How Your Launch Score Is Calculated (No Black Box)
Why the score is a pure function of what's actually on your page — and what that guarantees you.
Every scoring product eventually gets asked the same question: how much of this is real, and how much is a black box? For LaunchGraded the honest answer is that there is no black box — the score is 51 checks, each one a plain function that reads a fact (a scraped DOM value, a Lighthouse audit result) and returns pass or fail. Nothing in the scoring path is an LLM, a heuristic guess, or a judgment call made at scan time. That's a deliberate constraint, not a limitation we're working around.
Why determinism is the actual product
The entire value of a rescan — fix three things, run it again, see the number move — only means anything if the scoring is deterministic. If two identical pages could get different scores depending on which LLM call happened to run, or how a model felt about your copy that day, a rescan wouldn't prove improvement, it would just be noise you couldn't trust. Every check is a pure function: same input, same output, every time. That's the trust claim the whole product rests on.
The 9 dimensions, in short
The 51 checks are grouped into 9 dimensions, weighted by how much each one tends to matter for whether a visitor converts or a search engine can find the page at all:
- Messaging (13 points)
- Conversion (12 points)
- Trust (11 points)
- UX & Mobile (7 points)
- SEO (14 points)
- Performance (9 points)
- Accessibility (7 points)
- Technical (14 points)
- Discoverability (13 points)
Most checks are all-or-nothing. A handful are scaled — the Lighthouse performance and accessibility scores award partial credit proportional to the underlying Lighthouse ratio, so a 91/100 Lighthouse score isn't treated as a flat pass or fail, it's worth roughly 91% of that dimension's points. One check — broken links — is a gate: fail it and your verdict is capped at “Almost Ready” regardless of your total score, because a broken link on a page you're about to send traffic to is a different category of problem than a slightly weak meta description.
From points to a verdict
The 100 points roll up into four verdict bands: 90–100 is Launch Ready, 80–89 is Almost Ready, 60–79 is Needs Work, and anything below that is Not Ready. Any P0 blocker — a failing check flagged as critical, like a missing headline or no HTTPS — caps the verdict at Almost Ready no matter how high the underlying score is. A site can't score its way past a blocker; it has to actually fix it.
What this buys you
Because nothing in the scoring path involves a model reading your copy and forming an opinion, the same URL scanned twice with no changes in between will always return the same score. That's not a small technical detail — it's the difference between a checklist you can trust to measure real progress, and one that just tells you what an algorithm currently thinks of your page. The full rubric, check by check, is published on our methodology page and worked through by hand on the launch checklist.