eval Dev/Test)
- hyperopt/adapters/dqn.rs:1493 (full LobBar with real OFI)
- hyperopt/adapters/ppo.rs:1364 (zero-OFI LobBar — PPO lacks per-bar
OFI features; cost-net OFI-impact term degrades to 0; commission +
half-spread × position still apply)
11 new mapped-pinned buffers on GpuBacktestEvaluator:
Input (3): close_prices_buf, half_spread_buf, ofi_scalar_buf
Derivation (3): position_history_buf, side_ind_buf, rt_ind_buf
Output (5): cost_net_sharpe_buf, baseline_{buyhold,hold_only,
momentum,reversion}_sharpe_buf
5 new WindowMetrics fields (host-annualised via × annualization_factor):
- sharpe_cost_net (1.2.b cost-net sharpe)
- baseline_{buyhold,hold_only,momentum,reversion}_sharpe (1.4.b)
Per-window eval flow now: existing fused metrics kernel → 1.1.b sharpe →
position_history_derivation (one launch over all windows) →
cost_net_sharpe (per-window) → 4 × baseline_* (per-window).
commission_per_rt: host-computed constant per D3 resolution, formula
config.tx_cost_bps × 0.0001 × config.initial_capital, passed by-value
at each baseline + cost_net launcher invocation.
Wave 3a kernel signature follow-up: cost_net_sharpe_kernel side_ind /
rt_ind switched from unsigned int* → float* so the cost-net kernel
chains directly with the f32 streams emitted by
position_history_derivation_kernel (no u32→f32 adapter buffer; bit-pun
mismatch fixed). cost_net oracle test migrated MappedU32Buffer →
MappedF32Buffer accordingly.
Atomic per feedback_no_partial_refactor: constructor sig change + 3
production + 6 test call site migrations + 5 orphan launchers
eliminated + WindowMetrics field additions + cost_net kernel sig fix
+ audit doc all in this commit.
Closes 1.2.b + 1.4.b + position_history_derivation orphan launchers
per feedback_wire_everything_up.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Foxhunt
Production HFT trading system in Rust.
Architecture
The workspace contains 32 crates organized as follows:
Core Libraries (16)
| Crate | Purpose |
|---|---|
trading_engine |
Order processing, FIX 4.4, IB TWS, SIMD, RDTSC timing |
risk |
VaR, Kelly, circuit breakers, kill switches, compliance |
risk-data |
Risk data types and shared structures |
trading-data |
Trading data types |
ml |
DQN Rainbow, PPO, TFT, Mamba2, ensemble inference |
ml-data |
ML data types and feature definitions |
data |
Market data ingestion and storage |
backtesting |
Replay engine, strategy tester |
adaptive-strategy |
Ensemble execution, microstructure analysis |
common |
Shared types, resilience, error handling |
storage |
S3 and local model storage |
model_loader |
Model serialization and loading |
market-data |
Market data feed handlers |
database |
PostgreSQL access layer (SQLx) |
config |
Configuration management |
tli |
CLI commands and tooling |
Services (8)
| Service | Purpose |
|---|---|
backtesting_service |
gRPC backtesting service |
broker_gateway_service |
FIX routing, broker connectivity |
trading_service |
Core trading operations |
ml_training_service |
Model training orchestration |
data_acquisition_service |
Market data acquisition |
trading_agent_service |
Autonomous trading agents |
api_gateway |
gRPC API gateway with auth |
web-gateway |
Axum REST + WebSocket gateway |
Frontend
web-dashboard/ -- React 19 + TypeScript + Vite + TradingView charts.
Building
# Check compilation (no PostgreSQL required)
SQLX_OFFLINE=true cargo check --workspace
# Run tests for a specific crate
SQLX_OFFLINE=true cargo test -p <crate> --lib
# Clippy
SQLX_OFFLINE=true cargo clippy --workspace
ML Models
Four production model architectures on Candle v0.9.1 with CUDA:
- DQN Rainbow -- Deep Q-Network with prioritized replay, dueling heads, noisy nets
- PPO -- Proximal Policy Optimization with GAE and LSTM policies
- TFT -- Temporal Fusion Transformer for multi-horizon forecasting
- Mamba2 -- State space model for sequence prediction
Each model has a standalone trainer and a UnifiedTrainable adapter for the hyperopt pipeline.
Infrastructure
- Git: Gitea at
git.fxhnt.ai(Tailscale-only), Scaleway DEV1-S - Observability: OpenTelemetry OTLP (env
OTEL_EXPORTER_OTLP_ENDPOINT) - Database: PostgreSQL with SQLx offline mode for CI
License
Proprietary. All rights reserved.