Why We Publish Our Scoring Weights
Most keyword tools hide their formulas. We put ours in the docs — here's the reasoning, and what our numbers can and cannot tell you.
Every keyword tool gives you a difficulty score. Almost none of them will tell you how it's computed. You're asked to trust a black number, calibrate your instincts against it over months, and bet domains on it.
We think that's backwards — especially for the kind of player wordradar is built for.
The indie SEO bet is asymmetric
If you run a portfolio of small sites, you don't need a score that's right in absolute terms. You need one that's comparable and explainable: is this SERP weaker than that one, and why. A free-hosted subdomain ranking top-3 is a signal you can act on regardless of what any vendor's 0-100 scale says.
So proxy-kd is fully published: base 50, free subdomain -12, UGC -8, video -6, wiki -5, major brand +15, deep page -4, shallow +3 — the whole table, with honest caveats, versioned in a public-facing docs page that our tests lock to the code.
What the number can't tell you
- The weights are uncalibrated against any proprietary vendor. They encode observed SERP weakness patterns, not a regression on someone else's "true" difficulty.
- A first-seen date is when a word entered our corpus — usually days-to-weeks before meaningful volume, but not a guarantee of anything.
- Trends risers are cached snapshots with a timestamp, not a live feed.
We'd rather you know exactly how the sausage is made and push back when it tastes wrong — methodology bugs get priority in our issue tracker.
The rest of the radar
The ledger behind it is the other half: a million-word suggest corpus with first-seen dates, type classification (new / re-spiking / evergreen), direction tags, and a site-matching column that answers "which of my sites should own this word". Plus an HN new-site radar, domain timelines, and the Stripe referral leaderboard — all queryable over REST and native MCP.
Beta is invite-only for now. If that's your kind of game, request access.