Building CRAYVO, part 2: eleven cities of nights, and the rules that keep them true

A city app with no nights is a brochure. In the first days of October we had eight cities and about a thousand nights, and I asked for more: founders' meets, social mixers, game nights, trail walks, acting classes, in every city, through February. Here is how 779 new nights arrived in three days, and the four rules none of them may break.

The rules first

• Facts only, in our own words: what, when, where, what it costs. We never copy a host's description.
• The host is named and linked on every night. We sell nothing. The link goes to them.
• A host who asks gets the night taken down within 24 hours.
• No program of ours reads Instagram, Facebook, TikTok or X. Their terms forbid it, and so do ours.

Those rules cost us supply, especially in Pakistan, where most nights are announced only on Instagram. We chose them anyway, because a listings product lives or dies on whether hosts trust it.

How a night gets in

1. Read. A scout, or now the engine, opens the event's own public page and takes the facts the page publishes for machines. Robots rules are obeyed and one page is read every five seconds per site.
2. Check. A chain of checks: is the time a real time with a time zone, is the place a place, is it a duplicate of something already on file, does it advertise alcohol in a city where we list none, are the words ours, is the picture the host's.
3. Score. Fit 30 points, trust 30, host 20, demand 20. A weekly game night with 42 people signed up on its own page scores 86 and is an A. A lecture with a price nobody could confirm scores far lower, and a person looks at it first.
4. Draft. The engine files a draft. It cannot do more than that, by construction. Part 3 is about why.
5. Publish. A person. Always.
6. Re-check. Every live night's page is re-read weekly. Cancelled or gone comes down the same day.

The numbers from the three days

• 206 nights from a sweep for mixers, founders' meets and game nights; then 512 from a wider one; then 61 from Seoul and Colombo, two cities we had not opened.
• Of 779, the checks passed 639 clean, held 131 for a person to confirm one fact, and refused 9.
• 246 websites probed as possible sources. 57 publish event data a program can read. 20 refused us through robots.txt and were left alone.

What I would tell another founder

Supply is a legal and ethical problem before it is a technical one. Read the terms of every source. Write down what they say. Approve each one by name, or do not read it. We built the registry that forces this, and it is slower than scraping. It is also the only version of this product I would put my name on.

Next: the night a two-factor rule locked our own robots out of the database, and the door we built instead.

Everything on in Dubai this week · Run clubs in London