I just launched GuessCrowd, a small daily survey game: five everyday prompts ("Name a pizza topping", "Name something people lose all the time"), 20 seconds each, and you score by naming the answer most people would give.
The obvious problem with a game like this is the cold start. A survey game is only as good as its survey, and on launch day nobody has answered anything yet. Here is how I handled it.
Two sources of truth
Every board has two inputs:
- An editorial prior. For each prompt there is a starting board: a best estimate of how 100 people would answer, as weights out of 100. "Name a fruit that's yellow" starts as Banana 50, Lemon 24, Pineapple 8, Mango 5, Grapefruit 3.
- The live tally. Every answer a player locks in is counted under a normalized key, so "bananas" and "bannana" land on Banana.
The board a player sees is a blend of the two.
The blend
Let n be the number of live answers counted for a prompt. The weight given to the live share is
lambda = n / (n + 25)
share(answer) = (1 - lambda) * prior(answer) + lambda * live(answer)
With no players, lambda is 0 and the board is pure estimate. At 25 answers it is half and half. After a few hundred answers the prior barely matters, and the board belongs to the players. The constant 25 is a judgment call: small enough that a real crowd takes over within the first day, large enough that the first two or three players can't flip the board.
Scoring off the blended board
Points are relative to the top answer: the #1 answer is worth 100, and anything else scores in proportion to its share compared with #1. If half as many people say your answer, you get 50. Any locked-in answer gets at least 5 points, and letting the timer run out scores zero.
That proportional rule matters for fairness across the day. Early players are scored mostly against the estimate and later players mostly against the crowd, but in both cases the score measures the same thing: how close you were to the most common answer at the moment you played.
Fuzzy matching without surprises
People type fast under a 20-second timer, so close spellings and plurals snap to the nearest answer already on the board. Anything genuinely different stays as its own answer and starts its own count, which is how unexpected answers can climb onto the board over the day.
What I'd watch next
- Whether 25 is the right constant once real traffic arrives.
- Prompts where the prior was clearly wrong, which is the most useful feedback for writing better boards.
If you like this kind of game, today's board is at guesscrowd.com. It's free, runs in the browser and needs no account.
Originally published on Hashnode.
Top comments (0)