Currently, the adder element clamps sample values of floating point formats to the range [-1.0; 1.0] (-0dB).  This is incorrect: It might be perfectly possible to handle values exceeding -0dB in the pipeline.  Most probably by using a compressor/limiter element of choice, which would give a more desired result than just clipping, as it is currently done (implicitely).  If application developers do indeed just want to let it clip, the audioconvert element will just do it as part of the conversion to an integer format.  In no way should the adder element do the clipping on its own.