Run pattern discovery
5-min OHLCV → 60-bar rolling windows → normalize → encoder (128-d) → UMAP → HDBSCAN → pattern buckets → historical forward-return stats
Pattern buckets
Historical forward returns after each pattern (long from window close).
| Bucket | Count | Win rate | Avg return | Median | Std |
|---|
Analysis charts
Win rate vs avg return (bubble size = count)
Bucket size (top 15)
Bucket mix
Win rate & avg return by bucket
Forward return distribution (all windows)
Top bucket return ranges (min / median / max)
Cumulative forward return — top bucket vs all windows
Pattern bucket over time (sampled)
Forward return over time (sampled)
UMAP embedding (colored by bucket)
Mean normalized 60-bar close path per bucket (z-scored within each window). Shows the recurring candle shape HDBSCAN grouped together.
Top buckets — mean close path overlay
Bucket pattern detail
Per-bucket pattern cards (all buckets)
Discovers patterns on 4h, daily, weekly, and monthly bars, then matches normalized close-path shapes to each 5m bucket. High similarity (>0.7) suggests the same structural pattern recurs across time horizons.
Shape similarity heatmap (5m buckets × higher timeframes)
Recurring patterns (matched on 2+ scales)
| 5m bucket | Shape | Scales | 5m WR | 5m avg ret | Best matches |
|---|