Making Attention Computable
God's Eye View becomes a different engineering problem when several public sources occupy the same viewport. The question is no longer only where an aircraft, camera, or vessel is; it is whether the user can still read the important relationship among them. This chapter explains how the application turns world-anchored candidates into screen-space labels and cards, how it protects selected context without claiming completeness, and why its rendering budgets are part of the evidence a reader should attach to the view.
World anchors must end up in screen space
A data layer knows an object in world coordinates. A person reads the object's label, bracket, or card in a rectangular browser window. Those are different spaces. If a label is drawn at a fixed screen coordinate, it will drift away from its aircraft when the camera moves. If every source independently converts its world anchor to a screen position and solves its own collisions, labels from different layers will overlap or disagree about what should be visible.
God's Eye View keeps the anchor in the world and lets a shared overlay host perform the final projection. A source publishes a stable identity, a world position or position getter, a presentation variant, a title or detail, a priority, and—when appropriate—an activation callback. The host uses the camera's final view-projection state to place that entry in screen space. It can then test whether the anchor is behind the globe's horizon, whether it lies within the visible keyhole, and whether a candidate placement competes with dense interface chrome.
The timing matters. The overlay is painted after the scene has reached its final camera frame, so a bounding box tracks the same view that the user sees. A moving aircraft can update its display position once and have the trail, bracket, tracked readout, and label consume that same cached value. Reprojecting one consumer before the camera settles and another after it settles would create a subtle but damaging error: the text would appear to chase the object instead of belonging to it.
Horizon handling is also more than a viewport test. With the Cesium globe hidden behind Google Photorealistic 3D Tiles, the underlying globe may not write the far-side depth that would naturally hide a billboard. The overlay host therefore uses an ellipsoidal occluder to reject world points on the hidden side. A label near the limb can cross from visible terrain to sky, so the system uses a smooth keyhole or horizon fade rather than an abrupt pop. A cull is a geometric decision about visibility from the current camera; it is not a claim that the object is absent from the world.
The same host also manages interface collisions, but it does not treat the UI as a mask that cuts pixels out of the canvas. It records the rectangles of panels, docks, attribution, and other solid controls as placement constraints.