8d088cbc38
IQN and CQL SAXPY contributions into grad_buf are now scaled by adaptive
budget factors derived from learning_health (ISV[12]) and regime_stability
(ISV[11]):
iqn_budget = 0.10 + 0.30 × health (0.10..0.40)
cql_budget = 0.10 × (1−regime) × health (0 at collapse, volatile+healthy → 0.10)
ens_budget = 0.05 (constant)
c51_budget = 1 − iqn − cql − ens (C51 absorbs headroom at collapse → 0.85)
At collapse (health=0): IQN backed off to 10%, CQL off, C51 takes 85% —
stable directional learning when distributional components are unreliable.
Changes:
- GpuDqnTrainer: add 4 last_*_budget_eff fields (initialized to health=1 defaults)
- GpuDqnTrainer: add read_isv_health_and_regime() public accessor
- apply_iqn_trunk_gradient(): add iqn_budget param; scale = iqn_lambda × readiness × iqn_budget
- apply_cql_saxpy(): add cql_budget param; SAXPY alpha = cql_budget (was 1.0)
- FusedTrainingCtx: add compute_adaptive_budgets() — reads ISV, computes all 4, caches to trainer
- FusedTrainingCtx: add last_{iqn,cql,c51,ens}_budget_eff() accessors
- submit_aux_ops(): call compute_adaptive_budgets() once per step, thread to IQN/CQL sites
- training_loop.rs: B4/G5 propagation block for HEALTH_DIAG logging
Co-Authored-By: Claude Sonnet 4.6 <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.