@ItsDevMutai@Noise__Makr if you're pulling bars with copy_rates_from_pos, watch out that index -1 is the bar still forming, not the last closed one. an H1 engulfing can show up mid-hour and be gone by the close. dropping the last row before the pattern check keeps live agreeing with your backtest.
@_codewithemmy one thing to check before you trust the backtest numbers: if the signal is decided on a bar's close and you fill on that same close, you're reading one bar ahead. i ran an MA cross on random walk data and got +23.6%. shift(1) so it fills next bar and wins went 91/100 to 50/100.
@SystematicEdge1 one that bites on live history rather than tester reports: P&L sits on the closing deal, the opening one is 0, and commission/swap land as their own rows. partial closes split a position into several deals, so group by position_id or per-trade stats go wrong.
@kieran__duff the thing that gets you with an agent in the loop is it'll happily re-run the same history until the numbers look nice. by iteration 40 you've fit the data and nothing in the output says so. surfacing how often a strategy touched the same window would help a lot.
@EthersetHQ_ cron's gotcha is overlap. if a run outlives the interval the next one starts anyway, and two processes both verify against the same balance before placing. a lockfile around the whole run is boring but it's what stops the double fill.