How the Optimizer Works
Every ~22 hours, Amyth's backtest engine runs thousands of simulations to find the best parameters for your strategy. Here's exactly what happens under the hood.
The 22-hour cycle
Pull real candle data from Birdeye for BTC, ETH, and SOL across all supported cadences (5m, 15m, 30m, 1h, 2h).
Sample hundreds of parameter combinations across the full schema using Latin Hypercube Sampling, a smarter version of random search that covers the space more evenly. Ranked by Calmar (return ÷ max drawdown).
Take the top 10 candidates from Stage A and refine each with coordinate descent, tweaking one parameter at a time to find the local Calmar optimum.
Build a tight bounding box around the elite Stage-B winners and resample inside it, now ranking by raw return to maximize PnL within the proven-good region.
Coordinate descent again on the Stage-C winners, polishing parameters for maximum return.
Re-score the final candidates on the full window for an honest, unbiased PnL number. Quality gates run here before any winner ships.
Deliver the validated parameters to all running bots. Your vault picks up the new DNA on its next cron tick.
What the optimizer can and can't change
✓ Optimizer tunes these
- Entry thresholds (z-score, breakout period, etc.)
- Stop-loss distances (ATR multiples)
- Take-profit targets
- Indicator periods (moving average lengths, etc.)
- Conviction gates (minimum signal strength)
- Hold period limits
✗ These stay locked
- Your max leverage cap
- Your chosen strategy type
- Your selected assets (BTC/ETH/SOL)
- Your drawdown limit
- The cadence (evaluation frequency)
- Deposit and withdrawal permissions
Quality gates
Not every backtest result gets shipped. Each winning parameter set must clear these gates:
Every optimizer run is logged. You can see when your bot last received an optimizer update in the activity log. The backtest results that powered the update are visible in the app's strategy carousel, showing recent window performance for each preset×cadence cell.