But if you're not minimizing the confounding factors enough that the result can be repeated, your experiment failed. Your hypothesis has neither been provably confirmed nor denied. At that point you are, at best, still gathering evidence, and publishing results would be an error.