Stepping back from the bug list to look at what this template is actually for and how well its mechanics serve that purpose:
ROM-IT5L is built to teach vocabulary for instant-change concepts — states that transition with no meaningful intervening period, as distinct from the GT5L/GP3L families, which model gradual change over a duration. The visual language makes this distinction concrete rather than abstract: horizontal segments represent continuous, stable states; the connections between them are vertical drops, not diagonals. A diagonal line implies distance traveled, and therefore time passing — a vertical line implies none. That's a genuinely well-chosen piece of visual grammar for the concept it's representing, and it's the reason you called this one of the more universally useful templates in the family: most real-world instant changes (a light switching on, a decision being made, a door slamming) fit this shape naturally.
The learner drags the yellow circle along a five-segment path — three horizontal "state" segments joined by two vertical "instant transition" drops. Vocabulary is anchored at the midpoint of each horizontal segment, cycling TL → SL → Quiz on click, matching the project-wide convention that TL defaults so the learner has to work for meaning before SL confirms it.
Two "extension" positions exist beyond the core five — additional vocabulary tied to Line 1 and Line 3 specifically (a third, on Line 5, existed but you correctly identified it as unwanted and it's been removed). These function as a secondary layer of content without disrupting the core five-state rhythm, letting a template author add depth to a concept's states without needing to add more physical lines to the diagram.
This is the more architecturally ambitious half of the template. Rather than one linear path, the sliding text can move across three distinct paths — the main five-line route, plus two separate extension paths that branch off Lines 1 and 3. A learner drags the text along whichever path they're on, and the system tracks which path they're currently on (currentPath) to determine what vocabulary to display and where to file a correct answer once learned — into either the "Increasing" or "Decreasing" vocabulary array, depending on which direction of change that path represents.
This is a genuinely more complex data model than GT5L/GP3L ever needed, and most of what surfaced during testing traced back to that complexity: the geometric overlap between the main path and an extension path sharing the same coordinates, and the indexing collision in the vocabulary-tracking system once three separately-indexed data sources needed to feed into shared tracking arrays. Both were real architectural seams from supporting three paths rather than one, not surface-level bugs — which is exactly why they surfaced as "occasional and random" rather than as an obvious, everyday failure.
The Mode Indicator, drag-instruction panel, and vocabulary arrays now default to SL and use generic, translatable placeholder text rather than assuming any specific language pair — consistent with Conceptionary's language-agnostic design intent throughout. All three are repositionable and click-drag-droppable, matching the standard now established across the whole GT5L/GP3L/IT5L family.
Functionally, this template now does what it's designed to do cleanly: the visual metaphor is sound, the core States/Transitions mechanic works reliably, and the three-path Transitions system — while it's what generated most of the subtle bugs — is also what gives this template real flexibility that a simpler single-path design wouldn't have. The extension system in particular seems like a genuinely useful pattern: a way to attach additional vocabulary depth to specific states without redesigning the whole diagram every time a concept needs more than five items. Worth considering whether that same extension pattern is something the IT5LI (inverted) or IP3L templates should adopt too, if they don't already have an equivalent.