How the score works

The Hidden Gem Score is an opinion with its work shown. It’s a number from 0 to 100 for how much a place feels like a secret worth keeping, built from five marks you can see on every place’s page.

The five marks

Each mark is 0 to 100. The score is a weighted average, so the weights say what we care about most.

1. Uniqueness

   25%

   Could this exist anywhere else?

   The only place of its kind scores high. A great version of something common scores lower, however good it is.

2. Atmosphere

   20%

   What does it feel like to be in the room?

   Light, sound, the people, the clutter. Dim, strange and lived-in beats shiny and new.

3. Story

   20%

   Is there a real reason this place exists?

   A founder, a building, a recipe, a neighborhood. If we can’t find a story we can stand behind, it earns no points here.

4. Signature item

   15%

   Is there one thing you should order?

   It’s the “Order” line on every place. A place can be wonderful and still have no single dish worth a detour.

5. Under the radar

   20%

   How few people already know about it?

   Set by the label on every place: a deep cut gets 90, an open secret 60, a local legend 25. Legends can still score well. They just don’t win by default.

Why these weights? Uniqueness leads because it’s the whole point of the site. “Under the radar” was added after our first test batch, when the most famous places kept landing on top simply because more had been written about them. The signature item counts least because a great dish is the one thing every other guide already finds. These weights are a judgment call, not science, and you can see them in the calculator below.

Try the math

Move the sliders and pick how well-known the place is. This is the exact calculation the site uses: each mark times its weight, added up, rounded to a whole number.

Where the numbers come from

1. Find. Candidates come from OpenStreetMap, a free, community-built map, from local press and neighborhood writing, and from places people suggest to us. Chains are filtered out. We don’t use Google or Yelp data.
2. Research. An AI model reads what’s publicly written about each place, from local press and neighborhood guides to the place’s own pages, and drafts four of the marks, a short explanation, the story, the one-line summary and the “Order” line. It runs once per place and the results are saved. Nothing is computed per visitor.
3. Review. A person reads every draft, checks what they can, and approves, edits or drops it. Only approved places go live.
4. Stamp. Every place carries a “last verified” date. An old date means old information, so treat it that way.

What the score gets wrong

Said plainly, so you can decide how much to trust it.

• AI drafts can invent or misremember details. Review catches a lot, not everything. If a page disagrees with the sign on the door, believe the door.
• It’s taste, not measurement. Two sensible people would score the same place differently. The aim is to be consistent and open about the rules, not to be objective.
• Truly hidden places are the hardest to research. Almost nothing written means thin evidence, so those marks are less certain than the number suggests.
• It only knows what’s been written down. Places that are barely online get missed, and those are often run by people and communities who are less online. That’s a real bias and we haven’t solved it.
• It says nothing about cleanliness, food safety, price, accessibility or whether the place is open tonight. Check before you go. Some of these places are cash only.
• Sources age. An article from five years ago may describe a kitchen that has since changed hands. We look for recent signs a place is still open, and every page shows when it was last checked.
• A secret doesn’t stay secret. A score is a snapshot from its verified date. A place can stop being unlisted.