41292303dc
3-fold A/B sweep 2026-05-18 at commit 83546b5c3 falsified the simple
per-horizon Q_h attention pool:
mean_auc 0.7559 ± 0.0068 vs baseline 0.7749 ± 0.024 (Δ = -0.019)
h6000 0.7588 ± 0.0049 vs baseline 0.7591 ± 0.018 (Δ = -0.0003)
best_epoch on val_loss = 1 in 2/3 folds → calibration drift as α opens;
no horizon-distribution shift toward h6000. The per-horizon path
spends capacity on directions that hurt log-likelihood without lifting
ranking quality.
V1 of the v2 redesign deletes the falsified path entirely (per
feedback_no_partial_refactor; v2 spec/plan committed earlier today
captures the migration). Files removed:
cuda/per_horizon_attention_pool.cu
cuda/per_horizon_residual_head.cu
cuda/per_horizon_prob_blend.cu
src/per_horizon_attention_pool.rs
src/per_horizon_residual_head.rs
src/trainer/per_horizon_state.rs
tests/per_horizon_attention_pool_numgrad.rs
tests/per_horizon_residual_head_numgrad.rs
tests/per_horizon_full_pipeline_smoke.rs
perception.rs:
- struct field `per_horizon` removed
- new() initialization removed
- step_batched section 4.5 (forward_with_blend) → reserved comment
- step_batched section 5a (backward_through_blend) → reserved comment
- step_batched section 9 (adamw_step) → reserved comment
- existing 17 optimizer groups + BCE/attention-pool path untouched
- reduce_axis0 kernel kept (still used by existing param-grad reducers)
build.rs KERNELS: dropped the 3 per_horizon entries.
lib.rs + trainer/mod.rs: dropped per_horizon module declarations.
Workspace compiles clean (cargo check -p ml-alpha). Next: V2 builds
the horizon_token_attention_pool kernel as the v2 replacement.
Co-Authored-By: Claude Opus 4.7 <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.