goal: either web-based or lui-based UI where user does:
	1. initiate a session by picking a sentence from a tsdb profile
	2. that sentence is parsed by the backend, and the backend records the packed forest and stays open with that forest in memory
	3. the user is presented with a display of the sentence, and can mak decisions
	4. record decisions to the tsdb profile (or a copy of it)

improvements:
	- option to show *all* remaining discriminants (or, say, up to 20)
	- (option to?) show lexical types instead of lexeme names
	- item id display -- part of a prev/next/save interface
	- display parse tree


recording decisions and preferences?
	- first step: read existing preferences file
		- online display of whether preferred tree is currently in or out
	- then: read existing decisions file
		- mode/button to apply existing decisions and see how far that goes towards disambiguating
		- 'decisions' file format:
			d-state
				1 = explicitly require this labeling
				2 = explicitly ban this labeling
				3 = implicitly require this labeling
				4 = implicitly ban this labeling
					... itsdb code that generates it:
						1: 'toggle' == t
						2: 'toggle' == null
						3: 'toggle' is neither and 'state' is t
						4; 'toggle' is neither and 'state' is null
				... we may as well only look at types 1 and 2 for now.
			d-type
				2 = type	(apparently the lexical type of a lexeme)
				3 = constituent	(apparently a rule dominating that constituent)
			d-key
				symbolic label for the span... tends to be all caps :-/
			d-value
				sentence text with a || showing constituent break
			d-start and  d-end
				appear to match our token-based start/end positions
	- then: button to write a new decisions/preferences file pair