Hotel mapping and normalisation

This is the hardest problem in hotel distribution and the reason companies exist to solve only it. Three suppliers return the same Dubai hotel as 'Downtown Grand Dubai', 'Grand Downtown Hotel & Suites' and 'GRAND DOWNTOWN DXB', at three slightly different coordinates, with three different address formats and three unrelated internal ids. Without resolution, your search page shows the same hotel three times, your cheapest-rate logic is wrong, and your customer sees a mess.

TravelCore resolves properties through a layered match: exact identifier match where a shared identifier exists, then geospatial proximity within a tight radius, then normalised name similarity, then address and postal-code agreement, then corroborating signals such as star rating, room count and phone number. Each candidate pair carries a confidence score.

The part that matters commercially is what happens in the middle band. High-confidence matches merge automatically. Low-confidence pairs are rejected. The uncertain band goes to a human review queue rather than being guessed, because a wrong merge sells a room in the wrong building and that costs more than the review does.

This module is real code rather than a diagram, and it runs on this page. The engine below is the same TypeScript the test suite imports, covered by 37 tests over the cases that cost money, executing in your browser across a fixture catalogue. Move a threshold and the bands move, because there is no pre-computed result to move instead.