For trading bot developers · DEV0 → BT99
Your code runs. Does its backtest tell the truth?
Send your trading-bot repository, notebook or code excerpt. I check for one concrete bug within 24 hours, free. You get the file and line, why it changes the backtest, and the fix. No finding at the first look? I say so.
Six checks
What can flatter your backtest?
Look-ahead
A signal, fill or sort uses information from later in the test.
Missing costs
Spread and commission are absent from the result.
Impossible fills
Your code fills an order at a price the market never gave.
Gaps and time zones
Missing bars or clock differences change which data the rule sees.
Indicator start-up
Warm-up or recursive state changes a signal.
Platform parity
Your backtest behaves differently from MT5 or the exchange.
Our own case · HP36-02
A day-level stop guard missed an order race
Two same-day entries after a stop invalidated our first multi-symbol EA run. I discarded that result and added a history check directly before each order. These are excerpts from the actual EA, not a made-up before-and-after return chart.
Before: the outer retry guard alone
if(InpBlockStopUntilNextDay && g_stopped_day_bar == g_day_bar) return; // Outer retry guard; not enough alone.
After: recheck before each order
bool OpenSymbol(const int index)
{
if(index < 0 || index >= g_symbol_count)
return false;
if(StopOccurredToday())
return true;
// Select symbol and place order…
}The corrected guard reads today’s deal history for a stop exit. The outer guard remains as well. A second defect recalculated direction after partial closes. Direction is now frozen for the agreed holding block. Fixing mechanics is not evidence that a strategy makes money.
Read the exact source excerpts and provenanceBacktest Bug Hunt · BT99 · €99
One Python, JavaScript, MQL5 or Pine trading-bot repository, up to about 2,000 lines. All six classes checked. Each finding comes with its file and line, a failing case and the fix. If the code runs, I rerun its headline result with the fixes.
Delivery: 72 hours after code access and funding. Read-only source access, never live keys or accounts.
Kein Fund, keine Kosten. No finding in any of the six classes, no charge.
Send your code for a free first check
One finding, with the line and fix, within 24 hours. If the first look finds none, I tell you.
For freelancers and implementation partners
Have a client’s trading-bot case outside your scope? Send the case with the client’s permission. A first implementation can use OP-FIRST under the catalogue’s first-order terms. Referrals can use our standard partner commission: 30% in the first three months after agreement, then 20%, on released net service revenue for the first 90 days per introduced buyer. No exclusivity or upfront fee. Marketplace clients stay on their marketplace.
Partner options