Currently, the IL generated for operations on a bool? nb; like nb == true and nb is true is suboptimal. And since the runtime does not seem to understand the semantics of Nullable<T>, this results in suboptimal machine code being executed.

Namely:

1. The code nb == true generates branchless IL (see lifted comparison codegen can be improved #17261) equivalent to nb.GetValueOrDefault() == true & nb.HasValue, when it could be just nb.GetValueOrDefault().
2. The code nb is true generates branched IL equivalent to nb.HasValue ? nb.GetValueOrDefault() : false, when it could also be just nb.GetValueOrDefault().
3. The code nb is false generates branched IL equivalent to nb.HasValue ? !nb.GetValueOrDefault() : false when it could be the branchless !nb.GetValueOrDefault() & nb.HasValue.
4. The code nb ?? true generates branched IL equivalent to nb.HasValue ? nb.GetValueOrDefault() : true, when it could be the branchless nb.GetValueOrDefault() | !nb.HasValue.

Note that nb ?? false already produces the optimal code of nb.GetValueOrDefault().

Also note that the suggested code does not contain comparisons with true, which also result in unnecessary IL and machine code.

SharpLab link.