I have a guilty pleasure of watching Family Feud sometimes. It's fun to play along with at home, but it's not really challenging, by design. The highest-scoring answers are the most common ones. Name something you'd find in a kitchen. Refrigerator. Ten points. Next.
I figured it'd be more fun to do it the other way around: try to guess the least common answers in the 100-person survey. Those answers also tend to be a little unhinged, which is half the appeal. Someone out there said "bathtub" to that kitchen question, and I want to meet that person.
The judging problem
The hard part wasn't the game logic. It was the judge. If the board says "car" and someone types "jeep," how do you give them credit without doing exact string matching? You can't just check guess === answer. Real people type weird stuff.
So I built a tiny AI Steve Harvey. It gets the full context of the question and the answer list, and tries to figure out what the player actually meant. It's not perfect, but it handles the jeep-to-car leap well enough that it feels fair.
Vibe classification
I also wanted the host to react. If you type something really unhinged or shocking, Steve should make a face about it. That's half the show. So the judge also classifies the vibe of each guess and the host responds accordingly.
Play it
Live at familyfew-d.com. Three guesses per round, and the obvious ones don't count for much. Go find the weird answers.