Most bots run one strategy in every market and bleed when the regime turns. Adaptive Regime Switch Pro classifies the market every hour — trending, ranging, or risk-off — and switches its posture to match: it presses confirmed trends, stands aside in chop, and defends capital when volatility goes vertical. Traded on BTC & ETH USDT-margined perpetuals.
01
Performance vs. buy & hold
Full out-of-fee, no-look-ahead backtest. Benchmark = 50/50 BTC+ETH hold.
Equity curve — strategy vs. 50/50 hold · shading = perceived regime
Trend-upTrend-downRange — stand asideRisk-off — defend
Underwater — drawdown from peak
Walk-forward validation · no out-of-sample re-tuning
Logic fixed on the first 60% of history, then run untouched on the unseen last 40% — which includes the difficult 2026 bear-chop. Shown honestly, not hidden.
02
Does the “switch” earn its keep?
Same engine, components switched off one at a time — the evidence behind every design choice, including the ones we rejected.
Where the P&L came from · net by entry regime
Per-year breakdown
03
AI co-pilot in the loop
An LLM reviews every entry the rules engine proposes, using only point-in-time data, and can veto low-quality breakouts. Deterministic and cached, so the run is reproducible.
What the co-pilot did
Rules vs. Rules + AI co-pilot
The AI gate is applied through the same engine: vetoed entries are skipped, everything else is unchanged. This is the honest measure of whether the AI layer helps.
Live decision log · the agent's most recent entries, with the co-pilot's read
04
Robustness — Monte Carlo
Skill or one lucky path? We resampled the trade sequence 2,000× to map the distribution of outcomes.
Distribution of final return · 2,000 bootstrapped histories
What the distribution says
5th / 50th / 95th percentile of final return, the share of histories that ended in profit, and the 95th-percentile drawdown. A right-shifted, mostly-positive distribution is the signature of a real edge — not a single fortunate run.
Monthly returns · net strategy return by calendar month
05
Regime map & how it thinks
Price with the perceived regime shaded behind it, and the perceive → classify → act → defend loop.
Long entryShort entryExit· shading = regime (see equity legend)
1
Perceive
Measures trend strength (ADX), the volatility regime (ATR percentile over 30 days), the macro trend (EMA-200) and Donchian breakout channels on BTC & ETH perps.
2
Classify
Labels the hour Trend, Range or Risk-off — with persistence + a post-stop cooldown so it doesn't flip on noise.
3
Act
In a confirmed trend it takes the breakout (cleared by ≥· ATR, ADX ≥ ·), sized inversely to volatility. In a range it holds fire — chop has no edge.
4
Defend
Hard ATR stop on every trade, a Donchian trailing exit to bank trends, a risk-off rail that stops initiating into chaos, and a circuit breaker that halts after a ·% drawdown.