The rubric
How the score is built, and what it cannot tell you.
Most idea tools give you a list. Lists are cheap to generate and impossible to verify. Demand Radar does the opposite: it reads public text that already exists, counts it, shows you the raw quotes, and then asks you for the one thing a machine genuinely cannot know — whether you can build it.
The five dimensions
| Dimension | Points | Who assigns it |
|---|---|---|
| Demand volume | 20 | Automatic |
| Complaint intensity | 20 | Automatic |
| Competition gap | 20 | Automatic |
| Payment evidence | 20 | You |
| Buildability | 15 | You |
| Reuse | 5 | You |
Sixty points are computed from cited public data. Forty are yours. That split is deliberate: the failure mode of every automated scoring tool is false precision, and the one thing a scanner cannot see is whether a single person can ship the thing in two weeks.
Where each number comes from
Demand volume — 20 points
We count Hacker News comments matching your keyword in the last 12 months, plus Stack Overflow questions at 0.6, plus Reddit threads at 1.5. Reddit carries the highest per-item weight because a thread is a deliberate act of asking, not a passing mention. The sum is log-scaled, so the difference between 50 and 500 matters far more than between 5,000 and 50,000. Interest saturates; it does not compound.
Hacker News totals are approximate. Algolia reports an estimated hit count that can move by a few percent between identical calls; the comment sample we actually read is exact, the headline number is not.
Complaint intensity — 20 points
We pull up to 100 recent Hacker News comments and up to 100 Reddit threads, then read all of them for pain language: I wish there was, frustrating, wasted, gave up, why is there no, and 23 other markers. The score blends two things — the share of text that reads as a complaint, and the absolute count. A high ratio on a tiny sample is noise; a high ratio on a large sample is a market.
Reddit does most of the work here. It is where people describe a problem in their own words, at length, to an audience of people with the same problem. Hacker News is where they argue about it afterwards.
App Store reviews do not feed the score directly. They feed the evidence panel, because a 1-star review is worth more to you as a sentence than as an integer.
Competition gap — 20 points
We count apps in the US App Store whose title matches your keyword and that have at least 500 ratings. Five hundred is the line where an app has a real user base rather than a launch spike. GitHub repository counts are displayed but not scored: they are too noisy to price a market, and a failed GitHub call returning zero would make a crowded niche look empty — the worst direction for an error to point.
This is why a scan can show 3,000 repos and still read as an open shelf: solved technology plus no product winner is often the best place to stand.
Payment evidence — 20 points, yours
Two checks, ten points each: you found people already paying for a workaround, and a worse tool already charges money. Both are things you can verify in ten minutes on Fiverr, Upwork, or a competitor's pricing page. Neither is visible in public text, which is exactly why we do not pretend to compute it.
Buildability and reuse — 20 points, yours
Three checks for buildability (no license or regulated data, two-week core journey, no inventory or human ops) and one for reuse. These are the questions that kill more ideas than demand ever does.
Reading the verdict
70 and above — build it. Evidence is present and your execution profile holds up. Go build the smallest version and try to charge for it this month.
50 to 69 — worth watching. Usually this means real demand with a crowded shelf, or a clean shelf with thin evidence. Pick the one weakness and spend a week on it before writing code.
Below 50 — let it go. Not because the idea is bad, but because the evidence you can see does not support spending a quarter on it. Most ideas land here. That is the point.
What this score cannot see
- Non-technical audiences. Hacker News and Stack Overflow are developer-heavy. A clinic manager with a spreadsheet problem leaves no trace in either. Reddit narrows that gap but does not close it. A low volume score is a reason to look elsewhere for evidence, not a verdict.
- Reddit depends on a proxy, and Reddit fights proxies. Browser requests to Reddit are blocked, so scans route through a Cloudflare Pages Function. It authenticates with an app-only OAuth token where credentials are configured, and falls back to anonymous JSON where they are not. Reddit blocks many datacenter IP ranges either way. When the proxy cannot reach it, the scan marks the source as unavailable and scores from the rest rather than pretending the count is zero.
- App Store reviews can go quiet. Apple throttles its review feed by IP and signals the throttle with an empty feed rather than an error. The scan routes through a cached proxy and reports an empty feed as a failure, not as "this app has no complaints".
- Hacker News totals are estimates. Algolia's hit count is approximate by design. Two identical scans can differ by a few percent.
- Private markets. B2B niches that live in Slack groups and trade associations are invisible here.
- Timing. A score is a snapshot. A niche that scores 55 today can score 75 in six months when a platform changes its rules.
What we refuse to do
- We do not generate ideas with a language model and present them as research.
- We do not estimate market size. Every TAM number you have ever read was invented by someone.
- We do not store your scans on a server. Everything runs in your browser.