A world model doesn't collapse because it forgets a single word, it collapses because the room itself has changed. A language model can exist within static text, but a robot an embodied agent cannot. It requires a real-time 3D reading of the exact space it is about to traverse: the floor surface, the edge of a loading dock, the corner of a staircase, or the gap between rows of shelving. It cannot rely on a postcard image or an overhead view from orbit. Such details are transient; a weekend warehouse reorganization can turn last month’s spatial mesh into an operational liability. This is a gap that satellite imagery and fleet-based mapping cannot bridge. Satellites still see the roof, and mapping vehicles still see the roads, but the interior where geometry dictates collisions, path planning, and safety changes faster than traditional data collection cycles allow. A world model fails here not due to a lack of parameters, but because the system operates on outdated observations it mistakenly believes to be current. @vangrid_io targets this specific issue through on-demand recapture. A data requester specifies a location and sets a bounty. A field operator records a slow-moving trajectory using a smartphone, capturing the site exactly as it stands. Faces and license plates are blurred on-device, ensuring identifiable raw footage never leaves the operator's hands. Each clip is assigned a content hash, rough location data, and a timestamp, with the batch root recorded on Base. The timestamp here is not merely a marketing detail; it transforms "freshness" into a verifiable attribute of the observation. Buyers no longer need to take it on faith that old scraped data still matches the building; they can see that this specific geometry existed within that specific timeframe. A deeper analysis reveals that the core problem isn't a lack of 3D data, but a lack of data with temporal provenance for dynamic spaces. While models excel at interpolating from old data, interpolation fails when shelves have been moved, floors raised, or aisles narrowed. Verifiable freshness elevates recapture from a task of "taking a pretty picture" to an auditable signal layer precisely what an embodied system requires before it can trust an interior map. Satellite occupies the upper tier. Fleet Map occupies the street level. Vangrid is a layer that can be sent back inside specifically, right as the interior is shifting.