Above the root letter of a Tibetan syllable there may be a superscript, the mgo can or head letter, and three letters may take that position: ra, la and sa. No others.
All three are silent in modern Lhasa speech. What they do instead is set the tone: a syllable with a superscript is spoken high, even where the root letter would otherwise be low, so the head letter has become a pitch mark on a syllable rather than a consonant in it. In the classical language it is understood to have been pronounced, as part of an initial cluster, and clusters of this kind survive in Amdo speech today.
Two of the three change shape in the position. Superscript ra is written in a shortened form, and superscript la and sa are adapted to sit above the letter they govern. The stack is compact by design, because it has to be read as one unit and fit between two tsheg dots.
Which roots take which superscript is a property of the word and is learnt with it, not derived from a rule. That is the same situation as with the prefixes, and the two together are the reason a Tibetan word cannot be spelled from its pronunciation even by a native speaker who reads fluently. Spelling has to be known, and Tibetan schooling accordingly gives it a great deal of attention.