That is exactly why I treat the micro-lot run as a process test, not a signal test. At tiny size the dollar result matters less than whether the live spread and fills stay close to the assumptions, whether I follow the planned stop, and whether the drawdown behaves anything like the test. If I find myself wanting to intervene when the loss is only a few dollars, that is useful information to have before scaling.
I would log rule adherence and execution costs alongside P&L, then keep the size fixed for a pre-set sample or several weeks rather than increasing after one good patch. If the locked-out data and the micro run agree after spread and slippage, I would scale in small steps; if the fills or behavior diverge, pause and investigate instead of explaining it away. The boring part is what separates discomfort with small losses from evidence that the strategy itself has changed.