Toss out or Salvage data: a Modified Randomization Dilemna

We have an agricultural experiment at 24 sites across Europe to test a treatments effect on yield. The protocol and model stipulate a RCBD. One of the sites modified the RCBD and for ease of application created a block design with a pattern inside the blocks instead of completely randomized. There is no reason to believe the pattern has any bias that will affect the treatment effect.
But since this one site deviates from the rest there is concern on the team that this compromises the across site analysis. I am trying to identify a way to salvage what I expect is still relevant w low risk of biased data that could enhance the data set without degrading the integrity of an analysis that includes the site too much.
There is a risk to including the data, but there is also a cost to not including the data. Any thoughts on weighing the pros and cons or Stat methods to mitigate the deviation from RCBD for the one trial?