Reference board

The proposal was the method every region would follow. A development-only authoring page loaded the official interactive map's tiles under a drawing layer, and the author drew coastlines, ridges, rivers, cliff bands, plateaus, ground paint and every landmark's place over it. The map's pixels were converted to metres by timed runs between waypoints and a character's height used as a ruler, and heights were set by eye from views along ridges and glides from summits. Each landmark was photographed in Photo Mode by a fixed routine, each capture matched to a camera from its vanishing lines, and the recreation was judged by laying the capture over the live canvas and tuning shapes, kit parameters and colours until the misregistration looked gone.

Why not: every unknown it set by hand is exact data the game already holds. The open world's ground is a TerrainData per 1024-metre tile, its heights in metres on the game's own coordinates, and every object it places is a StreamGen record with its place and turn (game data formats), so a region's layout is read, not drawn, and needs no transform from a map's pixels. Heights set by eye and outlines drawn over a stencil are a search with the author as its score, and an overlay tuned by eye is the frame's score as the loop, judging every unknown through every other one. The method that replaces it is the scene derivation: each region laid out from its placements, each stand-in fitted against its own export, and each kind of unknown answered in its recreation pass by its own measure. The references it gathered are kept as parity keeps any, outside the repository and named in ParityReferenceMap, found by a search of published recordings before any of the game is recorded, with each camera solved on landmarks by genshin:parity pose rather than read from vanishing lines.