Dear Fanny,
My hypothesis in my last post was wrong: there is no problem for a dictionary-graph to match a sequence, and produce a lexical entry for a subsequence only.
The problem in your dictionary-graph is that the 'all matches' option does not work. I reproduced the same problem with test-Fanny-rba (attached graph): only the 3-letter matches are processed.
In order to work around this bug, you can make distinct graphs for 2-letter, 3-letter... and 5-letter abbreviations, as in test-Fanny-3-rba and test-Fanny-4-rba (attached graphs): with these dictionary-graphs, all the matches are processed.
I made my tests with version 3.2 alpha dated 27 June 2017.
The fact that the 'all matches' option does not work in spite of the 'a' in the end of the name of the graph is a bug. Can you report it as an issue in the GitHub platform?
Best,
Eric