We explore some of the similarity and differences between symbol-by-symbol
logarithmic loss, block logarithmic loss, equivocation, and list size
exponents.
We show with a few examples that - contrary to what one may suspect based on
recent results in multi-terminal source coding under logarithmic loss, and on
what seems to be tacitly assumed in some parts of the literature - the
fundamental limits may vary significantly across these performance criteria.