The U-Score™ explained: how we rate every venue 0–100

I spent the first year of building Unight asking the wrong question.

The question was: how do we rank restaurants and bars? Every answer I came up with — average user ratings, AI sentiment from reviews, expert panels, weighted star averages turned out to be a worse version of something that already existed and didn't work. Yelp had reviews. Google had ratings. TripAdvisor had badges. The internet was drowning in venue scores. None of them helped me decide where to actually go on a Thursday night.

Eventually, I figured out the question was wrong. The right question wasn't how do we rank venues. It was: what would a friend with great taste actually consider before telling me where to go?

The U-Score™ is our answer to that question. It's a 0–100 score we assign to every venue on Unight. It's not crowdsourced. It's not an average. It's a quality signal built the same way a friend would build one in their head — by thinking about a place across five separate dimensions, then deciding whether it adds up to something worth your time.

This is how it works.

Why we don't use stars

Before I explain what U-Score is, let me explain what it isn't.

It's not a rating average. A 4.3 on Google is a popularity number. It tells you a lot of people went somewhere and most of them weren't actively angry. It doesn't tell you whether you'll have a good night, whether the place is right for your mood, whether the bartender is any good, or whether the room will be empty on a Tuesday.

It's also not a critic score. A single voice — even a great one — can't be calibrated against your taste. Critics rate against their own benchmarks. The U-Score is calibrated against the question that actually matters: should this person, looking for this kind of night, go here?

We threw out the star system entirely. Then we built a new one.

The five pillars

Every venue on Unight gets scored across five dimensions. Each pillar gets a number 0–100. The U-Score is a weighted composite of those five.

The pillars are, in fixed order:

1. Experience. How good is the thing itself — the food, the drinks, the music, the show, whatever this venue does for a living? Experience is the floor. If the experience score is low, nothing else matters. A bar with great vibes and terrible cocktails has failed at being a bar.

2. Vibe. Atmosphere, lighting, music, energy, the room. Vibe is what people remember when they can't remember what they ate. It's the most subjective pillar, which is why we judge it as a team rather than letting any one person decide.

3. Value. Quality relative to price. A ¥40 noodle shop can have a higher Value score than a ¥800 omakase. Value is not "cheap." Value is what you got for what you paid. This is the pillar that most rating systems get wrong — they conflate price with quality.

4. Social. How well does this place work for the moments people actually use it for? Date night, group dinner, solo drink at the bar, after-work, brunch with parents. A bar that's great for first dates can be terrible for a birthday of twelve. The Social score captures fit.

5. Service. Staff, pacing, attentiveness, mistakes handled well. Service is the pillar people notice most when it's bad and least when it's good. It's also the most operationally telling — bad service usually means the venue isn't being run well, and that bleeds into everything else over time.

These five came out of nine years of running Unight in Shanghai. We tried more categories. We tried fewer. Five is what holds.

What scoring actually looks like

Here's a real example. Call it Maison Lune — a small natural wine bar in Jing'an.

• Experience: 92. The wine list is genuinely interesting. The small plates are well-executed and the kitchen has a point of view. This is a bar that cares about what it serves.

• Vibe: 89. Warm, low-lit, twelve seats at the counter and four small tables. Music never too loud. The kind of room that makes a conversation feel like a small event.

• Value: 78. Not cheap — drinks are ¥120, plates are ¥80–180. But the quality justifies the price. The score is lower than Experience because it's not a place to take someone if they're counting yuan.

• Social: 85. Excellent for a date, a small catch-up, or a solo glass at the counter. Less good for a group above four — the room simply isn't built for it.

• Service: 91. The owner is usually behind the bar. Staff know the wine list and don't oversell. Pacing is patient.

Weighted, that's a U-Score of 87.

That number on its own doesn't mean much. But when you see Maison Lune appear in your Unight feed for "natural wine, date night, Jing'an," and the 87 sits next to it, you know two things at once: this matches what I asked for, and this is genuinely good.

That's what the U-Score is for. It's not a ranking. It's a confidence signal.

Who actually scores venues

The U-Score is generated by both — AI and humans, working in sequence.

We've engineered a proprietary scoring system trained on nine years of operational experience: which venues delivered, which underperformed, which broke under pressure, which kept their edge as the market shifted. That system runs the first pass. It scores every venue across the five pillars using signals our content team would weigh — but at a speed and scale no human team could match.

Then the content team takes over. They validate. They walk through the AI's reasoning, push back on what doesn't match what they know, adjust scores where the system over-weights or under-weights, and approve the final number. Nothing publishes without human sign-off.

This matters. The moment scoring becomes fully automated, it stops being a quality signal and becomes an SEO target — venues optimize for the algorithm and the score loses its meaning. We've seen this happen to every platform that started as a curation product and ended as a ranked directory. We're not interested in that ending.

So we built it the other way: AI for the leverage, humans for the judgment. The system scores at scale; the team makes sure it's right. That's why we have 500 venues in Shanghai, not 5,000. Curation at this depth doesn't scale to "every restaurant in the city." It scales to "every restaurant worth your time."

What the U-Score is not

Three things worth being clear about.

It's not paid. Venues cannot pay to be scored higher. They cannot pay to be scored at all. Inclusion in Unight is editorial, full stop. Featured placement (visibility within already-curated content) is available as a paid product — clearly labeled, never affecting score. It's the same model the Michelin Guide has used for a century: you can't buy stars, but starred restaurants can buy advertising. The line is the whole product.

It's not fixed. Scores change. Venues evolve. A score from 2024 isn't a score in 2026. We re-score regularly, especially when we hear from users that something has shifted.

It's not a leaderboard. We don't publish a top-10 list of Shanghai venues by U-Score. The right venue for you depends on what you're looking for — and the AI uses U-Score as one input among several to find that match. A U-Score 91 cocktail bar is not the answer if you wanted brunch.

Why this is the thing we're building

I'll be honest about the strategy. The U-Score is the most defensible thing we have.

Anyone can build an app. Anyone can train an AI on scraped review data. Anyone can write copy that sounds like a curation platform. What no one else has is nine years of operating experience in one of the most competitive lifestyle markets in the world, feeding a scoring methodology calibrated by a human team that has refined it over thousands of venues.

That's the moat. It compounds with every venue we score and every user who tells us we got it right or wrong. The longer Unight operates, the harder it gets to copy.

When you see a U-Score next to a venue on Unight, you're not seeing a number. You're seeing the output of a system designed to do one thing: tell you whether somewhere is worth your time.

That's the only job that matters.

The U-Score™ is a proprietary methodology of Unight, operating in Shanghai since 2017. Unight is the AI companion that helps you decide where to spend your time in cities — starting in Shanghai, expanding globally in 2026.