These prototypes make a task's objectives explicit: its goal, success criteria, constraints, context, and assumptions.
The work explores how those representations can be used in neural training, across multiple tasks, and in text-based systems. Related prototypes attempt to identify when an objective leaves out part of a goal or conflicts with another requirement.
The central question is whether making the objective more explicit leads to a useful change in behavior. Representing a goal as structured data does not guarantee that a system understands it or can judge its own success.
The current work consists of exploratory implementations and notes. It needs concrete comparisons on defined tasks before broader conclusions are warranted.