GitHub stars aren't a quality signal

I kept running into the same list: “Best AI marketing skills,” ranked by GitHub stars.

Stars are honest about one thing: how many people found a repo interesting enough to bookmark. They’re silent about the thing that actually matters here — whether the output holds up.

There wasn’t a consistent way to answer that question, so I stopped looking for a better list and built a way to evaluate the work itself.

It’s called the Marketing Skills Standard.

The idea is fairly simple: six dimensions, a fixed rubric, and one rule that makes the evaluations reproducible. Every deduction needs a quoted line from the actual output.

Not “this feels generic.” The sentence that made it generic.

The harder problem appeared after a few evaluations had been scored. How do you know the score is actually supported by the evaluation, rather than just being a number typed into a field?

So the system checks itself.

Before an evaluation is published, a CI check recalculates the overall score from its six dimension scores and verifies that the numbers agree.

It already caught a real arithmetic slip in one of my own evaluations.

That’s the part I’m most pleased with — not the rubric, but the check that catches the rubric being wrong.

Stars aren’t a bad idea. They’re just easy to collect and only loosely related to what I actually care about.

The more systems I build, the more I notice that same problem: a convenient metric can quietly become a substitute for the thing it was supposed to measure. Popularity starts standing in for quality. Impressions can start standing in for trust.

I don’t think the answer is finding a perfect metric. It’s noticing when the measurement has become easier than the question, and being willing to build something more deliberate to get back to the thing that actually matters.