How the layout solver actually works
A deterministic slicing-tree algorithm and an AI variant strategy propose schemes in parallel — with a fallback so generation never depends entirely on an external provider.
When Floor Plan Studio generates layout schemes from a room program, two independent strategies run side by side. A deterministic layout solver recursively subdivides the buildable envelope using a guillotine slicing-tree algorithm — cutting the plot into progressively smaller rectangles sized to the program's area targets, the same family of approach used in floor-planning research for decades. In parallel, an AI variant strategy proposes its own layouts, trained on the same constraints but free to explore adjacencies and proportions the deterministic solver wouldn't try.
Every candidate scheme — from either strategy — is scored on the same objectives: area efficiency (how close each room lands to its target area), adjacency (whether rooms that should be near each other actually are), and compliance (does it pass the active ruleset before you've even opened it). The ranked list you see is sorted by that combined score, not by which strategy produced it.
The deterministic solver is also the fallback: if the AI provider is unavailable or returns something unusable, you still get a full set of ranked schemes from the slicing-tree alone. Generation never hard-depends on an external AI call succeeding — a design decision we made early, since a studio mid-deadline shouldn't be blocked by a third-party API outage.