daviddme sf-best-pick
The headline strategy picker. Pulls a wide HOT slice with sane quality filters, scores every candidate across seven factors, and returns the top N with a per-factor breakdown so you can explain on stream exactly why each one made the cut.
The score (each factor 0..1, then weighted)
| Factor | Weight | Floor / Target |
|---|---|---|
| Profit factor | 25% | 1.0 → 2.5 |
| Sharpe | 20% | 0.5 → 2.5 |
| Win rate | 10% | 40% → 70% |
| Net profit % | 15% | 10% → 100% |
| Drawdown | 15% | inverted: 30% (worst) → 8% (best) |
| Recency (30d) | 10% | 0% → 15% |
| Trades | 5% | 50 → 500 (sample size) |
Higher is better. A score of 0.8+ is exceptional; 0.6–0.8 is strong; below 0.5 means at least one factor is dragging hard.
When to use
- "Pick me the top 5 strategies on StrategyFactory"
- "Which strategies should I deploy this week?"
- "Find me three bangers I can run on Toobit"
How to run
python sf-best-pick/pick.py \
[--top N] # default 5
[--prescreen N] # how many HOT to score (default 100)
[--min-profit-factor X] # quality floor before scoring (default 1.3)
[--min-trades N] # sample-size floor (default 50)
[--max-drawdown %] # default 25
[--pair CSV] [--timeframe CSV]
[--explain] # multi-line "why" per pick
[--json]
Output
Table with rank, name, pair, timeframe, all the metrics, and the composite
score. --explain adds a paragraph per pick: which factors lifted it,
which dragged.
Notes
- Server-side filters are used as a quality floor; the real ranking happens client-side so the agent can adjust weights freely.
- For market-regime-aware picking, use
sf-market-fit. - For deployment after picking, use
sf-deploy.