Dear Liam,
There were several factors that led us to perform the manual evaluation only on the agent side (English-German). The customer source text was originally in English and was translated into German. We realised that showing the noisy English text to the evaluators as the reference could be misleading for them impacting their evaluations.
Moreover, in this setting it is usually fine for the agent to have a gist of the message; however there is extra responsibility in translating the agent since it has the company brand with it.
Therefore, we preferred to focus more on evaluating the agent translations more rigorously than to spend resources in evaluating the customer.
Hope this answers your question.
Best,
Amin