TL;DR — Brutally Honest
Ran LIVE on Binance Futures from April 2026. 90-day net: -$169 — and ~$113 of that was pure fees, not bad calls. So the wallet's back at $0 while we fix exactly that, and signals run on paper for now. Here's everything — numbers, bugs, what blew up, what survived. The red included.
Live Account Proof
Real Binance Futures. Real capital, from April 2026. Real positions — and a real loss we're not hiding. Zero backtests paraded as results. The live number is always at /api/stats.json.
The 90-day run, in numbers — live figure always at /api/stats.json
Source: Binance Futures /fapi/v1/income API (trailing 90-day window). Real exchange data, not state.cumPnL. This is the number that made us pull the wallet and rebuild.
Wallet
$0
funds pulled to fix fees
Net PnL (90d)
-$169
fees included
Commission
-$113
~2/3 of the loss — the real story
Trading loss
-$51
the calls alone
Funding
-$5
90-day
Current mode
PAPER
rebuilding the edge
Strategies in arena
3,400+
Arena WR (paper)
~53%
Live roster
10 champions
Live figures
→ /api/stats.json
What went wrong
Fees ate 78% of gross profit
On a small wallet trading many small positions, Binance taker fees (0.05% per side) compounded fast. Over the 90-day window commission alone hit ~$113 — more than double the -$51 trading loss — which is what dragged the net to -$169. We caught it in a deep fees audit. The fix isn't fancier order routing (the signal crosses the spread, so it's taker by nature) — it's trading far less, only when the edge clears the fees. Fee drag is the single biggest lever on net performance.
Multi-entry phantom losses (Pitfall #187)
When the state held multiple DCA entries for the same pair, reconciliation compared Binance's aggregate position against each entry individually — creating ghost losses that triggered infinite re-arm loops (~67 per day at peak). Fixed. State cumPnL was also corrupted, which is why we always use the Binance income API as source of truth.
Stop-loss substring bug silently killed trailing stops (#200/201)
A substring match "STOP" in order_type matched both STOP_MARKET and TRAILING_STOP_MARKET. Result: the trailing ratchet (Moonshot runner) was being silently cancelled on every BE-lock and trail tighten. Caught via a systematic audit, fixed with strict equality checks.
89.7% trail-kill signature in 30-day backtest replay
Counterfactual replay of 30 days / 1049 trades: 157 of 175 positions with trailing armed never had the trailing stop fire. That's structural proof of the #201 bug. Post-fix, the expected EV on runners recovered. Projected value unlocked: ~$10/week at current volume.
What's working
10-champion live roster, forward-selected only
The live roster promotes strategies purely on arena forward performance — never backtest. A strategy needs ≥15 arena trades, ≥25% WR, positive PnL, and PF ≥ 1.2 before touching the live roster. Some champions post 80%+ WR in recent arena sessions — but that's arena (paper): those numbers structurally run ahead of the live account (survivorship, no compounded fees). The live ledger is the humbler -$169 above. We show both.
53% arena win rate across 3,400+ competing strategies
The paper arena runs every 2 minutes, rotating through 80+ liquid pairs. 3,400+ strategies compete. The 53% WR at arena level, filtered down to the top 10 for live trading, creates the actual signal quality — not simulated data, not cherry-picked windows.
Multi-regime classification blocks hostile entries
Seven regimes: strong_bull → crash. Sizing adjusts 0.40× in crash to 1.20× in strong_bull. More importantly, hostile regime blocks new entries entirely — which likely saved the account from the crypto drawdowns in March–April 2026. The classifier has been running live since day one.
Hardware stop-losses on Binance (not software)
Every live position has a hardware STOP_MARKET order placed directly on Binance exchange. This order cannot be removed by the bot software alone — it persists even if the bot goes offline. A separate watchdog cron checks for naked positions every 5 minutes and re-arms any that are missing.
How to verify yourself
1. Hit the live API directly
curl https://darwintrade.com/api/stats.json | python3 -m json.toolReturns live wallet, trade count, WR, and net PnL — sourced from Binance income API, updated hourly.
2. Watch the FREE Telegram channel live
Join t.me/DarwinLabSignals. Every signal includes pair, direction, entry, SL, and TP prices — timestamped. Cross-check any entry against Binance Futures klines at that timestamp. The signal and the live order are generated by the same engine.
Why we show this
Most "crypto signal" services show cherry-picked Photoshop screenshots and Excel P&L sheets you could fabricate in 10 minutes. We ran the bot on our own money on a regulated exchange with public order-book access — and when it lost, we published the loss instead of deleting it. The wallet is at zero now while we fix the fee-drag, and the signals you see are on paper, timestamped and non-repainting, generated by the same engine. When the ledger turns green again, you'll have watched it happen in the open.
What you're looking at
- Account: Binance Futures USDT-M, isolated margin, hardware SL orders
- Engine: 3,400+ strategies evolved H24 by a genetic algorithm, filtered by a 48-feature XGBoost confidence gate (92% accuracy)
- Risk: Kelly ÷ 4 sizing, max 10 positions, regime-aware scaling (0.40× crash → 1.20× strong_bull)
What this is not
We're not licensed financial advisors. Crypto futures are high-risk. Signals can be wrong. Drawdowns happen — all historical losses are visible in the Kill Feed and Dashboard. Trade responsibly, size conservatively, never risk what you can't afford to lose.
Ready to receive the same signals?
FREE signals live on Telegram (graded ≥ 6.5/10). VIP gets live-only priority signals, DCA entries, and regime-dedicated analysis.
Past performance does not guarantee future results. Crypto futures trading involves substantial risk of loss. All stats sourced from Binance Futures /fapi/v1/income API. Darwin Lab is not a licensed financial advisor. Trade at your own risk.