Someone sent me a screenshot of my own board once, just to show me something. Forty-seven votes on a feature request. They were excited about it. And looking at it, I felt this small unease I couldn't quite name at first.
I wrote the code that produces that number. I know exactly what it counts. And that's the problem.
The count measures position more than demand
Here is the thing I keep watching happen. A feature sits at the top because it has votes. People vote for it because it's at the top. Because it already looks popular. Because they see it first.
And then that number gets quoted in a planning meeting as if it means something clean.
A request buried three screens down gathers a fraction of the votes, not because fewer people care about it, but because fewer people ever reached it. Put the same request at the top for a week and watch it accumulate. The number goes up. The demand does not move. What you're actually reading off the top of the board is position and momentum, with real demand somewhere underneath, quieter, harder to see.
I've seriously considered hiding the vote count on my own board. That's a strange thing for someone who built a voting tool to think about. The fact that it's even tempting tells you something.
A comment is worth two thirds of a vote, on purpose
I didn't let the raw count stand alone. The score that ranks a feature is votes times three, plus comments times two, plus views times a tenth. A written sentence about why you need something is harder to fake than a click on the most visible row. It costs a moment of actual thought. That small cost is enough to thin the reflexive votes from the considered ones.
The merge is the part I trust most. When two requests turn out to be the same idea said differently, you merge them, and the first thing the merge does is delete the duplicate votes from anyone who voted for both, then recount from scratch. I've watched a fifty vote feature come out the other side in the thirties, because a third of its votes were the same people endorsing two phrasings of one wish. Together they looked like a mandate. Deduplicated, they were an ordinary request with a loud echo.
The merge is the closest thing the product has to a lie detector. And most of what it catches is the count lying to you.
The other thing the count hides
Position and momentum are only half of why I distrust the number. The other half is that the count is anonymous by design. Forty-seven votes tells you forty-seven people clicked. It does not tell you whether they are forty-seven trials that expired last month or forty-seven accounts paying you every month. Those are completely different facts wearing the same number.
So the board does one more thing to argue with the count: it weights each vote by what the voter actually pays, pulled from Stripe. A feature with a handful of votes from your largest accounts can outrank one with a crowd of votes from people who will never pay you a cent. The raw count says popular. The weighted score says valuable. They are not the same word, and the gap between them is where most of the wrong features get built.
Stack that on top of the merge and the board gets honest in a hurry. The merge strips the duplicate votes, the weighting reranks what survives by revenue, and the feature that looked like a mandate, a big anonymous pile at the top, often turns out to be a modest request from people who were never going to pay. Three separate corrections aimed at one number, which should tell you how little I trust it standing alone.
This is also the part that separates the tool from most boards. Canny, Productboard, and the rest were built as engagement tools, so they count every vote the same. If you want the full reasoning, it is in revenue weighted voting. The short version: a number you cannot trace to revenue is a number you cannot trust to prioritize.
The honest use of the number
I still show it. The count is genuinely useful as a question: why is this near the top, who put it there, and what were they actually trying to do. It's dangerous only as an answer, when the biggest number ends the conversation instead of starting one.
So I kept the count and built the rest of the tool to argue with it. That is probably the strangest position to be in: knowing in exact detail how a number is produced, which is precisely why you won't let it have the last word.
