Methodology

Most restaurant guides are built from one viewpoint: a critic, a network of inspectors, or the raw volume of user reviews on a single platform. Each of those signals, on its own, can be wrong. A critic visits once. Inspectors miss neighbourhood rooms. Review platforms reward popularity over cooking.

Guidavera was built to resolve that. We eat in the cities we cover (not yet at every restaurant we list, and we’d rather say so than imply otherwise), and we ground that lived experience in a transparent consensus drawn across multiple independent sources. Every listing is verified by hand and ranked by a Consensus Score that reflects how a restaurant is regarded across the full picture, not one opinion within it.

Guidavera's consensus draws on several independent categories of signal:

• Authoritative international guides: recognition from established restaurant authorities whose awards carry weight in the industry.
• Professional critics and press: long-form reviews and features from trained food writers.
• Local food publications: editorial coverage from city-specific publications with meaningful editorial standards.
• Diner review platforms: large-scale review volume, used as a secondary signal after editorial recognition.
• Restaurant-verified data: hours, menus, contact details, and photos sourced directly from the restaurant and confirmed against their public channels.

The weighting behind the score is published in full below. We used to keep it private on the theory that published weights invite gaming. We changed our mind: the inputs aren’t ours to game. They belong to Michelin, Repsol, professional critics, and thousands of diners. A restaurant that wants to climb the ranking has exactly one lever, and it’s the one we want pulled: cook better.

Every ranked restaurant gets a score from 0 to 10, built from two blocks. A fixed block (55%) covers authority recognition, where not being listed is itself a signal. A city block covers the remainder: local critics and diner platforms, chosen per destination to match who actually covers that dining scene. These are the weights as of August 2026.

The city block differs by destination, because the critics who matter differ by destination. Barcelona has Macarfi and a standing Time Out editorial team; Mallorca doesn’t, so its stack is built on the island’s own authorities.

Three rules keep the score honest:

• Missing data is redistributed, not punished. When a city source hasn’t covered a restaurant, its weight is spread proportionally across the sources that have. Restaurants are compared on the quality of the data that exists, not penalised for coverage gaps.
• Review volume can’t buy a rank. Diner ratings are volume-weighted (Bayesian shrinkage): a rating backed by a handful of reviews is pulled toward the platform’s average until enough reviews exist to trust it. A 5.0 from 12 reviews does not beat a 4.7 from 3,000.
• Professional distinction always outranks its absence. Restaurants sort first by a distinction hierarchy (3 Michelin stars at the top, then 2 stars or 3 Soles, down through Bib Gourmand, Selected, and Recomendado), and the weighted score ranks them within each level. No volume of diner reviews lifts an unstarred restaurant above a starred one. Restaurants with no professional distinction are scored on city sources alone, since absence from Michelin is not a meaningful negative for a neighbourhood room Michelin never set out to review.

Weights are recalibrated as we learn (the section below explains how the system evolves), and this page is updated when they change. The inputs behind any individual score are printed on the restaurant’s own page, in its critic and diner ratings, so you can check our reading of a restaurant against the sources themselves.

Guidavera is built to keep getting better. The data we hold, the signals we draw on, and the way we weigh them are all in continuous improvement. A restaurant’s rank today is a snapshot, not a verdict for all time.

• Rankings shift as signals arrive. A new critic review, a fresh accolade, a price change, a shift in diner sentiment, or a confirmed closure all feed back into the score. Restaurants move up and down as the picture changes.
• The algorithm itself improves. We refine how we weigh sources, surface signal, and account for volume bias as we learn more about what genuine quality looks like across cuisines, neighbourhoods, and formats.
• New inputs broaden the picture. As we add new sources, finer-grained attributes, and richer on-the-ground data, the consensus sharpens. Restaurants that were previously underrepresented become visible, and ones that relied on a single popularity signal recede.

A consensus guide that never moved would mean either the world had stopped changing or we had stopped paying attention. Neither is true.