I built a launch-readiness scanner. Its first useful finding was a flaw in its own judgment.

I have built and tested many different digital products.

The products changed.

But one pattern kept returning:

Something could look finished without being ready.

A page could load correctly, look polished, and work on mobile — while still missing important signals related to security, accessibility, search visibility, or structure.

So I built Launch Reality Check.

The idea was simple:

Enter a public website URL, check a set of launch signals, and turn the findings into a prioritized action list.

The first website I tested was my own: Genius Brain.

The result was 44 out of 100.

It found:

• 9 passed checks
• 2 high-priority findings
• 4 medium-priority findings

Some findings were useful.

The page title was too long.

A canonical URL was not detected.

Several security and structured-data signals needed attention.

But the scan also exposed problems in the scanner itself.

A passed check still displayed a corrective action.

The interface said 14 checks, while the report contained 15.

More importantly, the scanner sometimes treated “I could not verify this from the public HTML” as “this does not exist.”

That distinction matters.

Not detected is not always the same as missing.

And uncertainty should not be presented as failure.

So the first real lesson from Launch Reality Check was not about my website.

It was about the responsibility of the diagnostic system.

A scanner should not only find problems.

It should also explain:

• what it directly observed
• what it inferred
• what it could not verify
• how much confidence it has in each finding

I am now revising the scoring model before making the tool public.

The next version will separate:

• Passed
• Failed
• Manual verification required
• Not applicable

It will also stop treating every technical signal as equally important to launch readiness.

I’m curious how other builders handle this problem.

When your product evaluates another product, how do you prevent the score from sounding more certain than the evidence actually is?

And what would you want a pre-launch scanner to verify before you trusted its result?