> ## Documentation Index
> Fetch the complete documentation index at: https://docs.glitchexecutor.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Backtest

> How the historical-bars simulator works and how to read the result.

Glitch Executor's backtester replays your strategy against historical bars from `glitch-trade-core`'s bar database (currently 6 symbols × \~3 years of H1 data). Same engine the platform uses internally for the walk-forward HPO sweeps that produced the seed strategies — exposed in the UI as a one-click run.

## How it works

For each historical bar in your selected window, the simulator:

1. Updates indicators
2. Evaluates entry/exit conditions from the strategy IR
3. Walks open positions through the bar's high/low for stop/TP fills
4. Records every closed trade
5. Tracks daily-state for breach math against the assigned firm rule set

Result is a single dict with verdict, total return, max day loss, trades closed, profitable days, ending balance, HWM, and (for quick rules with no stop-loss) `pending_positions` + `unrealised_pnl` + a `warnings` array.

## Reading the result

<AccordionGroup>
  <Accordion title="Verdict">
    `PASSES` means the strategy hit the firm's profit target without breach.
    `TERMINATED (breach)` means it hit a hard rule (daily loss / trailing DD / static DD).
    `Alive — not yet at target` means it survived the window but didn't reach the target.
  </Accordion>

  <Accordion title="Total return %">
    Realised PnL as a fraction of starting balance. Includes only closed trades.
    Open positions at end-of-window aren't counted (see `unrealised_pnl` below).
  </Accordion>

  <Accordion title="Max day loss">
    The worst single-day realised PnL during the window, as a fraction of starting
    balance. Compare against the firm's daily loss limit for breach distance.
  </Accordion>

  <Accordion title="Trades closed">
    Number of round-trip trades that completed. Open positions are NOT counted here.
  </Accordion>

  <Accordion title="Pending positions / unrealised PnL">
    If your strategy has no stop-loss (most quick rules) and a position opened but
    never reached the exit, it's still open at end of period. The simulator counts
    that as "pending" and marks it to market against the last close. Hidden losses
    show up here.
  </Accordion>
</AccordionGroup>

## Honest-backtest warnings

The simulator surfaces warnings when the result is artificially clean for known reasons:

<Warning>
  This rule has no stop-loss. N positions opened during the backtest are still open at
  end of period — capital frozen until price returns to the exit. Mark-to-market
  unrealised PnL: \$X.
</Warning>

<Warning>
  0% max day loss only because the rule has no stop-loss — losing positions never close
  and don't count against the daily-loss meter. Add a stop to stress-test honestly.
</Warning>

When you see these, don't trust the verdict. Add a stop-loss, partial-TP, or time-stop and re-run.

## Walk-forward (Pro+ tier)

Pro+ unlocks **walk-forward validation**: an Optuna sweep with train/test split. Pre-empts overfitting — a config that wins on training data but breaches on holdout is exposed before you fund a challenge. The platform's seed templates were chosen this way.