How the Evidence Chain works

Where the ideas come from

Every morning the pipeline pulls a shortlist of up to five ideas from five independent discovery sources: two from GitHub Trending, plus the top item from Show HN, Product Hunt, Reddit's idea boards and Betalist. A deliberately small shortlist — one idea wins the day, and a wide net beats a deep one for finding it.

Sources are checked in parallel and none of them is load-bearing. When one is unreachable, that day's report says so on its face rather than quietly shipping a thinner shortlist.

From idea to problem

Each candidate is enriched with whatever its source offers — repo metadata and README for GitHub, the launch post or tagline elsewhere. Then AI extracts the problem it actually solves — because no signal check makes sense until "immich" becomes "self-hosted photo backup". That problem statement is what the Evidence Chain is run against.

The same step assigns an app type from a fixed list — SaaS, Mobile App, Developer Tool, AI & ML, CRM & Sales, E-commerce, Fintech, Marketplace, Productivity, Analytics, Content & Media, Infrastructure, Security, Gaming, Other — describing what a product built on the idea would be, not the technology it uses. The list is deliberately closed: free-form labels drift into three spellings of the same thing and stop being filterable.

Nothing gets deleted

One idea wins each day, but every scored idea stays in thearchive with the score and signals exactly as published — no quiet re-scoring, no disappearing runners-up. You can search it by problem, app type or source.

The four signal families

One source is an anecdote. Four in agreement is evidence. Each idea's problem is cross-checked against four independent signal families — Built, Discussed, Used, Searched — in that order.

When the chain agrees, the score goes up. When it doesn't — we show you that too.

The three axes

The Opportunity Score is a weighted blend of three axes: Friction (30%), Viability (30%), and Market (40%). Market carries the most weight because demand is the hardest thing to fake — friction and viability tell you if you can build it; market tells you if anyone cares.

• Friction (30%) — How easy it is to build on or replicate this. Higher = easier.
• Viability (30%) — Technical soundness and sustainability of the project.
• Market (40%) — Evidence of real demand across the Evidence Chain.

The Deep Dive

The highest-scoring idea of the day becomes Today's Pick and gets a Deep Dive: why now, the market gap, the main competitor, and a build plan for a solo dev — with every claim sitting next to the signals that support it.

What we don't do

• We never invent market sizes.
• We never claim demand our signals can't show.
• When the data is thin, the score says so.

Known limitations

• English-language bias: most of our signal sources skew towards EN communities.
• Very recent projects have thin signal histories — their scores are less reliable.
• Download counts only exist for repos published as npm or PyPI packages.
• Repo metadata (stars, forks, license) only exists for GitHub-sourced ideas; items from Show HN, Product Hunt, Reddit or Betalist are judged without it.
• Some sources rate-limit or block automated access. A day's shortlist can be shorter than five — when that happens, the report names the source that failed.