Curation methodology · v1.0
How we rank
Our Top-10s are computed, not bought. Every ranking comes from a versioned algorithm scoring properties on data we own or license — and we publish the whole thing, so you can check our work. It's the only travel ranking you can audit.
What powers a rank — and what never does
Rankings run exclusively on data we already own or license: our members' first-party reviews and bookings, engagement signals, licensed TravelGate/GIATA content, open data (OpenStreetMap/Wikivoyage), a human editorial rubric, and verified sustainability certifications.
Third-party review scores from the likes of Google or TripAdvisor are display-only — shown live and attributed on a property's page where licensing allows, but never stored and never blended into a score.
The score, 0–100
Six families of evidence, each weighted. The weights sum to 100 and are published here verbatim.
Ratings from our own members' verified stays — Bayesian-shrunk toward the city average — plus aspect sentiment on the review text, the share of reviews from verified stays, and how helpful other members found them.
Real bookings, wishlist adds, and how often a search turns into a closer look — measured over the last 90 days.
How well the property fits the way you travel (wellness, family, culture, gastronomy…), from licensed TravelGate/GIATA content and our curated travel-style tags.
Verified-stay rating averages, service observations from trips our concierge fulfilled by hand, and how complete the property's content is (amenities, images).
A structured editorial rubric, human-scored under a named byline, where every point is tied to a citable fact — never an opinion on its own.
Verified third-party certifications (Green Key, EarthCheck, GSTC), stored as a certificate ID and registry link — never invented.
The honesty mechanics
- Fair to small samples
- Volume metrics are log-scaled and turned into within-city percentiles. Ratings are Bayesian-shrunk toward the city average, so a handful of five-star reviews can't leapfrog a property with a long, strong track record.
- Minimum evidence to count
- A family only counts once it clears its data floor — member signal needs 5 reviews, demand 20 interactions in 90 days, sentiment 8 review texts. Below the floor a family is inactive, its weight is shared out among the rest, and our stated confidence drops a grade.
- Freshness decays
- Evidence loses weight with age on an exponential half-life: reviews 180 days, demand 90 days, editorial 365 days. We re-verify — we never auto-renew.
- Deterministic ties
- Equal scores break by more active families, then a higher verified-stay share, then the earlier listing, then alphabetically. Never at random.
- We never pad a list
- A “Top 10” publishes only with at least 10 genuinely rank-eligible properties. Below that we show an unranked “Stunning Shortlist” instead of inventing places to fill a number.
- Every rank is explainable
- Each published rank stores its full family-by-family breakdown and the exact evidence rows behind it, so any ranking — today's or last month's — can be reconstructed from stored, licensed evidence.
Rankings are not for sale
The scoring engine reads zero paid or partner data — that data physically lives outside the tables the algorithm can see. Anything sponsored is clearly labelled and sits outside the ranked list. Score history is append-only, so any change is auditable. A property that pays us cannot move up a rank.