Evidence before code

Code is the most expensive way to answer "will anyone pay?" — you pay in months to learn something a conversation would have told you in minutes. The VALIDATE stage has one rule: no line of product code until a stranger has done something that cost them something.

The 2-week test

Pick the cheapest experiment that could disprove the idea, run it for two weeks, and write down the number before you start. Good experiments have three parts: a specific audience, a specific ask, and a metric with a target. "We'll see if people like it" is not an experiment. "Twenty freelance photographers see the pricing page; five click 'start trial'" is.

Landing-page 'validation' is fake

A waitlist of people who typed their email for a free thing proves curiosity, not demand. The internet is full of 4,000-person launch lists that converted to zero customers. The only validation that survives contact with reality is willingness to pay — or an expensive, verifiable substitute of it.

Ways to hear "yes" with money

• Pre-sell — a discounted annual for the first ten believers, refundable.
• Paid pilot — one customer pays a small amount for a hand-built version.
• Deposit / LOI — a signed letter with a number in it, for bigger fish.
• Manual service first — deliver the outcome by hand at a price. If nobody pays for the manual version, the automated version won't fix that.

Any one of these beats a thousand "this is cool!" replies.

Where the OS enforces this

• Experiments give every test a status, a metric, and a target — and keep the result on the record, including the failures. Especially the failures.
• Evidence you record here feeds the VALIDATE gate — you advance on proof, not enthusiasm.
• Override the gate anyway and the missing evidence is logged forever. The system doesn't normalize skipping homework.

Experiments are the unit of progress. Decided results — wins and losses — are the evidence the VALIDATE gate reads.

Next: Score it before you build it.