Personalization & Recommendation Platform

A guest views and saves campsites. Each interaction is captured by the shared behaviour layer as a weighted signal - a booking click counts for more than a hover. The ranker generates a candidate set, then blends booking co-occurrence, facility similarity, that behaviour affinity, review quality, freshness and loyalty into a single score.

It then applies the business-rule filter, drops anything below a minimum score, re-ranks, and enforces per-region diversity so the top slots aren't all from one place. The result fills the rails live. The "How these are picked" panel opens the inspector with the full $0 trace; an optional explanation narrates it in plain language.

The same engine turns a dead end into a saved booking. When a guest's chosen site is fully booked, the recovery flow anchors on it as the entry point, runs the identical business rules, and returns exactly three similar sites that pass them - delivered as a simulated email and recorded in the console's recovery log. Nothing about the recovery bends the rules that govern the everyday rails.