Tooling
Mirai Guide: anchoring booths to venue maps
Vision-assisted map review that stores normalized coordinates and hashes — never the uploaded bytes.
A venue map in Mirai Guide is not an image with pins drawn on it. It is a reviewed set of normalized booth coordinates anchored to a specific map revision, produced with vision assistance and approved by a human.
The anchoring contract
- A map belongs to exactly one venue, and it may only position booth-assignment IDs that the vendor-list stage already approved. Map extraction structurally cannot create or rename a booth — identity and position have different authorities.
- Venues with multiple pages, floors, and zones are supported, and each map has revisions with one active at a time, so a mid-event layout change supersedes rather than overwrites.
- Uploaded reference maps are transient vision inputs. Their bytes are never stored: only a content hash, the pixel dimensions, extraction diagnostics, and the reviewed normalized coordinates are retained. The published bundle carries clean data, not somebody’s photographed floor plan.
Venue identity across years
Event editions re-create their structure yearly, but the buildings do not move. A separate physical-venue registry records the addressable hall — not the whole building — with trilingual names, coordinates, and a capacity tier, and each edition’s venue rows link to it. A row is a hall rather than a building because capacity is a property of the hall: one complex can host differently-sized spaces, and tying the tier to the building would make it ambiguous.
That registry is what lets year-over-year data — sellout behavior, queue calibration, walk timings — attach to the place itself instead of being re-entered per edition and drifting.
Why normalized coordinates
Coordinates are stored normalized to the map raster rather than in pixels, so a re-exported raster at a different resolution doesn’t invalidate reviewed anchors, and every consumer — booth highlighting, the cart’s visit-plan line, and walkable navigation — reads the same anchor set at whatever scale it renders.