4a4953b01f
alpha-rl-zf6s5 (38a4aa15b b=1024 fold-1 step 5719 verdict, 2026-06-02):
the v5 scaffold_weight stayed pinned at 1.0 even when wr crossed
break-even (0.306 > 0.30). Root cause: `w_decay = min(1, 2 × frac_alerted)`
saturated at 1.0 because train-time Page-Hinkley naturally alerts on
75-95% of batches (per pearl_edge_decay_detector_phase1_validated_train_also_decays
discovered 2026-06-01); `max(w_competence, w_decay)` then kept the
scaffold engaged forever, defeating the fade mechanism.
FIX: `w_decay = max(0, 2 × frac_alerted − 1)` so re-engagement measures
EXCESS alertedness above 50% baseline, not absolute level:
frac_alerted v5 w_decay v5.1 w_decay
----------- ----------- ------------
0.50 1.0 0.00 ← v5 broken here
0.78 1.0 0.56
0.90 1.0 0.80
1.00 1.0 1.00
At zf6s5 step 5719 (frac_alerted=0.78), v5.1 gives w_decay=0.56 not 1.0;
w_competence=0.455 wins via max(), so w drops to ~0.56 letting the fade
begin properly.
Health signals from zf6s5 (informative even though terminated):
step 5719: wr=0.306, hold=14.4, entropy=1.82, realized=+$152M
trajectory: wr crossing 0.30 ✓, hold growing 12→14 ✓, entropy ↓ ✓
the agent IS becoming competent; the scaffold just couldn't fade.
Local invariant test:
reward_alignment_surfer_scaffold_invariants OK: 251 rows validated
train[0] bootstrap = 0.966 (unchanged — boot path identical)
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.