How it works

Buildwhat runs two daily proceedings. Ideas are generated from what people complain about on Hacker News, then put on trial; most are killed. Launches are real products posted as Show HN, each reviewed for whether people would use or pay for it.

Ideas: where they come from

Every day at 20:00 UTC we read the stories that reached the front page of a large tech community in the previous two days, keeping only discussions that have matured: at least 150 points and 60 comments. The source of every idea links back to its discussion.

The trial

Triage. We drop stories with no commercial pain in them: news, politics, obituaries, history, pure science, corporate drama. Interesting is not the same as buildable.
Deep read. Up to eight stories get read in full: the linked article and the top twenty comments. Ideas must come from what people actually said.
Hard gates. An idea is rejected outright if it lacks a specific buyer with a price or budget; if it comes from the comments and fewer than two different people described the problem; if it cites a comment that doesn't exist; if any score is 3 or lower; if it names no competitors or no gap; if it is framed as a generic "AI-powered" wrapper; or if it copies a commenter's words. We paraphrase, never quote.
Red team. A second, skeptical review tries to kill every survivor: does it already exist, is it a feature rather than a company, would the buyer really pay, is it a repeat of a recent idea? Its scores can only lower an idea's scores, never raise them.
The verdict. The top three that still score 6.0 or more after the red team survive. The red team scores harshly, so a 6 here is an idea that survived a determined attempt to kill it. Killed ideas are listed with the reason.

Scoring

Each idea is scored from 1 to 10 on six axes, against written anchors (what a 3, a 6 and a 9 look like). The radar shows the shape: two ideas with the same score can have very different weak spots.

Pain: How costly and urgent the problem is. 9 means people already pay for bad workarounds.
Willingness to pay: Whether a budget owner exists and spends money on this today.
Timing: Whether something changed in the last year that makes this newly possible or necessary.
Gap: How badly existing products serve the need.
Moat: What would stop a copycat: data, workflow lock-in, network effects, distribution.
Feasibility: Judged against the track. Indie: a sellable MVP from one to three people in under three months. Venture: a credible path to a fundable milestone within a year.

The overall score is a weighted mean computed by our code, not by a model: pain ×2, willingness to pay ×2, gap ×1.5, timing, moat and feasibility ×1.

Launches: the review

Every three hours we read each new Show HN post (the title plus the post text, or the site's own description if there is none) and score it on whether people would actually use or pay for it. Nothing is invented here: these are real products.

Problem: Is there a real, specific problem for a clear audience?
Demand: Would people adopt it, or pay for it? This counts double.
Novelty: How different is it from the tools people already use?
Credibility: Is it clearly real: a working product, a demo, open source, specifics rather than claims?

Launches scoring 6 or more are marked Strong, 5 to 6 Solid. Toys, clones, thin AI wrappers, vague pitches and non-products are Dismissed as noise and hidden unless you ask for them. Every morning a digest picks the strongest. The score only sees the title and description, so open the launch before you trust it.

Limits

Ideas and reviews come from public discussions, assessed by AI models and checked by code against the rules above. They are starting points for your own research, not advice. Verify the market before you commit to anything.