Where ideas go to die
Five gates, ordered from cheapest to most expensive to check. Each one eliminating. An idea that fails the first gate costs ten minutes; the same idea, coded, costs three weeks.
The price of opening a repo
85 MINUTES · FIVE GATES · ONE WAY THROUGH
The five gates
01 · The source · 10 min
Open licence, named public producer, published on a known schedule, granular enough to say something about one person. Scraping is not a source — it's a dependency that breaks without notice.
Killed by: no clear licence.
02 · Search intent · 15 min
The demand has to already exist. You don't create search intent alone and unfunded. One query above 5,000 a month, or three above 1,000.
Killed by: below the display threshold.
03 · The incumbent · 20 min
Does the results page serve articles, or tools? Articles mean the space is empty. A funded private tool in the top three means it isn't. This is where most good-looking ideas die — the stronger the personal stake, the likelier someone already owns it.
Killed by: an established tool in the top three.
04 · The verdict · 10 min
Finish this sentence in one line: I type ___ and the site tells me "___". The second half has to be about me, not about the world. If it can't be written, it's an article, not a product — and nobody shares a dashboard.
Killed by: the sentence won't write.
05 · Risk and load · 30 min
Does a wrong answer hurt someone — pensions, tax, health, benefits? Does it work without collecting anything? How many refreshes a year? Does it fit inside what I already pay for?
Killed by: more than two refreshes a year.
What doesn't count
A clean dataset. Personal interest in the subject. Technical difficulty. A beautiful chart. All four are seductive, none of them predicts anything.
The log
A rejected idea keeps one line and one reason. Without it, nothing stops me reopening it in six months and redoing the work.
Seven ideas run against the five gates. Property valuation: cleared gates 01 to 02, killed at gate 03. Pension simulator: cleared gates 01 to 02, killed at gate 03. Mood-tracking SEO pages: cleared gate 01, killed at gate 02. Ad-revenue extension: cleared gates 01 to 04, killed at gate 05. Yield analytics for publishers: cleared gates 01 to 02, killed at gate 03. Reviving an old side project: cleared gates 01 to 02, killed at gate 03. Salary comparison tool: cleared all five gates and became a repository.
Gate 03 does most of the killing. The stronger the personal stake, the likelier someone already owns the answer.
After the five gates
Passing isn't a project — it's a candidate. Two things come before any code: freeze the slugs, because URLs never get renamed after indexing, and name the distribution channel, then test it empty. Never build the landing surface before testing the channel.