Smart Gold Impulse

• Monte Carlo resamples the SIMULATED trades of this backtest. It inherits every weakness of the simulation (in-sample bias, regime luck, execution gap) and cannot add information -- it only reveals how fragile the observed path is.
• Same trades, different order: answers only how path-dependent the drawdown is. Treats trades as exchangeable -- grid and martingale sequences are not (their order is caused by the market path), so read this as a diagnostic, not a forecast.
• Draws trades independently with replacement (IID assumption). Real EAs violate IID through clustering and position chains; intervals are tighter than reality for such systems.
• Resamples 5-trading-day blocks of the end-of-day PnL series (circular: blocks wrap at the series end, so every day carries equal weight -- without wrapping, a crash on the final day would be undersampled): preserves short-range clustering and losing streaks up to the block length; longer regime shifts remain unmodeled.
• Floor breaches only, on block-bootstrap paths over a horizon equal to the tested history, profits above start swept (same convention as the prop-fit sweep). The firms' full rule sets (daily loss, consistency, position loss) are stricter -- treat every probability as a lower bound, and the safe-sizing answer as an upper bound on the truly safe size.
• The fan bands are pointwise percentiles across resampled paths -- the envelope is not a single achievable path. Paths use the tested sizing and are not stopped at account death.
• Challenge-pass probabilities run each resampled path until profit target, max-loss floor or a daily-loss day is hit (checked in that day's order: daily first, then floor). Where the audit has reconstructed equity, daily-loss and floor checks use the intraday equity drop below the day anchor (sampled at deal timestamps, server-day boundary) -- still a lower bound on breaches, but far closer to the firms' equity-based rules than EOD deltas. Phases are treated as independent resamples for the combined figure; the sizing curve keeps the targets fixed while scaling the trading, which is why less risk can also mean more undecided paths.

• Withdrawals are simulated monthly on the last trading day: everything above the start balance is taken out. A month without surplus pays nothing -- the dry-streak figure shows how long that can last. Paths are the OBSERVED backtest days; the withdrawal distribution adds resampled paths (same circular block bootstrap as the MC section, own seed).
• balance_scaled multiplies each day's PnL by balance/start -- an approximation: min-lot granularity and balance-coupled EA defaults do not scale linearly. Direction: optimistic for small accounts.
• Capital what-if keeps the observed USD path (fixed lots) and re-expresses risk against each capital. Margin requirements are NOT modeled -- a small account may be unable to hold the positions at all. Direction: optimistic. Under balance_scaled sizing, percent metrics are unchanged by construction; only the min-lot limit differs, which is why no table is shown.

A good backtest cannot prove an edge. This audit dissects the simulation you gave it. It can expose structural risks (martingale, grids, hidden concentration, cost fragility, rule conflicts), but it cannot tell you that the strategy will make money. Three limits are fundamental:

• In-sample bias: commercial EAs are typically released and updated after most of the tested window. A strong report may replay what the vendor optimized against, not what the market will do.
• Regime dependence: a profitable window can reward almost any mechanic aligned with the prevailing trend. Year slices and long/short splits above hint at this, but cannot settle it.
• Simulation gap: even with real ticks and execution delay, the tester does not model live spreads, slippage beyond the next tick, or broker-specific behavior.

This audit is software analysis, not investment advice, and contains no recommendation to buy or trade anything.