Maybe now that opaque is implemented in Agda 2.6.4 we should revisit our solution to #1753.
This is exactly the situation that the new opaque mechanism was designed for. What are people's thoughts? Too risky to adopt a cutting edge feature? In v3.0 I would imagine we will be making a lot of definitions opaque in the library?