An open method matters more than a mysterious score

How life trade-offs become city recommendations

This is not a national “most livable city” ranking. The same city can be a completely different answer for two people, so we first recover the life you want and then keep feasibility as a separate layer.

Two layers, so aspiration and feasibility do not blur together

Lifestyle asks where you are drawn. Reality asks what needs checking now.

Lifestyle fit

Find the city structure that resembles you

A/B choices build a preference vector across activity, nature, climate, transport, careers, housing, healthcare and social boundaries. The vector describes trade-offs, not a personality type.

Real-life feasibility

Keep time, budget and household constraints visible

Short stays use month, duration, party and work pattern. Long-term life adds income, industry, household and renting or buying. The Beta turns incomplete constraints into a checklist, not fake precision.

The current recommender is deterministic — no LLM chooses the city

The same answers produce the same result, and every contribution can be recalculated.

Answer preferences×Relative city traits×Seasonal month weight=Raw city score

1 · Answers become a vector

Each side adds fixed positive or negative weights to a few dimensions. All selected weights are summed.

2 · Compare the vector with every candidate city

Each city has 1–5 research scores on the same dimensions. The algorithm centers them at 2.5 and multiplies direction by preference.

3 · Month only amplifies relevant climate choices

Summer emphasizes heat, coolness and humidity. Winter emphasizes warmth, sunshine and humidity. It never decides whether you prefer hot or cold.

4 · Diversify the final three

The top raw score becomes the primary. Alternatives preferentially change region and city archetype so all three are not near-duplicates.

Why not let a large model pick from every city in China?

Language models can organize evidence, flag conflicts and explain results, but should not improvise a nationwide ranking from memory. The current system can only recommend from 18 profiled candidates. Expanding coverage requires dated, sourced, comparable city data — not merely a larger model.

Why vivid A/B choices instead of “Do you like big cities?”

Abstract questions let us want everything. Concrete details force priorities to surface.

Lived situations

Questions come from recurring city-life themes: commuting, weather, renting, buying, careers, nights, healthcare, children and distance from home.

Both sides must work

Each option should be attractive and costly: a smaller central home versus a complete home farther out, not an obvious right and wrong.

Ask as little as possible

Travel uses 9 core questions and long-term life 12. Children, parents, solo travel, remote work and housing plans trigger only relevant extras.

Community posts discover questions; they do not become statistics

Platforms such as Xiaohongshu can reveal what to ask, but not directly score a city.

Remove obvious promotion

Property, agent, tourism and commercial posts are filtered out, as are reposted lists without lived detail.

Look for independent repetition

A theme should recur across time, identities and opposing views. A complaint is not automatically a city fact.

Return to official definitions

Climate, transport, healthcare, housing and employment need official sources with year, geography and update date.

Current coverage

The Beta uses 18 deliberately distinct cities to validate the questions and logic. It does not pretend to cover all of China.

BeijingShanghaiShenzhenGuangzhouHangzhouSuzhouNanjingChengduChongqingWuhanChangshaQingdaoDalianKunmingXiamenXi’anGuiyangNingbo

Why not rush to sixty cities?

More cities make false precision more dangerous. New candidates need regional value, dated sources and a clear explanation of which life scenarios the data supports.

Priority order for formal data

National definitions first, local yearbooks and regulators second, commercial rankings only as leads.

Official public source ↗

Limits you should know

The result is a place to begin investigating, not an irreversible decision.

Differences within a city can be larger than differences between cities.
Rent, prices, jobs and transport change quickly and need a current check.
A short questionnaire cannot cover partners, hukou, school catchments or specific medical needs.
We do not claim a fake “92% match”; reasons, costs and confidence matter more.

Ready?

Find one direction worth understanding properly

Start choosing →