[ PROOF // CLUSTER-A2 ]
The receipts, not the pitch.
Cluster-A2 is our own system — dogfood, not product. Full parameters, full numbers, full caveats. The ongoing live record is at /papertrading.
BACKTEST REFERENCE // 6 TAHUN
- Period
- 2020-08 .. 2026-08 (6 tahun, 10 pair Bitget, Cluster-A2)
- Pairs
- 10 (BTC, ETH, SOL, BNB, XRP, AVAX, LINK, DOGE, ADA, HYPE) — Bitget spot
- Entry
- Close > highest high of the previous 20 days
- Exit
- Close < lowest low of the previous 10 days, or stop loss
- Stop loss
- Entry − 2×ATR(14), on every order, no exceptions
- Position sizing
- Risk 1% modal per trade ÷ stop distance
- Direction
- Long-only
- Cluster limit
- Max 2 positions per correlation cluster (Cluster-A2)
- Costs
- Fee 0.1% + slippage asumsi 0.05%
- Sharpe
- 0.82
- Max drawdown
- -26.45%
- Total return
- +152%
- Trades
- 94 (~15.7/tahun)
- Win rate
- 36.17%
- Avg win / avg loss
- +4.35R / -0.85R
- Profit factor
- 2.27
CAVEATS // BACA DULU
- Survivorship bias: SOL/BNB/XRP and the 7 other pairs are today's survivors, selected with hindsight — the figures are an upper expectation from historical data, not a promise.
- Concentration of returns: the top-5 trades contributed ~100% of net PnL; the single largest trade (BTC Oct 2020 → Mar 2021) was ~45% of the total. If no 2020–21-style supertrend recurs, future performance can be far flatter.
- Sharpe 0.82 from 94 trades has a wide confidence interval (reality could be ±0.3). The edge is cutting losses fast and letting winners run — not metric precision.
- The simulated 5% APY on paper idle cash adds return without changing the strategy — platform risk is not modeled.
- Live vs backtest comparison is only valid after ≥10 closed trades (Fase 2 criterion). Before that: monitor operations, not numbers.
REPRODUCE IT
All scripts, reports, and the decision log are open on GitHub. Don't trust the numbers — rerun them yourself.
decision_log.md →