A layered architecture that replaces browsing with structured elimination.
Most systems expose the full option set and expect the user to navigate it. GDL takes the opposite approach: reduce the decision space before the options are ever shown, using a small number of high-signal questions that each materially narrow what remains possible.
Translates ambiguous, subjective input — proxy language, mood, half-formed preference — into structured signal before any narrowing begins.
The core mechanism. Each question must materially change the outcome set — a question that only collects preference without narrowing options doesn't belong in the flow.
Converts a narrowed decision state into a small, concrete set of real options — never a single opaque answer, never an unfiltered list.
Checks that a recommendation actually holds up in the user's real-world constraints before it's delivered as final.
In practice
A user describes what they want in loose, subjective terms — not a spec, a feeling. Three to five questions later, the system has eliminated the vast majority of an option set the user never had to see, and arrives at a small number of choices with a clear reason for each. The user never browses. They arrive.
Additional layers handle discovery from partial memory, and mode-switching when the interaction's underlying conditions change mid-flow.