branch main commit 4fdecb8 built last · feat(ideas): replace git-based ideas with Cloudflare KV backend

Queued 7

Real-time regime detector for V66/MACD_div allocation
R121 P0 M
R120 PROVED regime-aware portfolio beats V66 5× on compound + matches min α. Need a classifier that detects 'recovery regime' (where MACD_div fails) in REAL TIME to deploy this.
Relaxed event filter sweep (notebook ready, GPU)
R122 P1 L
Tradeoff curve: how loose can the event filter go before min α collapses? 10 configs sweep, ~17 GPU-h.
Multi-horizon GRU ensemble (5min + 15min + 1h)
R123 P1 L
Each horizon sees different patterns; union of trade calendars → 2.5-3× V66 trades. Heavy GPU (12+ hours A100).
Phase 0 validation infrastructure (src/validation/)
R124 P0 L
Without it, no candidate can be safely promoted — blocks the entire promotion pipeline
RL agent (PPO/DQN) — Stable-baselines3
R125 P2 L
Unexplored paradigm — no R&D rounds yet, completely orthogonal to GRU
Sentiment / LLM features from crypto news
R126 P2 M · idea 002
Idea 002 — non-technical alpha source, complementary to GRU's price-only features
Transformer / TFT architecture
R127 P2 L
Last-tier exploration — only if all simpler paths plateau

In-flight 0

No experiments in flight. R120 ready to start.

Recently done 5

Portfolio V66 + MACD_div (regime-aware)
R120 ITERATE 2026-05-22
7 configs PASS — top: 362,991% compound (5.1× V66), min α matches 245, 16× more trades. Requires real-time regime detector (→ R121).
Multi-timeframe filter (4h context)
R119 REJECT 2026-05-21
All 5 filters degrade V66; best (no_extreme_RSI) +168 α vs +245 baseline
XGBoost alternative model
R118 REJECT 2026-05-21
+245% compound vs +70,576% V66 — trees crushed by GRU 287×; 0/4 beats B&H
Mean reversion strategies portfolio
R117 ITERATE 2026-05-21
MACD_div +179K% compound BUT min α -42 — became R120 portfolio candidate
Position sizing (Kelly + variants)
R116 REJECT 2026-05-20
All 9 sizing configs lose compound; 100% per trade confirmed optimal