Room to arrive and rest

Spatial refinement, 15 October 2026 Zurich. A free navigation center was not enough for the visible bench: the earlier proposed point beside the well would still intersect it. That point-only source was never deployed. The replacement requires a 3×3 neighborhood free of static obstacles and water, plus two metres from other furniture centers (three from beds). Migration, new bench construction, use and menu validation share that rule. Beds retain their separate existing center-only navigation limitation. Future visits choose validated endpoints within three metres of the landmark and at least 1.5 metres from other outdoor residents and active visit endpoints. Saved paths retain the reservation. The last step is rechecked; an occupied endpoint causes a real route adjustment, and a full destination leaves an existing journey visibly waiting without another camera or model call. New full-destination requests reject atomically. Private-room actors are excluded. Already completed idle visits are preserved, with no automatic forced walk. This checks destination spacing, not collision avoidance along every intermediate movement. Measured snapshot: 377 Python tests passed in 18.62 seconds, including 19 additional cases since the 358-test point-only source. Sources are private arrival-clearance-tests.xml and frozen arrival-clearance-source.json. An in-memory copy of the saved world moved one bench two grid cells to the nearest accepted placement. Its existing chosen approach used seven real grid steps and then the normal rest duration, with one completed activity and zero new model decisions. The copy receipt is arrival-clearance-saved-fixture.json; the actual database was unchanged. These are isolated tests, not a new native playtest. No new rendered frame was produced by this Python-only increment. Historical actual garden evidence shows the prior furnishings; it cannot verify the new placement or endpoint spacing. The next check is backed-up deployment followed by actual native placement and seating evidence before endurance. See docs/local-first-refinement.md.